Object state changes in video reveal critical cues about human and agent activity. However, existing methods are limited to temporal localization of when the object is in its initial state (e.g., cheese block) versus when it has completed a state change (e.g., grated cheese), offering no insight into where the change is unfolding. We propose to deepen the problem by introducing the spatially-progressing object state change segmentation task. The goal is to segment at the pixel-level those regions of an object that are actionable and those that are transformed. We show that state-of-the-art VLMs and video segmentation methods struggle at this task, underscoring its difficulty and novelty. As an initial baseline, we design a VLM-based pseudo-labeling approach, state-change dynamics constraints, and a novel WhereToChange benchmark built on in-the-wild Internet videos. Experiments on two datasets validate both the challenge of the new task as well as the promise of our model for localizing exactly where and how fast objects are changing in video. We further demonstrate useful implications for tracking activity progress to benefit robotic agents. Overall, our work positions spatial OSC segmentation as a new frontier task for video understanding: one that challenges current SOTA methods and invites the community to build more robust, state-change-sensitive representations.

Abstract An environment acoustic model represents how sound is transformed by the physical characteristics of an indoor environment, for any given source/receiver location. Whereas traditional methods for constructing such models assume dense geometry and/or sound measurements throughout the environment, we explore how to infer room impulse responses (RIRs) based on a sparse set of images and echoes observed in the space, as well as how to choose where to collect these audio-visual observations. Towards that goal, we first introduce a transformer-based method that uses self-attention to build a rich acoustic context, then infers the RIRs of arbitrary query source-receiver locations through cross-attention. Then, motivated by real-world physical constraints in collecting these observations, we further introduce active acoustic sampling , a new task in which a mobile agent jointly constructs the environment acoustic model and spatial occupancy map on-the-fly from sparse audio-visual observations. We train a reinforcement learning (RL) policy that guides agent navigation toward optimal acoustic data sampling positions, rewarding information gain for the full environment model. Evaluating on diverse unseen 3D indoor environments, our method outperforms the state-of-the-art and—in a major departure from traditional methods—generalizes to novel environments in a few-shot manner. Furthermore, when augmented with our active sampling policy, it successfully guides an embodied agent to acoustically informative positions given real-world exploration constraints, outperforming both traditional navigation agents and prior acoustic rendering methods. Project: http://vision.cs.utexas.edu/projects/fewShot-RIR .

Feedback is essential for learning a new skill or improving one's current skill-level. However, current methods for skill-assessment from video only provide scores or compare demonstrations, leaving the burden of knowing what to do differently on the user. We introduce a novel method to generate actionable feedback (AF) from video of a person doing a physical activity, such as basketball or soccer. Our method takes a video demonstration and its accompanying 3D body pose and generates (1) free-form expert commentary describing what the person is doing well and what they could improve, and (2) a visual expert demonstration that incorporates the required corrections. We show how to leverage Ego-Exo4D's [29] videos of skilled activity and expert commentary together with a strong language model to create a weakly-supervised training dataset for this task, and we devise a multimodal video-language model to infer coaching feedback. Our method is able to reason across multi-modal input combinations to output full-spectrum, actionable coaching—expert commentary, expert video retrieval, and expert pose generation—outperforming strong vision-language models on both established metrics and human preference studies.

Anticipating how a person will interact with objects in an environment is essential for activity understanding, but existing methods are limited to the 2D space of video frames—capturing physically ungrounded predictions of "what" and ignoring the "where" and "how". We introduce FICTION for 4D future interaction prediction from videos. Given an input video of a human activity, the goal is to predict which objects at what 3D locations the person will interact with in the next time period (e.g., cabinet, fridge), and how they will execute that interaction (e.g., poses for bending, reaching, pulling). Our novel model FIction fuses the past video observation of the person’s actions and their environment to predict both the "where" and "how" of future interactions. Through comprehensive experiments on a variety of activities and real-world environments in Ego-Exo4D, we show that our proposed approach outperforms prior autoregressive and (lifted) 2D video models substantially, with more than 30% relative gains.

Egocentric and exocentric perspectives of human action differ significantly, yet overcoming this extreme viewpoint gap is critical in augmented reality and robotics. We propose VIEWPOINTROSETTA, an approach that unlocks large-scale unpaired ego and exo video data to learn clip-level viewpoint-invariant video representations. Our framework introduces (1) a diffusion-based Rosetta Stone Translator (RST), which, leveraging a moderate amount of synchronized multi-view videos, serves as a translator in feature space to decipher the alignment between unpaired ego and exo data, and (2) a dual encoder that aligns unpaired data representations through contrastive learning with RST-based synthetic feature augmentation and soft alignment. To evaluate the learned features in a standardized setting, we construct a new cross-view benchmark using Ego-Exo4D, covering cross-view retrieval, action recognition, and skill assessment tasks. Our framework demonstrates superior cross-view understanding compared to previous view-invariant learning and ego video representation learning approaches, and opens the door to bringing vast amounts of traditional third-person video to bear on the more nascent first-person setting.

Given a multi-view video, which viewpoint is most informative for a human observer? Existing methods rely on heuristics or expensive "best-view" supervision to answer this question, limiting their applicability. We propose a weakly supervised approach that leverages language accompanying an instructional multi-view video as a means to recover its most informative viewpoint(s). Our key hypothesis is that the more accurately an individual view can predict a view-agnostic text summary, the more informative it is. To put this into action, we propose LANGVIEW, a framework that uses the relative accuracy of view-dependent caption predictions as a proxy for best view pseudo-labels. Then, those pseudo-labels are used to train a view selector, together with an auxiliary camera pose predictor that enhances view-sensitivity. During inference, our model takes as input only a multi-view video—no language or camera poses—and returns the best viewpoint to watch at each timestep. On two challenging datasets comprised of diverse multi-camera setups and how-to activities, our model consistently outperforms state-of-the-art baselines, both with quantitative metrics and human evaluation. Project: https://vision.cs.utexas.edu/projects/which-view-shows-it-best.

While image captioning provides isolated descriptions for individual images, and video captioning offers one single narrative for an entire video clip, our work explores an important middle ground: progress-aware video captioning at the frame level. This novel task aims to generate temporally fine-grained captions that not only accurately describe each frame but also capture the subtle progression of actions throughout a video sequence. Despite the strong capabilities of existing leading vision language models, they often struggle to discern the nuances of frame-wise differences. To address this, we propose ProgressCaptioner, a captioning model designed to capture the fine-grained temporal dynamics within an action sequence. Alongside, we develop the FrameCap dataset to support training and the Frame-CapEval benchmark to assess caption quality. The results demonstrate that ProgressCaptioner significantly surpasses leading captioning models, producing precise captions that accurately capture action progression and set a new standard for temporal precision in video captioning. Finally, we showcase practical applications of our approach, specifically in aiding keyframe selection and advancing video understanding, highlighting its broad utility.1

Skill assessment from video entails rating the quality of a person’s physical performance and explaining what could be done better. Today’s models specialize for an individual sport, and suffer from the high cost and scarcity of expert-level supervision across the long tail of sports. Towards closing that gap, we explore transferable video representations for skill assessment. Our CrossTrainer approach discovers skill-attributes—such as balance, control, and hand positioning—whose meaning transcends the boundaries of any given sport, then trains a multimodal language model to generate actionable feedback for a novel video, e.g., “lift hands more to generate more power” as well as its proficiency level, e.g., early expert. We validate the new model on multiple datasets for both cross-sport (transfer) and intra-sport (in-domain) settings, where it achieves gains up to 60% relative to the state of the art. By abstracting out the shared behaviors indicative of human skill, the proposed video representation generalizes substantially better than an array of existing techniques, enriching today’s multimodal large language models.

Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM–VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about ''what'', ''where'', ''when'', and ''how'' of a video. We make our work fully reproducible by providing data, training recipes, code & models.

Video reasoning, the task of enabling machines to infer from dynamic visual content through multi-step logic, is crucial for advanced AI. While the Chain-of-Thought (CoT) mechanism has enhanced reasoning in text-based tasks, its application to video understanding remains underexplored. This paper presents a systematic analysis revealing that CoT often degrades performance in video reasoning, generating verbose but misleading internal monologues, and leading to hallucinated visual details and overridden correct intuitions—a phenomenon we term "visual thinking drift." We explain this drift through a Bayesian lens, positing that CoT traces often diverge from actual visual evidence, instead amplifying internal biases or language priors, causing models to storytell rather than engage in grounded reasoning. To counteract this, we introduce Visual Evidence Reward (VER), a novel reinforcement learning framework that explicitly rewards the generation of reasoning traces that are verifiably grounded in visual evidence. Comprehensive evaluation across 10 diverse video understanding benchmarks demonstrates that our Video-VER model consistently achieves top performance. Our work sheds light on the distinct challenges of video-centric reasoning and encourages the development of AI that robustly grounds its inferences in visual evidence---for large multimodal models that not only "think before answering", but also "see while thinking".

The Arrow of Time (AoT)—time's irreversible flow shaping physical events—is fundamental to video comprehension, yet remains a significant challenge for modern large multimodal models (LMMs). Current LMMs struggle to perceive and utilize temporal directionality in video when responding to language queries, obstructing deeper temporal understanding. We tackle this deficiency by first providing a critical analysis of existing benchmarks and models. We then introduce ArrowRL, a reinforcement learning (RL)-based training strategy with an innovative reverse reward that instills AoT awareness by encouraging divergent video interpretations between forward and reversed visual frames. For rigorous evaluation, we additionally develop AoTBench, a new multi-faceted benchmark probing temporally challenging questions. Experiments show ArrowRL greatly advances temporal perception: it not only achieves substantial improvements on our challenging AoTBench but also demonstrably boosts performance on standard video question answering (VQA) benchmarks (with peak accuracy gains reaching over 20% and 10% respectively). This validates ArrowRL's effectiveness and highlights the critical need for dedicated AoT understanding in LMMs.

We present Ego-Exo4D, a diverse, large-scale multimodal multiview video dataset and benchmark challenge. Ego-Exo4D centers around simultaneously-captured egocentric and exocentric video of skilled human activities (e.g., sports, music, dance, bike repair). 740 participants from 13 cities worldwide performed these activities in 123 different natural scene contexts, yielding long-form captures from 1 to 42 minutes each and 1,286 hours of video combined. The multimodal nature of the dataset is unprecedented: the video is accompanied by multichannel audio, eye gaze, 3D point clouds, camera poses, IMU, and multiple paired language descriptions—including a novel “expert commentary” done by coaches and teachers and tailored to the skilled-activity domain. To push the frontier of first-person video understanding of skilled human activity, we also present a suite of benchmark tasks and their annotations, including fine-grained activity understanding, proficiency estimation, cross-view translation, and 3D hand/body pose. All resources are open sourced to fuel new research in the community. https://ego-exo4d-data.org/

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of daily-life activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards, with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception.

While object reconstruction has made great strides in recent years, current methods typically require densely captured images and/or known camera poses, and generalize poorly to novel object categories. To step toward object reconstruction in the wild, this work explores reconstructing general real-world objects from a few images without known camera poses or object categories. The crux of our work is solving two fundamental 3D vision problems — shape reconstruction and pose estimation — in a unified approach. Our approach captures the synergies of these two problems: reliable camera pose estimation gives rise to accurate shape reconstruction, and the accurate reconstruction, in turn, induces robust correlations between different views and facilitates pose estimation. Our method FORGE predicts 3D features from each view and leverages them in conjunction with the input images to establish cross-view correlations for estimating relative camera poses. The 3D features are then transformed by the estimated poses into a shared space and are fused into a neural radiance field. The reconstruction results are rendered by volume rendering techniques, enabling us to train the model without 3D shape groundtruth. Our experiments on both real and synthetic datasets, as well as in the wild images, show that FORGE reliably reconstructs objects from five views. Our pose estimation method outperforms existing ones by a large margin. The reconstruction results under predicted poses are comparable to the ones using ground-truth poses. And the performance on novel testing categories matches the results on categories seen during training.

We introduce the video detours problem for navigating instructional videos. Given a source video and a natural language query asking to alter the how-to video's current path of execution in a certain way, the goal is to find a related “detour video” that satisfies the requested alteration. To address this challenge, we propose VidDetours, a novel video-language approach that learns to retrieve the targeted temporal segments from a large repository of how-to's using video-and-text conditioned queries. Furthermore, we devise a language-based pipeline that exploits how-to video narration text to create weakly supervised training data. We demonstrate our idea applied to the domain of how-to cooking videos, where a user can detour from their current recipe to find steps with alternate ingredients, tools, and techniques. Validating on a ground truth annotated dataset of 16K samples, we show our model's significant improvements over best available methods for video retrieval and question answering, with recall rates exceeding the state of the art by 35%.

We propose a novel self-supervised embedding to learn how actions sound from narrated in-the-wild egocentric videos. Whereas existing methods rely on curated data with known audio-visual correspondence, our multimodal contrastive-consensus coding (M C3) embedding reinforces the associations between audio, language, and vision when all modality pairs agree, while diminishing those associations when anyone pair does not. We show our approach can successfully discover how the long tail of human actions sound from egocentric video, outperforming an array of recent multimodal embedding techniques on two datasets (Eg04D and EPIC-Sounds) and multiple cross-modal tasks.

We present Ego-Exo4D, a diverse, large-scale multi-modal multiview video dataset and benchmark challenge. Ego-Exo4D centers around simultaneously-captured ego-centric and exocentric video of skilled human activities (e.g., sports, music, dance, bike repair). 740 participants from 13 cities worldwide performed these activities in 123 different natural scene contexts, yielding long-form captures from 1 to 42 minutes each and 1,286 hours of video combined. The multimodal nature of the dataset is un-precedented: the video is accompanied by multichannel audio, eye gaze, 3D point clouds, camera poses, IMU, and multiple paired language descriptions-including a novel “expert commentary” done by coaches and teachers and tailored to the skilled-activity domain. To push the frontier of first-person video understanding of skilled human activity, we also present a suite of benchmark tasks and their annotations, including fine-grained activity understanding, proficiency estimation, cross-view translation, and 3D hand/body pose. All resources are open sourced to fuel new research in the community.

We propose a self-supervised method for learning repre- sentations based on spatial audio-visual correspondences in egocentric videos. Our method uses a masked auto- encoding framework to synthesize masked binaural audio through the synergy of audio and vision, thereby learning useful spatial relationships between the two modalities We use our pretrained features to tackle two downstream video tasks requiring spatial understanding in social scenar- ios: active speaker detection and spatial audio denoising. Through extensive experiments, we show that our features are generic enough to improve over multiple state-of-the- art baselines on both tasks on two challenging egocentric video datasets that offer binaural audio, EgoCom and Easy-Com. Project: http://vision.cs.utexas.edu/projects/ego_av_corr.

Object State Changes (OSCs) are pivotal for video understanding. While humans can effortlessly generalize OSC understanding from familiar to unknown objects, current approaches are confined to a closed vocabulary. Addressing this gap, we introduce a novel open-world formulation for the video OSC problem. The goal is to temporally localize the three stages of an OSC-the object's initial state, its transitioning state, and its end state-whether or not the object has been observed during training. Towards this end, we develop Vidosc, a holistic learning approach that: (1) leverages text and vision-language models for super-visory signals to obviate manually labeling OSC training data, and (2) abstracts fine-grained shared state representations from objects to enhance generalization. Furthermore, we present HowToChange, the first open-world benchmark for video OSC localization, which offers an order of mag-nitude increase in the label space and annotation volume compared to the best existing benchmark. Experimental results demonstrate the efficacy of our approach, in both tra-ditional closed-world and open-world scenarios.11Project webpage: https://vision.cs.utexas.edu/projects/VidOSC/.

Sim2real transfer has received increasing attention lately due to its success in transferring robotic policies learned in simulation to the real world. While significant progress has been made in transferring vision-based navigation policies, the current sim2real strategy for audio-visual navigation remains limited to basic data augmentation. Sound differs from light in that it spans across much wider frequencies and thus requires a different solution for sim2real. To understand how the acoustic sim2real gap varies with frequencies, we first define a novel acoustic field prediction (AFP) task that predicts the local sound pressure field. We then train frequency-specific AFP models in simulation and measure the prediction errors on collected real data. We propose a frequency-adaptive strategy that intelligently selects the best frequency band for prediction based on both the measured prior and the energy distribution of the received audio, which improves the generalization on real data. Coupled with waypoint navigation, we show the navigation policy not only improves navigation performance in simulation but also transfers successfully to real robots. This work demonstrates the potential of building autonomous agents that can see, hear, and act entirely from simulation, and transferring them to the real world.

An environment acoustic model represents how sound is transformed by the physical characteristics of an indoor environment, for any given source/receiver location. Traditional methods for constructing acoustic models involve expensive and time-consuming collection of large quantities of acoustic data at dense spatial locations in the space, or rely on privileged knowledge of scene geometry to intelligently select acoustic data sampling locations. We propose active acoustic sampling, a new task for efficiently building an environment acoustic model of an unmapped environment in which a mobile agent equipped with visual and acoustic sensors jointly constructs the environment acoustic model and the occupancy map on-the-fly. We introduce ActiveRIR, a reinforcement learning (RL) policy that leverages information from audio-visual sensor streams to guide agent navigation and determine optimal acoustic data sampling positions, yielding a high quality acoustic model of the environment from a minimal set of acoustic samples. We train our policy with a novel RL reward based on information gain in the environment acoustic model. Evaluating on diverse unseen indoor environments from a state-of-the-art acoustic simulation platform, ActiveRIR outperforms an array of methods—both traditional navigation agents based on spatial novelty and visual exploration as well as existing state-of-the-art methods.

We study the problem of precisely swapping objects in videos, with a focus on those interacted with by hands, given one user-provided reference object image. Despite the great advancements that diffusion models have made in video editing recently, these models often fall short in handling the intricacies of hand-object interactions (HOI), failing to produce realistic edits---especially when object swapping results in object shape or functionality changes. To bridge this gap, we present HOI-Swap, a novel diffusion-based video editing framework trained in a self-supervised manner. Designed in two stages, the first stage focuses on object swapping in a single frame with HOI awareness; the model learns to adjust the interaction patterns, such as the hand grasp, based on changes in the object's properties. The second stage extends the single-frame edit across the entire sequence; we achieve controllable motion alignment with the original video by: (1) warping a new sequence from the stage-I edited frame based on sampled motion points and (2) conditioning video generation on the warped sequence. Comprehensive qualitative and quantitative evaluations demonstrate that HOI-Swap significantly outperforms existing methods, delivering high-quality video edits with realistic HOIs.

Video-language embeddings are a promising avenue for injecting semantics into visual representations, but existing methods capture only short-term associations between seconds-long video clips and their accompanying text. We propose HierVL, a novel hierarchical video-language em-bedding that simultaneously accounts for both long-term and short-term associations. As training data, we take videos accompanied by timestamped text descriptions of human actions, together with a high-level text summary of the activity throughout the long video (as are available in Ego4D). We introduce a hierarchical contrastive training objective that encourages text-visual alignment at both the clip level and video level. While the clip-level constraints use the step-by-step descriptions to capture what is happening in that instant, the video-level constraints use the summary text to capture why it is happening, i.e., the broader context for the activity and the intent of the actor. Our hierarchical scheme yields a clip representation that outperforms its single-level counterpart as well as a long-term video representation that achieves SotA results on tasks requiring long-term video modeling. HierVL success-fully transfers to multiple challenging downstream tasks (in EPIC-KITCHENS-100, Charades-Ego, HowTo100M) in both zero-shot and fine-tuned settings.

We introduce the novel-view acoustic synthesis (NVAS) task: given the sight and sound observed at a source view-point, can we synthesize the sound of that scene from an unseen target viewpoint? We propose a neural rendering approach: Visually-Guided Acoustic Synthesis (ViGAS) network that learns to synthesize the sound of an arbitrary point in space by analyzing the input audio-visual cues. To benchmark this task, we collect two first-of-their-kind large-scale multi-view audio-visual datasets, one synthetic and one real. We show that our model successfully reasons about the spatial cues and synthesizes faithful audio on both datasets. To our knowledge, this work represents the very first formulation, dataset, and approach to solve the novel-view acoustic synthesis task, which has exciting potential applications ranging from AR/VR to art and design. Unlocked by this work, we believe that the future of novel-view synthesis is in multi-modal learning from videos.

Can conversational videos captured from multiple egocentric viewpoints reveal the map of a scene in a cost-efficient way? We seek to answer this question by proposing a new problem: efficiently building the map of a previously unseen 3D environment by exploiting shared information in the egocentric audio-visual observations of participants in a natural conversation. Our hypothesis is that as multiple people (“egos”) move in a scene and talk among themselves, they receive rich audio-visual cues that can help uncover the unseen areas of the scene. Given the high cost of continuously processing egocentric visual streams, we further explore how to actively coordinate the sampling of visual information, so as to minimize redundancy and reduce power use. To that end, we present an audio-visual deep reinforcement learning approach that works with our shared scene mapper to selectively turn on the camera to efficiently chart out the space. We evaluate the approach using a state-of-the-art audio-visual simulator for 3D scenes as well as real-world video. Our model outperforms previous state-of-the-art mapping methods, and achieves an excellent cost-accuracy tradeoff. Project: http://vision.cs.utexas.edu/projects/chat2map.

Searching long egocentric videos with natural language queries (NLQ) has compelling applications in augmented reality and robotics, where a fluid index into everything that a person (agent) has seen before could augment human memory and surface relevant information on demand. However, the structured nature of the learning problem (free-form text query inputs, localized video temporal window outputs) and its needle-in-a-haystack nature makes it both technically challenging and expensive to supervise. We introduce Narrations-as-Queries (NaQ), a data augmentation strategy that transforms standard video-text narrations into training data for a video query localization model. Validating our idea on the Ego4D benchmark, we find it has tremendous impact in practice. NaQ improves multiple top models by substantial margins (even doubling their accuracy), and yields the very best results to date on the Ego4D NLQ challenge, soundly outperforming all challenge winners in the CVPR and ECCV 2022 competitions and topping the current public leaderboard. Beyond achieving the state-of-the-art for NLQ, we also demonstrate unique properties of our approach such as the ability to perform zero-shot and fewshot NLQ, and improved performance on queries about long-tail object categories. Code and models: http://vision.cs.utexas.edu/projects/naq.

Different video understanding tasks are typically treated in isolation, and even with distinct types of curated data (e.g., classifying sports in one dataset, tracking animals in another). However, in wearable cameras, the immersive egocentric perspective of a person engaging with the world around them presents an interconnected web of video understanding tasks—hand-object manipulations, navigation in the space, or human-human interactions—that unfold continuously, driven by the person's goals. We argue that this calls for a much more unified approach. We propose EgoTask Translation (EgoT2), which takes a collection of models optimized on separate tasks and learns to translate their outputs for improved performance on any or all of them at once. Unlike traditional transfer or multi-task learning, EgoT2's “flipped design” entails separate task-specific backbones and a task translator shared across all tasks, which captures synergies between even heterogeneous tasks and mitigates task competition. Demonstrating our model on a wide array of video tasks from Ego4D, we show its advantages over existing transfer paradigms and achieve top-ranked results on four of the Ego4D 2022 benchmark challenges.11Project webpage: https://vision.cs.utexas.edu/projects/egot2/.

Reverberation not only degrades the quality of speech for human perception, but also severely impacts the accuracy of automatic speech recognition. Prior work attempts to remove reverberation based on the audio modality only. Our idea is to learn to dereverberate speech from audio-visual observations. The visual environment surrounding a human speaker reveals important cues about the room geometry, materials, and speaker location, all of which influence the precise reverberation effects. We introduce Visually-Informed Dereverberation of Audio (VIDA), an end-to-end approach that learns to remove reverberation based on both the observed monaural sound and visual scene. In support of this new task, we develop a large-scale dataset SoundSpaces-Speech that uses realistic acoustic renderings of speech in real-world 3D scans of homes offering a variety of room acoustics. Demonstrating our approach on both simulated and real imagery for speech enhancement, speech recognition, and speaker identification, we show it achieves state-of-the-art performance and substantially improves over audio-only methods.

The goal in episodic memory (EM) is to search a long egocentric video to answer a natural language query (e.g., “where did I leave my purse?”). Existing EM methods exhaustively extract expensive fixed-length clip features to look everywhere in the video for the answer, which is infeasible for long wearable-camera videos that span hours or even days. We propose SpotEM, an approach to achieve efficiency for a given EM method while maintaining good accuracy. SpotEM consists of three key ideas: 1) a novel clip selector that learns to identify promising video regions to search conditioned on the language query; 2) a set of low-cost semantic indexing features that capture the context of rooms, objects, and interactions that suggest where to look; and 3) distillation losses that address the optimization issues arising from end-to-end joint training of the clip selector and EM model. Our experiments on 200+ hours of video from the Ego4D EM Natural Language Queries benchmark and three different EM models demonstrate the effectiveness of our approach: computing only 10% – 25% of the clip features, we preserve 84% – 97% of the original EM model’s accuracy. Project page: https://vision.cs.utexas.edu/projects/spotem

Using visual SLAM to map new environments requires time-consuming visits to all regions for data collection. We propose an approach to estimate maps of areas beyond the visible regions using a cheap and readily available modality of data-sound. We introduce the idea of an active audio-visual mapping agent. Besides collecting visual data, the proposed agent emits sounds during navigation, captures the echoes, and uses them to accurately map unknown areas. We propose a reinforcement learning based method that simultaneously trains models to 1) estimate a map from the visual data, 2) output navigation actions, 3) output the decision to emit a sound and 4) refine estimated mans using the cantured audio. Our agent is trained and tested on 85 real-world homes from the Matterport3D dataset using the Habitat and SoundSpaces simulators for visual and audio data. Our method, unlike visual-data reliant approaches, yields more accurate maps with broader environmental coverage. In addition, compared to an agent that continually emits sounds, we observe that intelligently choosing when to emit sounds leads to accurate maps obatined with greater efficiency.

Procedural activity understanding requires perceiving human actions in terms of a broader task, where multiple keysteps are performed in sequence across a long video to reach a final goal state---such as the steps of a recipe or the steps of a DIY fix-it task. Prior work largely treats keystep recognition in isolation of this broader structure, or else rigidly confines keysteps to align with a particular sequential script. We propose discovering a task graph automatically from how-to videos to represent probabilistically how people tend to execute keysteps, then leverage this graph to regularize keystep recognition in novel videos. On multiple datasets of real-world instructional video, we show the impact: more reliable zero-shot keystep localization and improved video representation learning, exceeding the state of the art.

Visual Query Localization on long-form egocentric videos requires spatio-temporal search and localization of visually specified objects and is vital to build episodic memory systems. Prior work develops complex multi-stage pipelines that leverage well-established object detection and tracking methods to perform VQL. However, each stage is independently trained and the complexity of the pipeline results in slow inference speeds. We propose VQLoC, a novel single-stage VQL framework that is end-to-end trainable. Our key idea is to first build a holistic understanding of the query-video relationship and then perform spatio-temporal localization in a single shot manner. Specifically, we establish the query-video relationship by jointly considering query-to-frame correspondences between the query and each video frame and frame-to-frame correspondences between nearby video frames. Our experiments demonstrate that our approach outperforms prior VQL methods by $20$% accuracy while obtaining a $10\times$ improvement in inference speed. VQLoC is also the top entry on the Ego4D VQ2D challenge leaderboard.

First-person video highlights a camera-wearer's activities in the context of their persistent environment. However, current video understanding approaches reason over visual features from short video clips that are detached from the underlying physical space and capture only what is immediately visible. To facilitate human-centric environment understanding, we present an approach that links egocentric video and the environment by learning representations that are predictive of the camera-wearer's (potentially unseen) local surroundings. We train such models using videos from agents in simulated 3D environments where the environment is fully observable, and test them on human-captured real-world videos from unseen environments. On two human-centric video tasks, we show that models equipped with our environment-aware features consistently outperform their counterparts with traditional clip features. Moreover, despite being trained exclusively on simulated videos, our approach successfully handles real-world videos from HouseTours and Ego4D, and achieves state-of-the-art results on the Ego4D NLQ challenge.

Acoustic matching aims to re-synthesize an audio clip to sound as if it were recorded in a target acoustic environment. Existing methods assume access to paired training data, where the audio is observed in both source and target environments, but this limits the diversity of training data or requires the use of simulated data or heuristics to create paired samples. We propose a self-supervised approach to visual acoustic matching where training samples include only the target scene image and audio---without acoustically mismatched source audio for reference. Our approach jointly learns to disentangle room acoustics and re-synthesize audio into the target environment, via a conditional GAN framework and a novel metric that quantifies the level of residual acoustic information in the de-biased audio. Training with either in-the-wild web data or simulated data, we demonstrate it outperforms the state-of-the-art on multiple challenging datasets and a wide variety of real-world audio and environments.

Recent advances in egocentric video understanding models are promising, but their heavy computational expense is a barrier for many real-world applications. To address this challenge, we propose EgoDistill, a distillation-based approach that learns to reconstruct heavy ego-centric video clip features by combining the semantics from a sparse set of video frames with head motion from lightweight IMU readings. We further devise a novel IMU-based self-supervised pretraining strategy. Our method leads to significant improvements in efficiency, requiring 200× fewer GFLOPs than equivalent video models. We demonstrate its effectiveness on the Ego4D and EPIC- Kitchens datasets, where our method outperforms state-of-the-art efficient video understanding methods.

Visual object tracking is a key component to many egocentric vision problems. However, the full spectrum of challenges of egocentric tracking faced by an embodied AI is underrepresented in many existing datasets; these tend to focus on relatively short, third-person videos. Egocentric video has several distinguishing characteristics from those commonly found in past datasets: frequent large camera motions and hand interactions with objects commonly lead to occlusions or objects exiting the frame, and object appearance can change rapidly due to widely different points of view, scale, or object states. Embodied tracking is also naturally long-term, and being able to consistently (re-)associate objects to their appearances and disappearances over as long as a lifetime is critical. Previous datasets under-emphasize this re-detection problem, and their "framed" nature has led to adoption of various spatiotemporal priors that we find do not necessarily generalize to egocentric video. We thus introduce EgoTracks, a new dataset for long-term egocentric visual object tracking. Sourced from the Ego4D dataset, this new dataset presents a significant challenge to recent state-of-the-art single-object tracking models, which we find score poorly on traditional tracking metrics for our new dataset, compared to popular benchmarks. We further show improvements that can be made to a STARK tracker to significantly increase its performance on egocentric data, resulting in a baseline model we call EgoSTARK. We publicly release our annotations and benchmark, hoping our dataset leads to further advancements in tracking.

The egocentric and exocentric viewpoints of a human activity look dramatically different, yet invariant representations to link them are essential for many potential applications in robotics and augmented reality. Prior work is limited to learning view-invariant features from paired synchronized viewpoints. We relax that strong data assumption and propose to learn fine-grained action features that are invariant to the viewpoints by aligning egocentric and exocentric videos in time, even when not captured simultaneously or in the same environment. To this end, we propose AE2, a self-supervised embedding approach with two key designs: (1) an object-centric encoder that explicitly focuses on regions corresponding to hands and active objects; (2) a contrastive-based alignment objective that leverages temporally reversed frames as negative samples. For evaluation, we establish a benchmark for fine-grained video understanding in the ego-exo context, comprising four datasets---including an ego tennis forehand dataset we collected, along with dense per-frame labels we annotated for each dataset. On the four datasets, our AE2 method strongly outperforms prior work in a variety of fine-grained downstream tasks, both in regular and cross-view settings.

The papers in this special issue focus on egocentric perception. It gathers recent advances in this field that brings together multiple communities including computer vision, machine learning, and multimedia.

In reinforcement learning for visual navigation, it is common to develop a model for each new task, and train that model from scratch with task-specific interactions in 3D environments. However, this process is expensive; mas-sive amounts of interactions are needed for the model to generalize well. Moreover, this process is repeated when-ever there is a change in the task type or the goal modality. We present a unified approach to visual navigation using a novel modular transfer learning model. Our model can ef-fectively leverage its experience from one source task and apply it to multiple target tasks (e.g., ObjectNav, Room-Nav, Vi ewNav) with various goal modalities (e.g., image, sketch, audio, label). Furthermore, our model enables zero-shot experience learning, whereby it can solve the target tasks without receiving any task-specific interactive training. Our experiments on multiple photorealistic datasets and challenging tasks show that our approach learns faster, generalizes better, and outperforms SoTA models by a sig-nificant margin. Project page: https://vision.cs.utexas.edu/projects/zsel/

We introduce the visual acoustic matching task, in which an audio clip is transformed to sound like it was recorded in a target environment. Given an image of the target environment and a waveform for the source audio, the goal is to re-synthesize the audio to match the target room acoustics as suggested by its visible geometry and materials. To address this novel task, we propose a cross-modal transformer model that uses audio-visual attention to inject visual properties into the audio and generate realistic audio output. In addition, we devise a self-supervised training objective that can learn acoustic matching from in-the-wild Web videos, despite their lack of acoustically mismatched audio. We demonstrate that our approach successfully translates human speech to a variety of real-world environments depicted in images, outperforming both traditional acoustic matching and more heavily supervised baselines.

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of dailylife activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards, with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception. Project page: https://ego4d-data.org/

State-of-the-art approaches to ObjectGoal navigation (ObjectNav) rely on reinforcement learning and typically require significant computational resources and time for learning. We propose Potential functions for ObjectGoal Navigation with Interaction-free learning (PONI), a modular approach that disentangles the skills of 'where to look?’ for an object and 'how to navigate to$(x,\ y)$?’. Our key insight is that 'where to look?’ can be treated purely as a perception problem, and learned without environment interactions. To address this, we propose a network that predicts two complementary potential functions conditioned on a semantic map and uses them to decide where to look for an unseen object. We train the potential function network using supervised learning on a passive dataset of top-down semantic maps, and integrate it into a modular framework to perform ObjectNav. Experiments on Gibson and Matterport3D demonstrate that our method achieves the stateof-the-art for ObjectNav while incurring up to$1,600\times less$computational cost for training. Code and pre-trained models are available.11Website: https://vision.cs.utexas.edu/projects/poni/

We introduce SoundSpaces 2.0, a platform for on-the-fly geometry-based audio rendering for 3D environments. Given a 3D mesh of a real-world environment, SoundSpaces can generate highly realistic acoustics for arbitrary sounds captured from arbitrary microphone locations. Together with existing 3D visual assets, it supports an array of audio-visual research tasks, such as audio-visual navigation, mapping, source localization and separation, and acoustic matching. Compared to existing resources, SoundSpaces 2.0 has the advantages of allowing continuous spatial sampling, generalization to novel environments, and configurable microphone and material properties. To our knowledge, this is the first geometry-based acoustic simulation that offers high fidelity and realism while also being fast enough to use for embodied learning. We showcase the simulator's properties and benchmark its performance against real-world audio measurements. In addition, we demonstrate two downstream tasks---embodied navigation and far-field automatic speech recognition---and highlight sim2real performance for the latter. SoundSpaces 2.0 is publicly available to facilitate wider research for perceptual systems that can both see and hear.

Room impulse response (RIR) functions capture how the surrounding physical environment transforms the sounds heard by a listener, with implications for various applications in AR, VR, and robotics. Whereas traditional methods to estimate RIRs assume dense geometry and/or sound measurements throughout the environment, we explore how to infer RIRs based on a sparse set of images and echoes observed in the space. Towards that goal, we introduce a transformer-based method that uses self-attention to build a rich acoustic context, then predicts RIRs of arbitrary query source-receiver locations through cross-attention. Additionally, we design a novel training objective that improves the match in the acoustic signature between the RIR predictions and the targets. In experiments using a state-of-the-art audio-visual simulator for 3D environments, we demonstrate that our method successfully generates arbitrary RIRs, outperforming state-of-the-art methods and---in a major departure from traditional methods---generalizing to novel environments in a few-shot manner. Project: http://vision.cs.utexas.edu/projects/fs_rir

The fashion sense—meaning the clothing styles people wear—in a geographical region can reveal information about that region. For example, it can reflect the kind of activities people do there, or the type of crowds that frequently visit the region (e.g., tourist hot spot, student neighborhood, business center). We propose a method to create underground neighborhood maps of cities by analyzing how people dress. Using publicly available images from across a city, our method automatically segments the map into neighborhoods with a similar fashion sense. Our approach further allows discovering insights about a city, such as detecting distinct neighborhoods (what is the most unique region of NYC?) and answering analogy questions between cities (what is the "Downtown LA" of Bogota?). We also present two new underground map benchmarks derived from non-image data for 37 cities worldwide. Our method shows promising results on both these benchmarks as well as experiments with human judges."The map is not the thing mapped."—Eric Temple Bell

While 360 cameras offer tremendous new possibilities in vision, graphics, and augmented reality, the spherical images they produce make visual recognition non-trivial. Ideally, 360 imagery could inherit the convolutional neural networks (CNNs) trained with great success on perspective projection images. However, existing methods to transfer CNNs from perspective to spherical images introduce significant computational costs and/or degradations in accuracy. We propose to learn a Spherical Convolution Network (SphConv) that translates a planar CNN to the equirectangular projection of 360 images. Given a source CNN for perspective images, SphConv learns to reproduce the flat filter outputs on 360 data. The key benefits are 1) efficient and accurate recognition for 360 images, and 2) the ability to leverage pre-trained networks for perspective images. We propose two instantiations of SphConv---Spherical Kernel, which learns location dependent kernels on the sphere, and Kernel Transformer Network, which learns a functional transformation that generates SphConv from the source CNN. Validating our approach with multiple source CNNs and datasets, we show that it successfully preserves the source CNN's accuracy, while offering efficiency, transferability, and scalability to typical image resolutions. We further introduce a spherical Faster R-CNN based on SphConv and show that we can learn a spherical object detector without any object annotations in 360 images.

Recent work on audio-visual navigation assumes a constantly-sounding target and restricts the role of audio to signaling the target’s position. We introduce semantic audio-visual navigation, where objects in the environment make sounds consistent with their semantic meaning (e.g., toilet flushing, door creaking) and acoustic events are sporadic or short in duration. We propose a transformer-based model to tackle this new semantic AudioGoal task, incorporating an inferred goal descriptor that captures both spatial and semantic properties of the target. Our model’s persistent multimodal memory enables it to reach the goal even long after the acoustic event stops. In support of the new task, we also expand the SoundSpaces audio simulations to provide semantically grounded sounds for an array of objects in Matterport3D. Our method strongly outperforms existing audio-visual navigation methods by learning to associate semantic, acoustic, and visual cues.1

We introduce a new approach for audio-visual speech separation. Given a video, the goal is to extract the speech associated with a face in spite of simultaneous back-ground sounds and/or other human speakers. Whereas existing methods focus on learning the alignment between the speaker’s lip movements and the sounds they generate, we propose to leverage the speaker’s face appearance as an additional prior to isolate the corresponding vocal qualities they are likely to produce. Our approach jointly learns audio-visual speech separation and cross-modal speaker embeddings from unlabeled video. It yields state-of-the-art results on five benchmark datasets for audio-visual speech separation and enhancement, and generalizes well to challenging real-world videos of diverse scenarios. Our video results and code: http://vision.cs.utexas.edu/projects/VisualVoice/.

We introduce an approach for pre-training egocentric video models using large-scale third-person video datasets. Learning from purely egocentric data is limited by low dataset scale and diversity, while using purely exocentric (third-person) data introduces a large domain mismatch. Our idea is to discover latent signals in third-person video that are predictive of key egocentric-specific properties. Incorporating these signals as knowledge distillation losses during pre-training results in models that benefit from both the scale and diversity of third-person video data, as well as representations that capture salient egocentric properties. Our experiments show that our "Ego-Exo" framework can be seamlessly integrated into standard video models; it outperforms all baselines when fine-tuned for egocentric activity recognition, achieving state-of-the-art results on Charades-Ego and EPIC-Kitchens-100.

Conversational interfaces for the detail-oriented retail fashion domain are more natural, expressive, and user friendly than classical keyword-based search interfaces. In this paper, we introduce the Fashion IQ dataset to support and advance research on interactive fashion image retrieval. Fashion IQ is the first fashion dataset to provide human-generated captions that distinguish similar pairs of garment images together with side-information consisting of real-world product descriptions and derived visual attribute labels for these images. We provide a detailed analysis of the characteristics of the Fashion IQ data, and present a transformer-based user simulator and interactive image retriever that can seamlessly integrate visual attributes with image features, user feedback, and dialog history, leading to improved performance over the state of the art in dialog-based image retrieval. We believe that our dataset will encourage further work on developing more natural and real-world applicable conversational shopping assistants.1

We propose Anticipative Video Transformer (AVT), an end-to-end attention-based video modeling architecture that attends to the previously observed video in order to anticipate future actions. We train the model jointly to predict the next action in a video sequence, while also learning frame feature encoders that are predictive of successive future frames’ features. Compared to existing temporal aggregation strategies, AVT has the advantage of both maintaining the sequential progression of observed actions while still capturing long-range dependencies—both critical for the anticipation task. Through extensive experiments, we show that AVT obtains the best reported performance on four popular action anticipation benchmarks: EpicKitchens-55, EpicKitchens-100, EGTEA Gaze+, and 50-Salads; and it wins first place in the EpicKitchens-100 CVPR’21 challenge.

Fashion is intertwined with external cultural factors, but identifying these links remains a manual process limited to only the most salient phenomena. We propose a data-driven approach to identify specific cultural factors affecting the clothes people wear. Using large-scale datasets of news articles and vintage photos spanning a century, we present a multi-modal statistical model to detect influence relationships between happenings in the world and people’s choice of clothing. Furthermore, on two image datasets we apply our model to improve the concrete vision tasks of visual style forecasting and photo timestamping. Our work is a first step towards a computational, scalable, and easily refreshable approach to link culture to clothing.

We introduce the active audio-visual source separation problem, where an agent must move intelligently in order to better isolate the sounds coming from an object of interest in its environment. The agent hears multiple audio sources simultaneously (e.g., a person speaking down the hall in a noisy household) and it must use its eyes and ears to automatically separate out the sounds originating from a target object within a limited time budget. Towards this goal, we introduce a reinforcement learning approach that trains movement policies controlling the agent’s camera and microphone placement over time, guided by the improvement in predicted audio separation quality. We demonstrate our approach in scenarios motivated by both augmented reality (system is already co-located with the target object) and mobile robotics (agent begins arbitrarily far from the target object). Using state-of-the-art realistic audio-visual simulations in 3D environments, we demonstrate our model’s ability to find minimal movement sequences with maximal payoff for audio source separation. Project: http://vision.cs.utexas.edu/projects/move2hear.

Given only a few glimpses of an environment, how much can we infer about its entire floorplan? Existing methods can map only what is visible or immediately apparent from context, and thus require substantial movements through a space to fully map it. We explore how both audio and visual sensing together can provide rapid floorplan reconstruction from limited viewpoints. Audio not only helps sense geometry outside the camera’s field of view, but it also reveals the existence of distant freespace (e.g., a dog barking in another room) and suggests the presence of rooms not visible to the camera (e.g., a dishwasher humming in what must be the kitchen to the left). We introduce AV-Map, a novel multi-modal encoder-decoder framework that reasons jointly about audio and vision to reconstruct a floorplan from a short input video sequence. We train our model to predict both the interior structure of the environment and the associated rooms’ semantic labels. Our results on 85 large real-world environments show the impact: with just a few glimpses spanning 26% of an area, we can estimate the whole area with 66% accuracy—substantially better than the state of the art approach for extrapolating visual maps.

We present a multiview pseudo-labeling approach to video learning, a novel framework that uses complementary views in the form of appearance and motion information for semi-supervised learning in video. The complementary views help obtain more reliable "pseudo-labels" on unlabeled video, to learn stronger video representations than from purely supervised data. Though our method capitalizes on multiple views, it nonetheless trains a model that is shared across appearance and motion input and thus, by design, incurs no additional computation overhead at inference time. On multiple video recognition datasets, our method substantially outperforms its supervised counterpart, and compares favorably to previous work on standard benchmarks in self-supervised video representation learning.

Dexterous robotic hands are appealing for their agility and human-like morphology, yet their high degree of freedom makes learning to manipulate challenging. We introduce an approach for learning dexterous grasping. Our key idea is to embed an object-centric visual affordance model within a deep reinforcement learning loop to learn grasping policies that favor the same object regions favored by people. Unlike traditional approaches that learn from human demonstration trajectories (e.g., hand joint sequences captured with a glove), the proposed prior is object-centric and image-based, allowing the agent to anticipate useful affordance regions for objects unseen during policy learning. We demonstrate our idea with a 30-DoF five-fingered robotic hand simulator on 40 objects from two datasets, where it successfully and efficiently learns policies for stable functional grasps. Our affordance-guided policies are significantly more effective, generalize better to novel objects, train 3× faster than the baselines, and are more robust to noisy sensor readings and actuation. Our work offers a step towards manipulation agents that learn by watching how people use objects, without requiring state and action information about the human body. Project website with videos: http://vision.cs.utexas.edu/projects/graff-dexterous-affordance-grasp.

Complex physical tasks entail a sequence of object interactions, each with its own preconditions -- which can be difficult for robotic agents to learn efficiently solely through their own experience. We introduce an approach to discover activity-context priors from in-the-wild egocentric video captured with human worn cameras. For a given object, an activity-context prior represents the set of other compatible objects that are required for activities to succeed (e.g., a knife and cutting board brought together with a tomato are conducive to cutting). We encode our video-based prior as an auxiliary reward function that encourages an agent to bring compatible objects together before attempting an interaction. In this way, our model translates everyday human experience into embodied agent skills. We demonstrate our idea using egocentric EPIC-Kitchens video of people performing unscripted kitchen activities to benefit virtual household robotic agents performing various complex tasks in AI2-iTHOR, significantly accelerating agent learning.

Standard video encoders developed for conventional narrow field-of-view video are widely applied to 360°video as well, with reasonable results. However, while this approach commits arbitrarily to a projection of the spherical frames, we observe that some orientations of a 360°video, once projected, are more compressible than others. We introduce an approach to predict the sphere rotation that will yield the maximal compression rate. Given video clips in their original encoding, a convolutional neural network learns the association between a clip's visual content and its compressibility at different rotations of a cubemap projection. Given a novel video, our learning-based approach efficiently infers the most compressible direction in one shot, without repeated rendering and compression of the source video. We validate our idea on thousands of video clips and multiple popular video codecs. The results show that this untapped dimension of 360°compression has substantial potential-“good” rotations are typically 8-18 percent more compressible than bad ones, and our learning approach can predict them reliably 78 percent of the time.

The evolution of clothing styles and their migration across the world is intriguing, yet difficult to describe quantitatively. We propose to discover and quantify fashion influences from catalog and social media photos. We explore fashion influence along two channels: geolocation and fashion brands. We introduce an approach that detects which of these entities influence which other entities in terms of propagating their styles. We then leverage the discovered influence patterns to inform a novel forecasting model that predicts the future popularity of any given style within any given city or brand. To demonstrate our idea, we leverage public large-scale datasets of 7.7M Instagram photos from 44 major world cities (where styles are worn with variable frequency) as well as 41K Amazon product photos (where styles are purchased with variable frequency). Our model learns directly from the image data how styles move between locations and how certain brands affect each other's designs in a predictable way. The discovered influence relationships reveal how both cities and brands exert and receive fashion influence for an array of visual styles inferred from the images. Furthermore, the proposed forecasting model achieves state-of-the-art results for challenging style forecasting tasks. Our results indicate the advantage of grounding visual style evolution both spatially and temporally, and for the first time, they quantify the propagation of inter-brand and inter-city influences. Project page:https://www.cs.utexas.edu/~ziad/influence_from_photos.html

The evolution of clothing styles and their migration across the world is intriguing, yet difficult to describe quantitatively. We propose to discover and quantify fashion influences from everyday images of people wearing clothes. We introduce an approach that detects which cities influence which other cities in terms of propagating their styles. We then leverage the discovered influence patterns to inform a forecasting model that predicts the popularity of any given style at any given city into the future. Demonstrating our idea with GeoStyle-a large-scale dataset of 7.7M images covering 44 major world cities, we present the discovered influence relationships, revealing how cities exert and receive fashion influence for an array of 50 observed visual styles. Furthermore, the proposed forecasting model achieves state-of-the-art results for a challenging style forecasting task, showing the advantage of grounding visual style evolution both spatially and temporally.

In the face of the video data deluge, today's expensive clip-level classifiers are increasingly impractical. We propose a framework for efficient action recognition in untrimmed video that uses audio as a preview mechanism to eliminate both short-term and long-term visual redundancies. First, we devise an ImgAud2Vid framework that hallucinates clip-level features by distilling from lighter modalities---a single frame and its accompanying audio---reducing short-term temporal redundancy for efficient clip-level recognition. Second, building on ImgAud2Vid, we further propose ImgAud-Skimming, an attention-based long short-term memory network that iteratively selects useful moments in untrimmed videos, reducing long-term temporal redundancy for efficient video-level recognition. Extensive experiments on four action recognition datasets demonstrate that our method achieves the state-of-the-art in terms of both recognition accuracy and speed.

Body shape plays an important role in determining what garments will best suit a given person, yet today's clothing recommendation methods take a “one shape fits all” approach. These body-agnostic vision methods and datasets are a barrier to inclusion, ill-equipped to provide good suggestions for diverse body shapes. We introduce ViBE, a VIsual Body-aware Embedding that captures clothing's affinity with different body shapes. Given an image of a person, the proposed embedding identifies garments that will flatter her specific body shape. We show how to learn the embedding from an online catalog displaying fashion models of various shapes and sizes wearing the products, and we devise a method to explain the algorithm's suggestions for well-fitting garments. We apply our approach to a dataset of diverse subjects, and demonstrate its strong advantages over status quo body-agnostic recommendation, both according to automated metrics and human opinion.

First-person video naturally brings the use of a physical environment to the forefront, since it shows the camera wearer interacting fluidly in a space based on his intentions. However, current methods largely separate the observed actions from the persistent space itself. We introduce a model for environment affordances that is learned directly from egocentric video. The main idea is to gain a human-centric model of a physical space (such as a kitchen) that captures (1) the primary spatial zones of interaction and (2) the likely activities they support. Our approach decomposes a space into a topological map derived from first-person activity, organizing an ego-video into a series of visits to the different zones. Further, we show how to link zones across multiple related environments (e.g., from videos of multiple kitchens) to obtain a consolidated representation of environment functionality. On EPIC-Kitchens and EGTEA+, we demonstrate our approach for learning scene affordances and anticipating future actions in long-form video.

The body pose of a person wearing a camera is of great interest for applications in augmented reality, healthcare, and robotics, yet much of the person's body is out of view for a typical wearable camera. We propose a learning-based approach to estimate the camera wearer's 3D body pose from egocentric video sequences. Our key insight is to leverage interactions with another person---whose body pose we can directly observe---as a signal inherently linked to the body pose of the first-person subject. We show that since interactions between individuals often induce a well-ordered series of back-and-forth responses, it is possible to learn a temporal model of the interlinked poses even though one party is largely out of view. We demonstrate our idea on a variety of domains with dyadic interaction and show the substantial impact on egocentric body pose estimation, which improves the state of the art.

Existing models often leverage co-occurrences between objects and their context to improve recognition accuracy. However, strongly relying on context risks a model's generalizability, especially when typical co-occurrence patterns are absent. This work focuses on addressing such contextual biases to improve the robustness of the learnt feature representations. Our goal is to accurately recognize a category in the absence of its context, without compromising on performance when it co-occurs with context. Our key idea is to decorrelate feature representations of a category from its co-occurring context. We achieve this by learning a feature subspace that explicitly represents categories occurring in the absence of context along side a joint feature subspace that represents both categories and context. Our very simple yet effective method is extensible to two multi-label tasks -- object and attribute classification. On 4 challenging datasets, we demonstrate the effectiveness of our method in reducing contextual bias.

Embodied agents operating in human spaces must be able to master how their environment works: what objects can the agent use, and how can it use them? We introduce a reinforcement learning approach for exploration for interaction, whereby an embodied agent autonomously discovers the affordance landscape of a new unmapped 3D environment (such as an unfamiliar kitchen). Given an egocentric RGB-D camera and a high-level action space, the agent is rewarded for maximizing successful interactions while simultaneously training an image-based affordance segmentation model. The former yields a policy for acting efficiently in new environments to prepare for downstream interaction tasks, while the latter yields a convolutional neural network that maps image regions to the likelihood they permit each action, densifying the rewards for exploration. We demonstrate our idea with AI2-iTHOR. The results show agents can learn how to use new home environments intelligently and that it prepares them to rapidly address various downstream tasks like "find a knife and put it in the drawer." Project page: http://vision.cs.utexas.edu/projects/interaction-exploration/

The fashion domain is a magnet for computer vision. New vision problems are emerging in step with the fashion industry's rapid evolution towards an online, social, and personalized business. Style models, trend forecasting, interactive search, and recommendation all require visual understanding with rich detail and subtlety. As a result, research in this area is poised to have great influence on how people shop, how the fashion industry analyzes its enterprise, and how we model the cultural trends revealed by what people wear.

Binaural audio provides a listener with 3D sound sensation, allowing a rich perceptual experience of the scene. However, binaural recordings are scarcely available and require nontrivial expertise and equipment to obtain. We propose to convert common monaural audio into binaural audio by leveraging video. The key idea is that visual frames reveal significant spatial cues that, while explicitly lacking in the accompanying single-channel audio, are strongly linked to it. Our multi-modal approach recovers this link from unlabeled video. We devise a deep convolutional neural network that learns to decode the monaural (single-channel) soundtrack into its binaural counterpart by injecting visual information about object and scene configurations. We call the resulting output 2.5D visual sound-the visual stream helps "lift" the flat single channel audio into spatialized sound. In addition to sound generation, we show the self-supervised representation learned by our network benefits audio-visual source separation. Our video results: http://vision.cs.utexas.edu/projects/2.5D_visual_sound/.

Transfer learning, which allows a source task to affect the inductive bias of the target task, is widely used in computer vision. The typical way of conducting transfer learning with deep neural networks is to fine-tune a model pretrained on the source task using data from the target task. In this paper, we propose an adaptive fine-tuning approach, called SpotTune, which finds the optimal fine-tuning strategy per instance for the target data. In SpotTune, given an image from the target task, a policy network is used to make routing decisions on whether to pass the image through the fine-tuned layers or the pre-trained layers. We conduct extensive experiments to demonstrate the effectiveness of the proposed approach. Our method outperforms the traditional fine-tuning approach on 12 out of 14 standard datasets. We also compare SpotTune with other state-of-the-art fine-tuning strategies, showing superior performance. On the Visual Decathlon datasets, our method achieves the highest score across the board without bells and whistles.

Ideally, 360° imagery could inherit the deep convolutional neural networks (CNNs) already trained with great success on perspective projection images. However, existing methods to transfer CNNs from perspective to spherical images introduce significant computational costs and/or degradations in accuracy. We present the Kernel Transformer Network (KTN) to efficiently transfer convolution kernels from perspective images to the equirectangular projection of 360° images. Given a source CNN for perspective images as input, the KTN produces a function parameterized by a polar angle and kernel as output. Given a novel 360° image, that function in turn can compute convolutions for arbitrary layers and kernels as would the source CNN on the corresponding tangent plane projections. Distinct from all existing methods, KTNs allow model transfer: the same model can be applied to different source CNNs with the same base architecture. This enables application to multiple recognition tasks without re-training the KTN. Validating our approach with multiple source CNNs and datasets, we show that KTNs improve the state of the art for spherical convolution. KTNs successfully preserve the source CNN's accuracy, while offering transferability, scalability to typical image resolutions, and, in many cases, a substantially lower memory footprint.

Highlight detection has the potential to significantly ease video browsing, but existing methods often suffer from expensive supervision requirements, where human viewers must manually identify highlights in training videos. We propose a scalable unsupervised solution that exploits video duration as an implicit supervision signal. Our key insight is that video segments from shorter user-generated videos are more likely to be highlights than those from longer videos, since users tend to be more selective about the content when capturing shorter videos. Leveraging this insight, we introduce a novel ranking framework that prefers segments from shorter videos, while properly accounting for the inherent noise in the (unlabeled) training data. We use it to train a highlight detector with 10M hashtagged Instagram videos. In experiments on two challenging public video highlight detection benchmarks, our method substantially improves the state-of-the-art for unsupervised highlight detection.

Estimating the relative rigid pose between two RGB-D scans of the same underlying environment is a fundamental problem in computer vision, robotics, and computer graphics. Most existing approaches allow only limited maximum relative pose changes since they require considerable overlap between the input scans. We introduce a novel approach that extends the scope to extreme relative poses, with little or even no overlap between the input scans. The key idea is to infer more complete scene information about the underlying environment and match on the completed scans. In particular, instead of only performing scene completion from each individual scan, our approach alternates between relative pose estimation and scene completion. This allows us to perform scene completion by utilizing information from both input scans at late iterations, resulting in better results for both scene completion and relative pose estimation. Experimental results on benchmark datasets show that our approach leads to considerable improvements over state-of-the-art approaches for relative pose estimation. In particular, our approach provides encouraging relative pose estimates even between non-overlapping scans.

Current wisdom suggests more labeled image data is always better, and obtaining labels is the bottleneck. Yet curating a pool of sufficiently diverse and informative images is itself a challenge. In particular, training image curation is problematic for fine-grained attributes, where the subtle visual differences of interest may be rare within traditional image sources. We propose an active image generation approach to address this issue. The main idea is to jointly learn the attribute ranking task while also learning to generate novel realistic image samples that will benefit that task. We introduce an end-to-end framework that dynamically "imagines" image pairs that would confuse the current model, presents them to human annotators for labeling, then improves the predictive model with the new examples. On two datasets, we show that by thinking outside the pool of real images, our approach gains generalization accuracy on challenging fine-grained attribute comparisons.

Learning how objects sound from video is challenging, since they often heavily overlap in a single audio channel. Current methods for visually-guided audio source separation sidestep the issue by training with artificially mixed video clips, but this puts unwieldy restrictions on training data collection and may even prevent learning the properties of "true" mixed sounds. We introduce a co-separation training paradigm that permits learning object-level sounds from unlabeled multi-source videos. Our novel training objective requires that the deep neural network's separated audio for similar-looking objects be consistently identifiable, while simultaneously reproducing accurate video-level audio tracks for each source training pair. Our approach disentangles sounds in realistic test videos, even in cases where an object was not observed individually during training. We obtain state-of-the-art results on visually-guided audio source separation and audio denoising for the MUSIC, AudioSet, and AV-Bench datasets.

Given an outfit, what small changes would most improve its fashionability? This question presents an intriguing new computer vision challenge. We introduce Fashion++, an approach that proposes minimal adjustments to a full-body clothing outfit that will have maximal impact on its fashionability. Our model consists of a deep image generation neural network that learns to synthesize clothing conditioned on learned per-garment encodings. The latent encodings are explicitly factorized according to shape and texture, thereby allowing direct edits for both fit/presentation and color/patterns/material, respectively. We show how to bootstrap Web photos to automatically train a fashionability model, and develop an activation maximization-style approach to transform the input image into its more fashionable self. The edits suggested range from swapping in a new garment to tweaking its color, how it is worn (e.g., rolling up sleeves), or its fit (e.g., making pants baggier). Experiments demonstrate that Fashion++ provides successful edits, both according to automated metrics and human opinion.

Learning how to interact with objects is an important step towards embodied visual intelligence, but existing techniques suffer from heavy supervision or sensing requirements. We propose an approach to learn human-object interaction "hotspots" directly from video. Rather than treat affordances as a manually supervised semantic segmentation task, our approach learns about interactions by watching videos of real human behavior and anticipating afforded actions. Given a novel image or video, our model infers a spatial hotspot map indicating where an object would be manipulated in a potential interaction even if the object is currently at rest. Through results with both first and third person video, we show the value of grounding affordances in real human-object interactions. Not only are our weakly supervised hotspots competitive with strongly supervised affordance methods, but they can also anticipate object interaction for novel object categories. Project page: http://vision.cs.utexas.edu/projects/interaction-hotspots/.

Visual recognition systems mounted on autonomous moving agents face the challenge of unconstrained data, but simultaneously have the opportunity to improve their performance by moving to acquire new views at test time. In this work, we first show how a recurrent neural network-based system may be trained to perform end-to-end learning of motion policies suited for this "active recognition" setting. Further, we hypothesize that active vision requires an agent to have the capacity to reason about the effects of its motions on its view of the world. To verify this hypothesis, we attempt to induce this capacity in our active recognition pipeline, by simultaneously learning to forecast the effects of the agent's motions on its internal representation of the environment conditional on all past views. Results across three challenging datasets confirm both that our end-to-end system successfully learns meaningful policies for active category recognition, and that "learning to look ahead" further boosts recognition performance.

We propose an end-to-end learning framework for segmenting generic objects in both images and videos. Given a novel image or video, our approach produces a pixel-level mask for all "object-like" regions-even for object categories never seen during training. We formulate the task as a structured prediction problem of assigning an object/background label to each pixel, implemented using a deep fully convolutional network. When applied to a video, our model further incorporates a motion stream, and the network learns to combine both appearance and motion and attempts to extract all prominent objects whether they are moving or not. Beyond the core model, a second contribution of our approach is how it leverages varying strengths of training annotations. Pixel-level annotations are quite difficult to obtain, yet crucial for training a deep network approach for segmentation. Thus we propose ways to exploit weakly labeled data for learning dense foreground segmentation. For images, we show the value in mixing object category examples with image-level labels together with relatively few images with boundary-level annotations. For video, we show how to bootstrap weakly annotated videos together with the network trained for image segmentation. Through experiments on multiple challenging image and video segmentation benchmarks, our method offers consistently strong results and improves the state-of-the-art for fully automatic segmentation of generic (unseen) objects. In addition, we demonstrate how our approach benefits image retrieval and image retargeting, both of which flourish when given our high-quality foreground maps. Code, models, and videos are at: http://vision.cs.utexas.edu/projects/pixelobjectness/.

Our interviews with people who have visual impairments show clothes shopping is an important activity in their lives. Unfortunately, clothes shopping web sites remain largely inaccessible. We propose design recommendations to address online accessibility issues reported by visually impaired study participants and an implementation, which we call BrowseWithMe, to address these issues. BrowseWithMe employs artificial intelligence to automatically convert a product web page into a structured representation that enables a user to interactively ask the BrowseWithMe system what the user wants to learn about a product (e.g., What is the price? Can I see a magnified image of the pants?). This enables people to be active solicitors of the specific information they are seeking rather than passive listeners of unparsed information. Experiments demonstrate BrowseWithMe can make online clothes shopping more accessible and produce accurate image descriptions.

Relative attribute models can compare images in terms of all detected properties or attributes, exhaustively predicting which image is fancier, more natural, and so on without any regard to ordering. However, when humans compare images, certain differences will naturally stick out and come to mind first. These most noticeable differences, or prominent differences, are likely to be described first. In addition, many differences, although present, may not be mentioned at all. In this work, we introduce and model prominent differences, a rich new functionality for comparing images. We collect instance-level annotations of most noticeable differences, and build a model trained on relative attribute features that predicts prominent differences for unseen pairs. We test our model on the challenging UT-Zap50K shoes and LFW10 faces datasets, and outperform an array of baseline methods. We then demonstrate how our prominence model improves two vision tasks, image search and description generation, enabling more natural communication between people and vision systems.

Existing methods to recognize actions in static images take the images at their face value, learning the appearances-objects, scenes, and body poses-that distinguish each action class. However, such models are deprived of the rich dynamic structure and motions that also define human activity. We propose an approach that hallucinates the unobserved future motion implied by a single snapshot to help static-image action recognition. The key idea is to learn a prior over short-term dynamics from thousands of unlabeled videos, infer the anticipated optical flow on novel static images, and then train discriminative models that exploit both streams of information. Our main contributions are twofold. First, we devise an encoder-decoder convolutional neural network and a novel optical flow encoding that can translate a static image into an accurate flow map. Second, we show the power of hallucinated flow for recognition, successfully transferring the learned motion into a standard two-stream network for activity recognition. On seven datasets, we demonstrate the power of the approach. It not only achieves state-of-the-art accuracy for dense optical flow prediction, but also consistently enhances recognition of actions and dynamic scenes.

The study of algorithms to automatically answer visual questions currently is motivated by visual question answering (VQA) datasets constructed in artificial VQA settings. We propose VizWiz, the first goal-oriented VQA dataset arising from a natural VQA setting. VizWiz consists of over 31,000 visual questions originating from blind people who each took a picture using a mobile phone and recorded a spoken question about it, together with 10 crowdsourced answers per visual question. VizWiz differs from the many existing VQA datasets because (1) images are captured by blind photographers and so are often poor quality, (2) questions are spoken and so are more conversational, and (3) often visual questions cannot be answered. Evaluation of modern algorithms for answering visual questions and deciding if a visual question is answerable reveals that VizWiz is a challenging dataset. We introduce this dataset to encourage a larger community to develop more generalized algorithms that can assist blind people.

We propose to automatically create capsule wardrobes. Given an inventory of candidate garments and accessories, the algorithm must assemble a minimal set of items that provides maximal mix-and-match outfits. We pose the task as a subset selection problem. To permit efficient subset selection over the space of all outfit combinations, we develop submodular objective functions capturing the key ingredients of visual compatibility, versatility, and user-specific preference. Since adding garments to a capsule only expands its possible outfits, we devise an iterative approach to allow near-optimal submodular function maximization. Finally, we present an unsupervised approach to learn visual compatibility from "in the wild" full body outfit photos; the compatibility metric translates well to cleaner catalog photos and improves over existing methods. Our results on thousands of pieces from popular fashion websites show that automatic capsule creation has potential to mimic skilled fashionistas in assembling flexible wardrobes, while being significantly more scalable.

It is common to implicitly assume access to intelligently captured inputs (e.g., photos from a human photographer), yet autonomously capturing good observations is itself a major challenge. We address the problem of learning to look around: if an agent has the ability to voluntarily acquire new views to observe its environment, how can it learn efficient exploratory behaviors to acquire informative visual observations? We propose a reinforcement learning solution, where the agent is rewarded for actions that reduce its uncertainty about the unobserved portions of its environment. Based on this principle, we develop a recurrent neural network-based approach to perform active completion of panoramic natural scenes and 3D object shapes. Crucially, the learned policies are not tied to any recognition task nor to the particular semantic content seen during training. As a result, 1) the learned "look around" behavior is relevant even for new tasks in unseen environments, and 2) training data acquisition involves no manual labeling. Through tests in diverse settings, we demonstrate that our approach learns useful generic policies that transfer to new unseen tasks and environments.

Standard video encoders developed for conventional narrow field-of-view video are widely applied to 360° video as well, with reasonable results. However, while this approach commits arbitrarily to a projection of the spherical frames, we observe that some orientations of a 360° video, once projected, are more compressible than others. We introduce an approach to predict the sphere rotation that will yield the maximal compression rate. Given video clips in their original encoding, a convolutional neural network learns the association between a clip's visual content and its compressibility at different rotations of a cubemap projection. Given a novel video, our learning-based approach efficiently infers the most compressible direction in one shot, without repeated rendering and compression of the source video. We validate our idea on thousands of video clips and multiple popular video codecs. The results show that this untapped dimension of 360° compression has substantial potential-"good" rotations are typically 8-10% more compressible than bad ones, and our learning approach can predict them reliably 82% of the time.

Very deep convolutional neural networks offer excellent recognition results, yet their computational expense limits their impact for many real-world applications. We introduce BlockDrop, an approach that learns to dynamically choose which layers of a deep network to execute during inference so as to best reduce total computation without degrading prediction accuracy. Exploiting the robustness of Residual Networks (ResNets) to layer dropping, our framework selects on-the-fly which residual blocks to evaluate for a given novel image. In particular, given a pretrained ResNet, we train a policy network in an associative reinforcement learning setting for the dual reward of utilizing a minimal number of blocks while preserving recognition accuracy. We conduct extensive experiments on CIFAR and ImageNet. The results provide strong quantitative and qualitative evidence that these learned policies not only accelerate inference but also encode meaningful visual information. Built upon a ResNet-101 model, our method achieves a speedup of 20% on average, going as high as 36% for some images, while maintaining the same 76.4% top-1 accuracy on ImageNet.

Visual questions (VQs) can lead multiple people to respond with different answers rather than a single, agreed upon response. Moreover, the answers from a crowd can include different numbers of unique answers that arise with different relative frequencies. Such answer diversity arises for a variety of reasons including that VQs are subjective, difficult, or ambiguous. We propose a new problem of predicting the answer distribution that would be observed from a crowd for any given VQ; i.e., the number of unique answers and their relative frequencies. Our experiments confirm that the answer distribution can be predicted accurately for VQs asked by both blind and sighted people. We then propose a novel crowd-powered VQA system that uses the answer distribution predictions to reason about how many answers are needed to capture the diversity of possible human responses. Experiments demonstrate this proposed system accelerates capturing the diversity of answers with considerably less human effort than is required with a state-of-art system.

Visual question answering systems empower users to ask any question about any image and receive a valid answer. However, existing systems do not yet account for the fact that a visual question can lead to a single answer or multiple different answers. While a crowd often agrees, disagreements do arise for many reasons including that visual questions are ambiguous, subjective, or difficult. We propose a model, CrowdVerge, for automatically predicting from a visual question whether a crowd would agree on one answer. We then propose how to exploit these predictions in a novel application to efficiently collect all valid answers to visual questions. Specifically, we solicit fewer human responses when answer agreement is expected and more human responses otherwise. Experiments on 121,811 visual questions asked by sighted and blind people show that, compared to existing crowdsourcing systems, our system captures the same answer diversity with typically 14-23% less crowd involvement.

We propose an end-to-end learning framework for segmenting generic objects in videos. Our method learns to combine appearance and motion information to produce pixel level segmentation masks for all prominent objects in videos. We formulate this task as a structured prediction problem and design a two-stream fully convolutional neural network which fuses together motion and appearance in a unified framework. Since large-scale video datasets with pixel level segmentations are problematic, we show how to bootstrap weakly annotated videos together with existing image recognition datasets for training. Through experiments on three challenging video segmentation benchmarks, our method substantially improves the state-of-the-art for segmenting generic (unseen) objects. Code and pre-trained models are available on the project website.

Todays person detection methods work best when people are in common upright poses and appear reasonably well spaced out in the image. However, in many real images, thats not what people do. People often appear quite close to each other, e.g., with limbs linked or heads touching, and their poses are often not pedestrian-like. We propose an approach to detangle people in multi-person images. We formulate the task as a region assembly problem. Starting from a large set of overlapping regions from body part semantic segmentation and generic object proposals, our optimization approach reassembles those pieces together into multiple person instances. Since optimal region assembly is a challenging combinatorial problem, we present a Lagrangian relaxation method to accelerate the lower bound estimation, thereby enabling a fast branch and bound solution for the global optimum. As output, our method produces a pixel-level map indicating both 1) the body part labels (arm, leg, torso, and head), and 2) which parts belong to which individual person. Our results on challenging datasets show our method is robust to clutter, occlusion, and complex poses. It outperforms a variety of competing methods, including existing detector CRF methods and region CNN approaches. In addition, we demonstrate its impact on a proxemics recognition task, which demands a precise representation of whose body part is where in crowded images.

Understanding the camera wearers activity is central to egocentric vision, yet one key facet of that activity is inherently invisible to the camera-the wearers body pose. Prior work focuses on estimating the pose of hands and arms when they come into view, but this 1) gives an incomplete view of the full body posture, and 2) prevents any pose estimate at all in many frames, since the hands are only visible in a fraction of daily life activities. We propose to infer the invisible pose of a person behind the egocentric camera. Given a single video, our efficient learning-based approach returns the full body 3D joint positions for each frame. Our method exploits cues from the dynamic motion signatures of the surrounding scene-which change predictably as a function of body pose-as well as static scene structures that reveal the viewpoint (e.g., sitting vs. standing). We further introduce a novel energy minimization scheme to infer the pose sequence. It uses soft predictions of the poses per time instant together with a non-parametric model of human pose dynamics over longer windows. Our method outperforms an array of possible alternatives, including typical deep learning approaches for direct pose regression from images.

360° Video requires human viewers to actively control where to look while watching the video. Although it provides a more immersive experience of the visual content, it also introduces additional burden for viewers, awkward interfaces to navigate the video lead to suboptimal viewing experiences. Virtual cinematography is an appealing direction to remedy these problems, but conventional methods are limited to virtual environments or rely on hand-crafted heuristics. We propose a new algorithm for virtual cinematography that automatically controls a virtual camera within a 360° video. Compared to the state of the art, our algorithm allows more general camera control, avoids redundant outputs, and extracts its output videos substantially more efficiently. Experimental results on over 7 hours of real in the wild video show that our generalized camera control is crucial for viewing 360° video, while the proposed efficient algorithm is essential for making the generalized control computationally tractable.

What is the future of fashion? Tackling this question from a data-driven vision perspective, we propose to forecast visual style trends before they occur. We introduce the first approach to predict the future popularity of styles discovered from fashion images in an unsupervised manner. Using these styles as a basis, we train a forecasting model to represent their trends over time. The resulting model can hypothesize new mixtures of styles that will become popular in the future, discover style dynamics (trendy vs. classic), and name the key visual attributes that will dominate tomorrow's fashion. We demonstrate our idea applied to three datasets encapsulating 80,000 fashion products sold across six years on Amazon. Results indicate that fashion forecasting benefits greatly from visual analysis, much more than textual or meta-data cues surrounding products.

While machine learning approaches to image restoration offer great promise, current methods risk training models fixated on performing well only for image corruption of a particular level of difficulty-such as a certain level of noise or blur. First, we examine the weakness of conventional “fixated” models and demonstrate that training general models to handle arbitrary levels of corruption is indeed non-trivial. Then, we propose an on-demand learning algorithm for training image restoration models with deep convolutional neural networks. The main idea is to exploit a feedback mechanism to self-generate training instances where they are needed most, thereby learning models that can generalize across difficulty levels. On four restoration tasks-image inpainting, pixel interpolation, image deblurring, and image denoising-and three diverse datasets, our approach consistently outperforms both the status quo training procedure and curriculum learning alternatives.

Distinguishing subtle differences in attributes is valuable, yet learning to make visual comparisons remains nontrivial. Not only is the number of possible comparisons quadratic in the number of training images, but also access to images adequately spanning the space of fine-grained visual differences is limited. We propose to overcome the sparsity of supervision problem via synthetically generated images. Building on a state-of-the-art image generation engine, we sample pairs of training images exhibiting slight modifications of individual attributes. Augmenting real training image pairs with these examples, we then train attribute ranking models to predict the relative strength of an attribute in novel pairs of real images. Our results on datasets of faces and fashion images show the great promise of bootstrapping imperfect image generators to counteract sample sparsity for learning to rank.

While 360° cameras offer tremendous new possibilities in vision, graphics, and augmented reality, the spherical images they produce make core feature extraction non-trivial. Convolutional neural networks (CNNs) trained on images from perspective cameras yield “flat" filters, yet 360° images cannot be projected to a single plane without significant distortion. A naive solution that repeatedly projects the viewing sphere to all tangent planes is accurate, but much too computationally intensive for real problems. We propose to learn a spherical convolutional network that translates a planar CNN to process 360° imagery directly in its equirectangular projection. Our approach learns to reproduce the flat filter outputs on 360° data, sensitive to the varying distortion effects across the viewing sphere. The key benefits are 1) efficient feature extraction for 360° images and video, and 2) the ability to leverage powerful pre-trained networks researchers have carefully honed (together with massive labeled image training sets) for perspective images. We validate our approach compared to several alternative methods in terms of both raw CNN output accuracy as well as applying a state-of-the-art “flat" object detector to 360° data. Our method yields the most accurate results while saving orders of magnitude in computation versus the existing exact reprojection solution.

We propose an efficient approach for activity detection in video that unifies activity categorization with space-time localization. The main idea is to pose activity detection as a maximum-weight connected subgraph problem. Offline, we learn a binary classifier for an activity category using positive video exemplars that are "trimmed" in time to the activity of interest. Then, given a novel untrimmed video sequence, we decompose it into a 3D array of space-time nodes, which are weighted based on the extent to which their component features support the learned activity model. To perform detection, we then directly localize instances of the activity by solving for the maximum-weight connected subgraph in the test video's space-time graph. We show that this detection strategy permits an efficient branch-and-cut solution for the best-scoring-and possibly non-cubically shaped-portion of the video for a given activity classifier. The upshot is a fast method that can search a broader space of space-time region candidates than was previously practical, which we find often leads to more accurate detection. We demonstrate the proposed algorithm on four datasets, and we show its speed and accuracy advantages over multiple existing search strategies.

Foreground object segmentation is a critical step for many image analysis tasks. While automated methods can produce high-quality results, their failures disappoint users in need of practical solutions. We propose a resource allocation framework for predicting how best to allocate a fixed budget of human annotation effort in order to collect higher quality segmentations for a given batch of images and automated methods. The framework is based on a proposed prediction module that estimates the quality of given algorithm-drawn segmentations. We demonstrate the value of the framework for two novel tasks related to "pulling the plug" on computer and human annotators. Specifically, we implement two systems that automatically decide, for a batch of images, when to replace 1) humans with computers to create coarse segmentations required to initialize segmentation tools and 2) computers with humans to create final, fine-grained segmentations. Experiments demonstrate the advantage of relying on a mix of human and computer efforts over relying on either resource alone for segmenting objects in three diverse datasets representing visible, phase contrast microscopy, and fluorescence microscopy images.

We propose a semi-automatic method to obtain foreground object masks for a large set of related images. We develop a stagewise active approach to propagation: in each stage, we actively determine the images that appear most valuable for human annotation, then revise the foreground estimates in all unlabeled images accordingly. In order to identify images that, once annotated, will propagate well to other examples, we introduce an active selection procedure that operates on the joint segmentation graph over all images. It prioritizes human intervention for those images that are uncertain and influential in the graph, while also mutually diverse. We apply our method to obtain foreground masks for over 1 million images. Our method yields state-of-the-art accuracy on the ImageNet and MIT Object Discovery datasets, and it focuses human attention more effectively than existing propagation strategies.

How can unlabeled video augment visual learning? Existing methods perform "slow" feature analysis, encouraging the representations of temporally close frames to exhibit only small differences. While this standard approach captures the fact that high-level visual signals change slowly over time, it fails to capture how the visual content changes. We propose to generalize slow feature analysis to "steady" feature analysis. The key idea is to impose a prior that higher order derivatives in the learned feature space must be small. To this end, we train a convolutional neural network with a regularizer on tuples of sequential frames from unlabeled video. It encourages feature changes over time to be smooth, i.e., similar to the most recent changes. Using five diverse datasets, including unlabeled YouTube and KITTI videos, we demonstrate our method's impact on object, scene, and action recognition tasks. We further show that our features learned from unlabeled video can even surpass a standard heavily supervised pretraining approach.

Video summarization has unprecedented importance to help us digest, browse, and search today's ever-growing video collections. We propose a novel subset selection technique that leverages supervision in the form of humancreated summaries to perform automatic keyframe-based video summarization. The main idea is to nonparametrically transfer summary structures from annotated videos to unseen test videos. We show how to extend our method to exploit semantic side information about the video's category/ genre to guide the transfer process by those training videos semantically consistent with the test input. We also show how to generalize our method to subshot-based summarization, which not only reduces computational costs but also provides more flexible ways of defining visual similarity across subshots spanning several frames. We conduct extensive evaluation on several benchmarks and demonstrate promising results, outperforming existing methods in several settings.

We present a novel form of interactive video object segmentation where a few clicks by the user helps the system produce a full spatio-temporal segmentation of the object of interest. Whereas conventional interactive pipelines take the user's initialization as a starting point, we show the value in the system taking the lead even in initialization. In particular, for a given video frame, the system precomputes a ranked list of thousands of possible segmentation hypotheses (also referred to as object region proposals) using image and motion cues. Then, the user looks at the top ranked proposals, and clicks on the object boundary to carve away erroneous ones. This process iterates (typically 2-3 times), and each time the system revises the top ranked proposal set, until the user is satisfied with a resulting segmentation mask. Finally, the mask is propagated across the video to produce a spatio-temporal object tube. On three challenging datasets, we provide extensive comparisons with both existing work and simpler alternative methods. In all, the proposed Click Carving approach strikes an excellent balance of accuracy and human effort. It outperforms all similarly fast methods, and is competitive or better than those requiring 2 to 12 times the effort.

Text detection in stores has valuable applications that could transform the shopping experience, yet cluttered store environments present distinct challenges for existing techniques. We propose a strategy for text detection in stores that exploits a repetition prior. Leveraging the fact that shops typically display multiple instances of the same product on the shelf, our approach localizes text regions with a global view of the image, preferring instances that have repeated support in the scene. On two challenging real-world datasets taken with a mobile phone and wearable camera, we demonstrate our method's substantial advantages compared to several state-of-the-art techniques in grocery store environments.

Understanding how images of objects and scenes behave in response to specific ego-motions is a crucial aspect of proper visual development, yet existing visual learning methods are conspicuously disconnected from the physical source of their images. We propose to exploit proprioceptive motor signals to provide unsupervised regularization in convolutional neural networks to learn visual representations from egocentric video. Specifically, we enforce that our learned features exhibit equivariance, i.e, they respond predictably to transformations associated with distinct ego-motions. With three datasets, we show that our unsupervised feature learning approach significantly outperforms previous approaches on visual recognition and next-best-view prediction tasks. In the most challenging test, we show that features learned from video captured on an autonomous driving platform improve large-scale scene recognition in static images from a disjoint domain.

We explore the problem of predicting "just noticeable differences" in a visual attribute. While some pairs of images have a clear ordering for an attribute (e.g., A is more sporty than B), for others the difference may be indistinguishable to human observers. However, existing relative attribute models are unequipped to infer partial orders on novel data. Attempting to map relative attribute ranks to equality predictions is non-trivial, particularly since the span of indistinguishable pairs in attribute space may vary in different parts of the feature space. We develop a Bayesian local learning strategy to infer when images are indistinguishable for a given attribute. On the UT-Zap50K shoes and LFW-10 faces datasets, we outperform a variety of alternative methods. In addition, we show the practical impact on fine-grained visual search.

We propose a dense local region detector to extract features suitable for image matching and object recognition tasks. Whereas traditional local interest operators rely on repeatable structures that often cross object boundaries (e.g., corners, scale-space blobs), our sampling strategy is driven by segmentation, and thus preserves object boundaries and shape. At the same time, whereas existing region-based representations are sensitive to segmentation parameters and object deformations, our novel approach to robustly sample dense sites and determine their connectivity offers better repeatability. In extensive experiments, we find that the proposed region detector provides significantly better repeatability and localization accuracy for object matching compared to an array of existing feature detectors. In addition, we show our regions lead to excellent results on two benchmark tasks that require good feature matching: weakly supervised foreground discovery and nearest neighbor-based object recognition.

The appearance of an attribute can vary considerably from class to class (e.g., a "fluffy" dog vs. a "fluffy" towel), making standard class-independent attribute models break down. Yet, training object-specific models for each attribute can be impractical, and defeats the purpose of using attributes to bridge category boundaries. We propose a novel form of transfer learning that addresses this dilemma. We develop a tensor factorization approach which, given a sparse set of class-specific attribute classifiers, can infer new ones for object-attribute pairs unobserved during training. For example, even though the system has no labeled images of striped dogs, it can use its knowledge of other attributes and objects to tailor "stripedness" to the dog category. With two large-scale datasets, we demonstrate both the need for category-sensitive attributes as well as our method's successful transfer. Our inferred attribute classifiers perform similarly well to those trained with the luxury of labeled class-specific instances, and much better than those restricted to traditional modes of transfer.

We pose unseen view synthesis as a probabilistic tensor completion problem. Given images of people organized by their rough viewpoint, we form a 3D appearance tensor indexed by images (pose examples), viewpoints, and image positions. After discovering the low-dimensional latent factors that approximate that tensor, we can impute its missing entries. In this way, we generate novel synthetic views of people -- even when they are observed from just one camera viewpoint. We show that the inferred views are both visually and quantitatively accurate. Furthermore, we demonstrate their value for recognizing actions in unseen views and estimating viewpoint in novel images. While existing methods are often forced to choose between data that is either realistic or multi-view, our virtual views offer both, thereby allowing greater robustness to viewpoint in novel images.

Existing methods to learn visual attributes are prone to learning the wrong thing -- namely, properties that are correlated with the attribute of interest among training samples. Yet, many proposed applications of attributes rely on being able to learn the correct semantic concept corresponding to each attribute. We propose to resolve such confusions by jointly learning decorrelated, discriminative attribute models. Leveraging side information about semantic relatedness, we develop a multi-task learning approach that uses structured sparsity to encourage feature competition among unrelated attributes and feature sharing among related attributes. On three challenging datasets, we show that accounting for structure in the visual attribute space is key to learning attribute models that preserve semantics, yielding improved generalizability that helps in the recognition and discovery of unseen object categories.

It is useful to automatically compare images based on their visual properties - to predict which image is brighter, more feminine, more blurry, etc. However, comparative models are inherently more costly to train than their classification counterparts. Manually labeling all pairwise comparisons is intractable, so which pairs should a human supervisor compare? We explore active learning strategies for training relative attribute ranking functions, with the goal of requesting human comparisons only where they are most informative. We introduce a novel criterion that requests a partial ordering for a set of examples that minimizes the total rank margin in attribute space, subject to a visual diversity constraint. The setwise criterion helps amortize effort by identifying mutually informative comparisons, and the diversity requirement safeguards against requests a human viewer will find ambiguous. We develop an efficient strategy to search for sets that meet this criterion. On three challenging datasets and experiments with "live" online annotators, the proposed method outperforms both traditional passive learning as well as existing active rank learning methods.

Given two images, we want to predict which exhibits a particular visual attribute more than the other-even when the two images are quite similar. Existing relative attribute methods rely on global ranking functions; yet rarely will the visual cues relevant to a comparison be constant for all data, nor will humans' perception of the attribute necessarily permit a global ordering. To address these issues, we propose a local learning approach for fine-grained visual comparisons. Given a novel pair of images, we learn a local ranking model on the fly, using only analogous training comparisons. We show how to identify these analogous pairs using learned metrics. With results on three challenging datasets-including a large newly curated dataset for fine-grained comparisons-our method outperforms stateof-the-art methods for relative attribute prediction.

Biographical details and talk summaries for the keynote talks presented at the AAAI Conference on Human Computation and Crowdsourcing.

We consider the problem of retrieving the database points nearest to a given hyperplane query without exhaustively scanning the entire database. For this problem, we propose two hashing-based solutions. Our first approach maps the data to 2-bit binary keys that are locality sensitive for the angle between the hyperplane normal and a database point. Our second approach embeds the data into a vector space where the euclidean norm reflects the desired distance between the original points and hyperplane query. Both use hashing to retrieve near points in sublinear time. Our first method's preprocessing stage is more efficient, while the second has stronger accuracy guarantees. We apply both to pool-based active learning: Taking the current hyperplane classifier as a query, our algorithm identifies those points (approximately) satisfying the well-known minimal distance-to-hyperplane selection criterion. We empirically demonstrate our methods' tradeoffs and show that they make it practical to perform active selection with millions of unlabeled points.

Activities in egocentric video are largely defined by the objects with which the cam-era wearer interacts, making representations that summarize the objects in view quite informative. Beyond simply recording how frequently each object occurs in a single histogram, spatio-temporal binning approaches can capture the objects ’ relative layout and ordering. However, existing methods use hand-crafted binning schemes (e.g., a uni-formly spaced pyramid of partitions), which may fail to capture the relationships that best distinguish certain activities. We propose to learn the spatio-temporal partitions that are discriminative for a set of egocentric activity classes. We devise a boosting approach that automatically selects a small set of useful spatio-temporal pyramid histograms among a randomized pool of candidate partitions. In order to efficiently focus the candidate par-titions, we further propose an “object-centric ” cutting scheme that prefers sampling bin boundaries near those objects prominently involved in the egocentric activities. In this way, we specialize the randomized pool of partitions to the egocentric setting and im-prove the training efficiency for boosting. Our approach yields state-of-the-art accuracy for recognition of challenging activities of daily living. 1

We propose an approach to learn action categories from static images that leverages prior observations of generic human motion to augment its training process. Using unlabeled video containing various human activities, the system first learns how body pose tends to change locally in time. Then, given a small number of labeled static images, it uses that model to extrapolate beyond the given exemplars and generate "synthetic" training examples-poses that could link the observed images and/or immediately precede or follow them in time. In this way, we expand the training set without requiring additional manually labeled examples. We explore both example-based and manifold-based methods to implement our idea. Applying our approach to recognize actions in both images and video, we show it enhances a state-of-the-art technique when very few labeled training examples are available.

We introduce a fast deformable spatial pyramid (DSP) matching algorithm for computing dense pixel correspondences. Dense matching methods typically enforce both appearance agreement between matched pixels as well as geometric smoothness between neighboring pixels. Whereas the prevailing approaches operate at the pixel level, we propose a pyramid graph model that simultaneously regularizes match consistency at multiple spatial extents-ranging from an entire image, to coarse grid cells, to every single pixel. This novel regularization substantially improves pixel-level matching in the face of challenging image variations, while the "deformable" aspect of our model overcomes the strict rigidity of traditional spatial pyramids. Results on Label Me and Caltech show our approach outperforms state-of-the-art methods (SIFT Flow [15] and Patch-Match [2]), both in terms of accuracy and run time.

We present a video summarization approach that discovers the story of an egocentric video. Given a long input video, our method selects a short chain of video sub shots depicting the essential events. Inspired by work in text analysis that links news articles over time, we define a random-walk based metric of influence between sub shots that reflects how visual objects contribute to the progression of events. Using this influence metric, we define an objective for the optimal k-subs hot summary. Whereas traditional methods optimize a summary's diversity or representative ness, ours explicitly accounts for how one sub-event "leads to" another-which, critically, captures event connectivity beyond simple object co-occurrence. As a result, our summaries provide a better sense of story. We apply our approach to over 12 hours of daily activity video taken from 23 unique camera wearers, and systematically evaluate its quality compared to multiple baselines with 34 human subjects.

Collecting and annotating videos of realistic human actions is tedious, yet critical for training action recognition systems. We propose a method to actively request the most useful video annotations among a large set of unlabeled videos. Predicting the utility of annotating unlabeled video is not trivial, since any given clip may contain multiple actions of interest, and it need not be trimmed to temporal regions of interest. To deal with this problem, we propose a detection-based active learner to train action category models. We develop a voting-based framework to localize likely intervals of interest in an unlabeled clip, and use them to estimate the total reduction in uncertainty that annotating that clip would yield. On three datasets, we show our approach can learn accurate action detectors more efficiently than alternative active learning strategies that fail to accommodate the "untrimmed" nature of real video data.

The mode of manual annotation used in an interactive segmentation algorithm affects both its accuracy and ease-of-use. For example, bounding boxes are fast to supply, yet may be too coarse to get good results on difficult images, freehand outlines are slower to supply and more specific, yet they may be overkill for simple images. Whereas existing methods assume a fixed form of input no matter the image, we propose to predict the tradeoff between accuracy and effort. Our approach learns whether a graph cuts segmentation will succeed if initialized with a given annotation mode, based on the image's visual separability and foreground uncertainty. Using these predictions, we optimize the mode of input requested on new images a user wants segmented. Whether given a single image that should be segmented as quickly as possible, or a batch of images that must be segmented within a specified time budget, we show how to select the easiest modality that will be sufficiently strong to yield high quality segmentations. Extensive results with real users and three datasets demonstrate the impact.

In interactive image search, a user iteratively refines his results by giving feedback on exemplar images. Active selection methods aim to elicit useful feedback, but traditional approaches suffer from expensive selection criteria and cannot predict in formativeness reliably due to the imprecision of relevance feedback. To address these drawbacks, we propose to actively select "pivot" exemplars for which feedback in the form of a visual comparison will most reduce the system's uncertainty. For example, the system might ask, "Is your target image more or less crowded than this image?" Our approach relies on a series of binary search trees in relative attribute space, together with a selection function that predicts the information gain were the user to compare his envisioned target to the next node deeper in a given attribute's tree. It makes interactive search more efficient than existing strategies-both in terms of the system's selection time as well as the user's feedback effort.

Current methods learn monolithic attribute predictors, with the assumption that a single model is sufficient to reflect human understanding of a visual attribute. However, in reality, humans vary in how they perceive the association between a named property and image content. For example, two people may have slightly different internal models for what makes a shoe look "formal", or they may disagree on which of two scenes looks "more cluttered". Rather than discount these differences as noise, we propose to learn user-specific attribute models. We adapt a generic model trained with annotations from multiple users, tailoring it to satisfy user-specific labels. Furthermore, we propose novel techniques to infer user-specific labels based on transitivity and contradictions in the user's search history. We demonstrate that adapted attributes improve accuracy over both existing monolithic models as well as models that learn from scratch with user-specific data alone. In addition, we show how adapted attributes are useful to personalize image search, whether with binary or relative attributes.

User feedback helps an image search system refine its relevance predictions, tailoring the search towards the user's preferences. Existing methods simply take feedback at face value: clicking on an image means the user wants things like it, commenting that an image lacks a specific attribute means the user wants things that have it. However, we expect there is actually more information behind the user's literal feedback. In particular, a user's (possibly subconscious) search strategy leads him to comment on certain images rather than others, based on how any of the visible candidate images compare to the desired content. For example, he may be more likely to give negative feedback on an irrelevant image that is relatively close to his target, as opposed to bothering with one that is altogether different. We introduce novel features to capitalize on such implied feedback cues, and learn a ranking function that uses them to improve the system's relevance estimates. We validate the approach with real users searching for shoes, faces, or scenes using two different modes of feedback: binary relevance feedback and relative attributes-based feedback. The results show that retrieval improves significantly when the system accounts for the learned behaviors. We show that the nuances learned are domain-invariant, and useful for both generic user-independent search as well as personalized user-specific search.

Learning domain-invariant features is of vital importance to unsupervised domain adaptation, where classifiers trained on the source domain need to be adapted to a different target domain for which no labeled examples are available. In this paper, we propose a novel approach for learning such features. The central idea is to exploit the existence of landmarks, which are a subset of labeled data instances in the source domain that are distributed most similarly to the target domain. Our approach automatically discovers the landmarks and use them to bridge the source to the target by constructing provably easier auxiliary domain adaptation tasks. The solutions of those auxiliary tasks form the basis to compose invariant features for the original task. We show how this composition can be optimized discriminatively without requiring labels from the target domain. We validate the method on standard benchmark datasets for visual object recognition and sentiment analysis of text. Empirical results show the proposed method outperforms the state-of-the-art significantly.

In multi-class categorization tasks, knowledge about the classes’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the semantic distances between classes (e.g., according to a given object taxonomy for visual recognition), limiting the influence to pairwise structures. We propose to model \emphanalogies that reflect the relationships between multiple pairs of classes simultaneously, in the form “p is to q, as r is to s"". We translate semantic analogies into higher-order geometric constraints called \emphanalogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embedding. Furthermore, we show how to discover analogies from attribute-based class descriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over existing approaches, both in terms of recognition accuracy and analogy completion.

In visual recognition problems, the common data distribution mismatches between training and testing make domain adaptation essential. However, image data is difficult to manually divide into the discrete domains required by adaptation algorithms, and the standard practice of equating datasets with domains is a weak proxy for all the real conditions that alter the statistics in complex ways (lighting, pose, background, resolution, etc.) We propose an approach to automatically discover latent domains in image or video datasets. Our formulation imposes two key properties on domains: maximum distinctiveness and maximum learnability. By maximum distinctiveness, we require the underlying distributions of the identified domains to be different from each other; by maximum learnability, we ensure that a strong discriminative model can be learned from the domain. We devise a nonparametric representation and efficient optimization procedure for distinctiveness, which, when coupled with our learnability constraint, can successfully discover domains among both training and test data. We extensively evaluate our approach on object recognition and human activity recognition tasks.

Secure Computation of Face Identification (SCiFI) [20] is a recently developed secure face recognition system that ensures the list of faces it can identify (e.g., a terrorist watch list) remains private. In this work, we study the consequences of malformed input attacks on the system - from both a security and computer vision standpoint. In particular, we present 1) a cryptographic attack that allows a dishonest user to undetectably obtain a coded representation of faces on the list, and 2) a visualization approach that exploits this breach, turning the lossy recovered codes into human-identifiable face sketches. We evaluate our approach on two challenging datasets, with face identification tasks given to a computer and human subjects. Whereas prior work considered security in the setting of honest inputs and protocol execution, the success of our approach underscores the risk posed by malicious adversaries to todays automatic face recognition systems.

We propose to model relative attributes that capture the relationships between images and objects in terms of human-nameable visual properties. For example, the models can capture that animal A is 'furrier' than animal B, or image X is 'brighter' than image B. Given training data stating how object/scene categories relate according to different attributes, we learn a ranking function per attribute. The learned ranking functions predict the relative strength of each property in novel images. We show how these relative attribute predictions enable a variety of novel applications, including zero-shot learning from relative comparisons, automatic image description, image search with interactive feedback, and active learning of discriminative classifiers. We overview results demonstrating these applications with images of faces and natural scenes. Overall, we find that relative attributes enhance the precision of communication between humans and computer vision algorithms, providing the richer language needed to fluidly "teach" a system about visual concepts.

We propose an efficient approach that unifies activity categorization with space-time localization. The main idea is to pose activity detection as a maximum-weight connected subgraph problem over a learned space-time graph constructed on the test sequence. We show this permits an efficient branch-and-cut solution for the best-scoring - and possibly non-cubically shaped - portion of the video for a given activity classifier. The upshot is a fast method that can evaluate a broader space of candidates than was previously practical, which we find often leads to more accurate detection. We demonstrate the proposed algorithm on three datasets, and show its speed and accuracy advantages over multiple existing search strategies.

Attributes are visual concepts that can be detected by machines, understood by humans, and shared across categories. They are particularly useful for fine-grained domains where categories are closely related to one other (e.g. bird species recognition). In such scenarios, relevant attributes are often local (e.g. “white belly”), but the question of how to choose these local attributes remains largely unexplored. In this paper, we propose an interactive approach that discovers local attributes that are both discriminative and semantically meaningful from image datasets annotated only with fine-grained category labels and object bounding boxes. Our approach uses a latent conditional random field model to discover candidate attributes that are detectable and discriminative, and then employs a recommender system that selects attributes likely to be semantically meaningful. Human interaction is used to provide semantic names for the discovered attributes. We demonstrate our method on two challenging datasets, Caltech-UCSD Birds-200-2011 and Leeds Butterflies, and find that our discovered attributes outperform those generated by traditional approaches.

In real-world applications of visual recognition, many factors - such as pose, illumination, or image quality - can cause a significant mismatch between the source domain on which classifiers are trained and the target domain to which those classifiers are applied. As such, the classifiers often perform poorly on the target domain. Domain adaptation techniques aim to correct the mismatch. Existing approaches have concentrated on learning feature representations that are invariant across domains, and they often do not directly exploit low-dimensional structures that are intrinsic to many vision datasets. In this paper, we propose a new kernel-based method that takes advantage of such structures. Our geodesic flow kernel models domain shift by integrating an infinite number of subspaces that characterize changes in geometric and statistical properties from the source to the target domain. Our approach is computationally advantageous, automatically inferring important algorithmic parameters without requiring extensive cross-validation or labeled data from either domain. We also introduce a metric that reliably measures the adaptability between a pair of source and target domains. For a given target domain and several source domains, the metric can be used to automatically select the optimal source domain to adapt and avoid less desirable ones. Empirical studies on standard datasets demonstrate the advantages of our approach over competing methods.

We propose a novel mode of feedback for image search, where a user describes which properties of exemplar images should be adjusted in order to more closely match his/her mental model of the image(s) sought. For example, perusing image results for a query “black shoes”, the user might state, “Show me shoe images like these, but sportier.” Offline, our approach first learns a set of ranking functions, each of which predicts the relative strength of a nameable attribute in an image (`sportiness', `furriness', etc.). At query time, the system presents an initial set of reference images, and the user selects among them to provide relative attribute feedback. Using the resulting constraints in the multi-dimensional attribute space, our method updates its relevance function and re-ranks the pool of images. This procedure iterates using the accumulated constraints until the top ranked images are acceptably close to the user's envisioned target. In this way, our approach allows a user to efficiently “whittle away” irrelevant portions of the visual feature space, using semantic language to precisely communicate her preferences to the system. We demonstrate the technique for refining image search for people, products, and scenes, and show it outperforms traditional binary relevance feedback in terms of search speed and accuracy.

We present a video summarization approach for egocentric or “wearable” camera data. Given hours of video, the proposed method produces a compact storyboard summary of the camera wearer's day. In contrast to traditional keyframe selection techniques, the resulting summary focuses on the most important objects and people with which the camera wearer interacts. To accomplish this, we develop region cues indicative of high-level saliency in egocentric video — such as the nearness to hands, gaze, and frequency of occurrence — and learn a regressor to predict the relative importance of any new region based on these cues. Using these predictions and a simple form of temporal event detection, our method selects frames for the storyboard that reflect the key object-driven happenings. Critically, the approach is neither camera-wearer-specific nor object-specific; that means the learned importance metric need not be trained for a given user or context, and it can predict the importance of objects and people that have never been seen previously. Our results with 17 hours of egocentric data show the method's promise relative to existing techniques for saliency and summarization.

When learning features for complex visual recognition problems, labeled image exemplars alone can be insufficient. While an \emph{object taxonomy} specifying the categories' semantic relationships could bolster the learning process, not all relationships are relevant to a given visual classification task, nor does a single taxonomy capture all ties that \emph{are} relevant. In light of these issues, we propose a discriminative feature learning approach that leverages \emph{multiple} hierarchical taxonomies representing different semantic views of the object categories (e.g., for animal classes, one taxonomy could reflect their phylogenic ties, while another could reflect their habitats). For each taxonomy, we first learn a tree of semantic kernels, where each node has a Mahalanobis kernel optimized to distinguish between the classes in its children nodes. Then, using the resulting \emph{semantic kernel forest}, we learn class-specific kernel combinations to select only those relationships relevant to recognize each object class. To learn the weights, we introduce a novel hierarchical regularization term that further exploits the taxonomies' structure. We demonstrate our method on challenging object recognition datasets, and show that interleaving multiple taxonomic views yields significant accuracy improvements.

Current uses of tagged images typically exploit only the most explicit information: the link between the nouns named and the objects present somewhere in the image. We propose to leverage "unspoken" cues that rest within an ordered list of image tags so as to improve object localization. We define three novel implicit features from an image's tags-the relative prominence of each object as signified by its order of mention, the scale constraints implied by unnamed objects, and the loose spatial links hinted at by the proximity of names on the list. By learning a conditional density over the localization parameters (position and scale) given these cues, we show how to improve both accuracy and efficiency when detecting the tagged objects. Furthermore, we show how the localization density can be learned in a semantic space shared by the visual and tag-based features, which makes the technique applicable for detection in untagged input images. We validate our approach on the PASCAL VOC, LabelMe, and Flickr image data sets, and demonstrate its effectiveness relative to both traditional sliding windows as well as a visual context baseline. Our algorithm improves state-of-the-art methods, successfully translating insights about human viewing behavior (such as attention, perceived importance, or gaze) into enhanced object detection.

Fast retrieval methods are critical for many large-scale and data-driven vision applications. Recent work has explored ways to embed high-dimensional features or complex distance functions into a low-dimensional Hamming space where items can be efficiently searched. However, existing methods do not apply for high-dimensional kernelized data when the underlying feature embedding for the kernel is unknown. We show how to generalize locality-sensitive hashing to accommodate arbitrary kernel functions, making it possible to preserve the algorithm's sublinear time similarity search guarantees for a wide class of useful similarity functions. Since a number of successful image-based kernels have unknown or incomputable embeddings, this is especially valuable for image retrieval tasks. We validate our technique on several data sets, and show that it enables accurate and fast performance for several vision problems, including example-based object classification, local feature matching, and content-based retrieval.

How can knowing about some categories help us to discover new ones in unlabeled images? Unsupervised visual category discovery is useful to mine for recurring objects without human supervision, but existing methods assume no prior information and thus tend to perform poorly for cluttered scenes with multiple objects. We propose to leverage knowledge about previously learned categories to enable more accurate discovery, and address challenges in estimating their familiarity in unsegmented, unlabeled images. We introduce two variants of a novel object-graph descriptor to encode the 2D and 3D spatial layout of object-level co-occurrence patterns relative to an unfamiliar region and show that by using them to model the interaction between an image’s known and unknown objects, we can better detect new visual categories. Rather than mine for all categories from scratch, our method identifies new objects while drawing on useful cues from familiar ones. We evaluate our approach on several benchmark data sets and demonstrate clear improvements in discovery over conventional purely appearance-based baselines.

We present an approach to discover novel faces in untagged photo collections by leveraging the “social context” of co-occurring people. Our idea exploits the social nature of consumer photos, in which people of the same clique (family, team, class, friends) often appear together. Initially, the system trains detector s for any individuals with tagged instances in the collection. Then, for each untagged image, it isolates any unfamiliar faces. Among those, it discovers novel face clusters by leveraging both their appearance, as well as descriptors encoding the (predicted) familiar fa ces with which the unfamiliar faces co-occur. The resulting discovered people can then be presented to a user for nametagging, thereby efficiently propagating manually provide d labels. Our experiments with real consumer photo collections demonstrate that the system outperforms baseline approaches that either lack any social context model, or rely solely on the appearance of co-occurring faces. Furthermore, we show it can successfully use the discovered models it forms to auto-tag unseen faces in a new collection.

Image-based location estimation methods typically recognize every photo independently, and their resulting reliance on strong visual feature matches makes them most suited for distinctive landmark scenes. We observe that when touring a city, people tend to follow common travel patterns - for example, a stroll down Wall Street might be followed by a ferry ride, then a visit to the Statue of Liberty. We propose an approach that learns these trends directly from online image data, and then leverages them within a Hidden Markov Model to robustly estimate locations for novel sequences of tourist photos. We further devise a set-to-set matching-based likelihood that treats each “burst” of photos from the same camera as a single observation, thereby better accommodating images that may not contain particularly distinctive scenes. Our experiments with two large datasets of major tourist cities clearly demonstrate the approach's advantages over methods that recognize each photo individually, as well as a simpler HMM baseline that lacks the proposed burst-based observation model.

Visual attributes expose human-defined semantics to object recognition models, but existing work largely restricts their influence to mid-level cues during classifier training. Rather than treat attributes as intermediate features, we consider how learning visual properties in concert with object categories can regularize the models for both. Given a low-level visual feature space together with attribute-and object-labeled image data, we learn a shared lower-dimensional representation by optimizing a joint loss function that favors common sparsity patterns across both types of prediction tasks. We adopt a recent kernelized formulation of convex multi-task feature learning, in which one alternates between learning the common features and learning task-specific classifier parameters on top of those features. In this way, our approach discovers any structure among the image descriptors that is relevant to both tasks, and allows the top-down semantics to restrict the hypothesis space of the ultimate object classifiers. We validate the approach on datasets of animals and outdoor scenes, and show significant improvements over traditional multi-class object classifiers and direct attribute prediction models.

We propose a dense local region detector to extract features suitable for image matching and object recognition tasks. Whereas traditional local interest operators rely on repeatable structures that often cross object boundaries (e.g., corners, scale-space blobs), our sampling strategy is driven by segmentation, and thus preserves object boundaries and shape. At the same time, whereas existing region-based representations are sensitive to segmentation parameters and object deformations, our novel approach to robustly sample dense sites and determine their connectivity offers better repeatability. In extensive experiments, we find that the proposed region detector provides significantly better repeatability and localization accuracy for object matching compared to an array of existing detectors. In addition, we show our regions lead to excellent results on two benchmark tasks that require good feature matching: weakly supervised foreground discovery, and nearest neighbor-based object recognition.

Objects vary in their visual complexity, yet existing discovery methods perform “batch” clustering, paying equal attention to all instances simultaneously — regardless of the strength of their appearance or context cues. We propose a self-paced approach that instead focuses on the easiest instances first, and progressively expands its repertoire to include more complex objects. Easier regions are defined as those with both high likelihood of generic objectness and high familiarity of surrounding objects. At each cycle of the discovery process, we re-estimate the easiness of each subwindow in the pool of unlabeled images, and then retrieve a single prominent cluster from among the easiest instances. Critically, as the system gradually accumulates models, each new (more difficult) discovery benefits from the context provided by earlier discoveries. Our experiments demonstrate the clear advantages of self-paced discovery relative to conventional batch approaches, including both more accurate summarization as well as stronger predictive models for novel data.

Human-nameable visual attributes offer many advantages when used as mid-level features for object recognition, but existing techniques to gather relevant attributes can be inefficient (costing substantial effort or expertise) and/or insufficient (descriptive properties need not be discriminative). We introduce an approach to define a vocabulary of attributes that is both human understandable and discriminative. The system takes object/scene-labeled images as input, and returns as output a set of attributes elicited from human annotators that distinguish the categories of interest. To ensure a compact vocabulary and efficient use of annotators' effort, we 1) show how to actively augment the vocabulary such that new attributes resolve inter-class confusions, and 2) propose a novel “nameability” manifold that prioritizes candidate attributes by their likelihood of being associated with a nameable property. We demonstrate the approach with multiple datasets, and show its clear advantages over baselines that lack a nameability model or rely on a list of expert-provided attributes.

We propose a branch-and-cut strategy for efficient region-based object detection. Given an oversegmented image, our method determines the subset of spatially contiguous regions whose collective features will maximize a classifier's score. We formulate the objective as an instance of the prize-collecting Steiner tree problem, and show that for a family of additive classifiers this enables fast search for the optimal object region via a branch-and-cut algorithm. Unlike existing branch-and-bounddetection methods designed for bounding boxes, our approach allows scoring of irregular shapes - which is especially critical for objects that do not conform to a rectangular window. We provide results on three challenging object detection datasets, and demonstrate the advantage of rapidly seeking best-scoring regions rather than subwindow rectangles.

Active learning and crowdsourcing are promising ways to efficiently build up training sets for object recognition, but thus far techniques are tested in artificially controlled settings. Typically the vision researcher has already determined the dataset's scope, the labels “actively” obtained are in fact already known, and/or the crowd-sourced collection process is iteratively fine-tuned. We present an approach for live learning of object detectors, in which the system autonomously refines its models by actively requesting crowd-sourced annotations on images crawled from the Web. To address the technical issues such a large-scale system entails, we introduce a novel part-based detector amenable to linear classifiers, and show how to identify its most uncertain instances in sub-linear time with a hashing-based solution. We demonstrate the approach with experiments of unprecedented scale and autonomy, and show it successfully improves the state-of-the-art for the most challenging objects in the PASCAL benchmark. In addition, we show our detector competes well with popular nonlinear classifiers that are much more expensive to train.

Traditional supervised visual learning simply asks annotators “what” label an image should have. We propose an approach for image classification problems requiring subjective judgment that also asks “why”, and uses that information to enrich the learned model. We develop two forms of visual annotator rationales: in the first, the annotator highlights the spatial region of interest he found most influential to the label selected, and in the second, he comments on the visual attributes that were most important. For either case, we show how to map the response to synthetic contrast examples, and then exploit an existing large-margin learning technique to refine the decision boundary accordingly. Results on multiple scene categorization and human attractiveness tasks show the promise of our approach, which can more accurately learn complex categories with the explanations behind the label choices.

We present an active learning approach to choose image annotation requests among both object category labels and the objects' attribute labels. The goal is to solicit those labels that will best use human effort when training a multi-class object recognition model. In contrast to previous work in active visual category learning, our approach directly exploits the dependencies between human-nameable visual attributes and the objects they describe, shifting its requests in either label space accordingly. We adopt a discriminative latent model that captures object-attribute and attribute-attribute relationships, and then define a suitable entropy reduction selection criterion to predict the influence a new label might have throughout those connections. On three challenging datasets, we demonstrate that the method can more successfully accelerate object learning relative to both passive learning and traditional active learning approaches.

We present an approach to discover and segment foreground object(s) in video. Given an unannotated video sequence, the method first identifies object-like regions in any frame according to both static and dynamic cues. We then compute a series of binary partitions among those candidate “key-segments” to discover hypothesis groups with persistent appearance and motion. Finally, using each ranked hypothesis in turn, we estimate a pixel-level object labeling across all frames, where (a) the foreground likelihood depends on both the hypothesis's appearance as well as a novel localization prior based on partial shape matching, and (b) the background likelihood depends on cues pulled from the key-segments' (possibly diverse) surroundings observed across the sequence. Compared to existing methods, our approach automatically focuses on the persistent foreground regions of interest while resisting oversegmentation. We apply our method to challenging benchmark videos, and show competitive or better results than the state-of-the-art.

Human-nameable visual “attributes” can benefit various recognition tasks. However, existing techniques restrict these properties to categorical labels (for example, a person is `smiling' or not, a scene is `dry' or not), and thus fail to capture more general semantic relationships. We propose to model relative attributes. Given training data stating how object/scene categories relate according to different attributes, we learn a ranking function per attribute. The learned ranking functions predict the relative strength of each property in novel images. We then build a generative model over the joint space of attribute ranking outputs, and propose a novel form of zero-shot learning in which the supervisor relates the unseen object category to previously seen objects via attributes (for example, `bears are furrier than giraffes'). We further show how the proposed relative attributes enable richer textual descriptions for new images, which in practice are more precise for human interpretation. We demonstrate the approach on datasets of faces and natural scenes, and show its clear advantages over traditional binary attribute prediction for these new tasks.

We introduce an approach to learn discriminative visual representations while exploiting external semantic knowledge about object category relationships. Given a hierarchical taxonomy that captures semantic similarity between the objects, we learn a corresponding tree of metrics (ToM). In this tree, we have one metric for each non-leaf node of the object hierarchy, and each metric is responsible for discriminating among its immediate subcategory children. Specifically, a Mahalanobis metric learned for a given node must satisfy the appropriate (dis)similarity constraints generated only among its subtree members' training instances. To further exploit the semantics, we introduce a novel regularizer coupling the metrics that prefers a sparse disjoint set of features to be selected for each metric relative to its ancestor supercategory nodes' metrics. Intuitively, this reflects that visual cues most useful to distinguish the generic classes (e.g., feline vs. canine) should be different than those cues most useful to distinguish their component fine-grained classes (e.g., Persian cat vs. Siamese cat). We validate our approach with multiple image datasets using the WordNet taxonomy, show its advantages over alternative metric learning approaches, and analyze the meaning of attribute features selected by our algorithm.

We introduce a method for image retrieval that leverages the implicit information about object importance conveyed by the list of keyword tags a person supplies for an image.We propose an unsupervised learning procedure based on Kernel Canonical Correlation Analysis that discovers the relationship between how humans tag images (e.g., the order in which words are mentioned) and the relative importance of objects and their layout in the scene.Using this discovered connection, we show how to boost accuracy for novel queries, such that the search results may more closely match the user's mental image of the scene being sought.We evaluate our approach on two datasets, and show clear improvements over both an approach relying on image features alone, as well as a baseline that uses words and image features, but ignores the implied importance cues.

Current uses of tagged images typically exploit only the most explicit information: the link between the nouns named and the objects present somewhere in the image. We propose to leverage “unspoken” cues that rest within an ordered list of image tags so as to improve object localization. We define three novel implicit features from an image's tags - the relative prominence of each object as signified by its order of mention, the scale constraints implied by unnamed objects, and the loose spatial links hinted by the proximity of names on the list. By learning a conditional density over the localization parameters (position and scale) given these cues, we show how to improve both accuracy and efficiency when detecting the tagged objects. We validate our approach with 25 object categories from the PASCAL VOC and LabelMe datasets, and demonstrate its effectiveness relative to both traditional sliding windows as well as a visual context baseline.

We present a feature matching algorithm that leverages bottom-up segmentation. Unlike conventional image-to-image or region-to-region matching algorithms, our method finds corresponding points in an “asymmetric” manner, matching features within each region of a segmented image to a second unsegmented image. We develop a dynamic programming solution to efficiently identify corresponding points for each region, so as to maximize both geometric consistency and appearance similarity. The final matching score between two images is determined by the union of corresponding points obtained from each region-to-image match. Our encoding for the geometric constraints makes the algorithm flexible when matching objects exhibiting non-rigid deformations or intra-class appearance variation. We demonstrate our image matching approach applied to object category recognition, and show on the Caltech-256 and 101 datasets that it outperforms existing image matching measures by 10~20% in nearest-neighbor recognition tests.

Recent work shows how to use local spatio-temporal features to learn models of realistic human actions from video. However, existing methods typically rely on a predefined spatial binning of the local descriptors to impose spatial information beyond a pure “bag-of-words” model, and thus may fail to capture the most informative space-time relationships. We propose to learn the shapes of space-time feature neighborhoods that are most discriminative for a given action category. Given a set of training videos, our method first extracts local motion and appearance features, quantizes them to a visual vocabulary, and then forms candidate neighborhoods consisting of the words associated with nearby points and their orientation with respect to the central interest point. Rather than dictate a particular scaling of the spatial and temporal dimensions to determine which points are near, we show how to learn the class-specific distance functions that form the most informative configurations. Descriptors for these variable-sized neighborhoods are then recursively mapped to higher-level vocabularies, producing a hierarchy of space-time configurations at successively broader scales. Our approach yields state-of-the-art performance on the UCF Sports and KTH datasets.

How can knowing about some categories help us to discover new ones in unlabeled images? Unsupervised visual category discovery is useful to mine for recurring objects without human supervision, but existing methods assume no prior information and thus tend to perform poorly for cluttered scenes with multiple objects. We propose to leverage knowledge about previously learned categories to enable more accurate discovery. We introduce a novel object-graph descriptor to encode the layout of object-level co-occurrence patterns relative to an unfamiliar region, and show that by using it to model the interaction between an image's known and unknown objects we can better detect new visual categories. Rather than mine for all categories from scratch, our method identifies new objects while drawing on useful cues from familiar ones. We evaluate our approach on benchmark datasets and demonstrate clear improvements in discovery over conventional purely appearance-based baselines.

We present a method to segment a collection of unlabeled images while exploiting automatically discovered appearance patterns shared between them. Given an unlabeled pool of multi-object images, we first detect any visual clusters present among their sub-regions, where inter-region similarity is measured according to both appearance and contextual layout. Then, using each initial segment as a seed, we solve a graph cuts problem to refine its boundary — enforcing preferences to include nearby regions that agree with an ensemble of representative regions discovered for that cluster, and exclude those regions that resemble familiar objects. Through extensive experiments, we show that the segmentations computed jointly on the collection agree more closely with true object boundaries, when compared to either a bottom-up baseline or a graph cuts foreground segmentation that can only access cues from a single image.

Active learning methods aim to select the most informative unlabeled instances to label first, and can help to focus image or video annotations on the examples that will most improve a recognition system. However, most existing methods only make myopic queries for a single label at a time, retraining at each iteration. We consider the problem where at each iteration the active learner must select a set of examples meeting a given budget of supervision, where the budget is determined by the funds (or time) available to spend on annotation. We formulate the budgeted selection task as a continuous optimization problem where we determine which subset of possible queries should maximize the improvement to the classifier's objective, without overspending the budget. To ensure far-sighted batch requests, we show how to incorporate the predicted change in the model that the candidate examples will induce. We demonstrate the proposed algorithm on three datasets for object recognition, activity recognition, and content-based retrieval, and we show its clear practical advantages over random, myopic, and batch selection baselines.

We consider the problem of retrieving the database points nearest to a given {\em hyperplane} query without exhaustively scanning the database. We propose two hashing-based solutions. Our first approach maps the data to two-bit binary keys that are locality-sensitive for the angle between the hyperplane normal and a database point. Our second approach embeds the data into a vector space where the Euclidean norm reflects the desired distance between the original points and hyperplane query. Both use hashing to retrieve near points in sub-linear time. Our first method's preprocessing stage is more efficient, while the second has stronger accuracy guarantees. We apply both to pool-based active learning: taking the current hyperplane classifier as a query, our algorithm identifies those points (approximately) satisfying the well-known minimal distance-to-hyperplane selection criterion. We empirically demonstrate our methods' tradeoffs, and show that they make it practical to perform active selection with millions of unlabeled points.

Discriminative methods for visual object category recognition are typically non-probabilistic, predicting class labels but not directly providing an estimate of uncertainty. Gaussian Processes (GPs) provide a framework for deriving regression techniques with explicit uncertainty models; we show here how Gaussian Processes with covariance functions defined based on a Pyramid Match Kernel (PMK) can be used for probabilistic object category recognition. Our probabilistic formulation provides a principled way to learn hyperparameters, which we utilize to learn an optimal combination of multiple covariance functions. It also offers confidence estimates at test points, and naturally allows for an active learning paradigm in which points are optimally selected for interactive labeling. We show that with an appropriate combination of kernels a significant boost in classification performance is possible. Further, our experiments indicate the utility of active learning with probabilistic predictive models, especially when the amount of training data labels that may be sought for a category is ultimately very small.

We propose a space-time Markov random field (MRF) model to detect abnormal activities in video. The nodes in the MRF graph correspond to a grid of local regions in the video frames, and neighboring nodes in both space and time are associated with links. To learn normal patterns of activity at each local node, we capture the distribution of its typical optical flow with a mixture of probabilistic principal component analyzers. For any new optical flow patterns detected in incoming video clips, we use the learned model and MRF graph to compute a maximum a posteriori estimate of the degree of normality at each local node. Further, we show how to incrementally update the current model parameters as new video observations stream in, so that the model can efficiently adapt to visual context changes over a long period of time. Experimental results on surveillance videos show that our space-time MRF model robustly detects abnormal activities both in a local and global sense: not only does it accurately localize the atomic abnormal activities in a crowded video, but at the same time it captures the global-level abnormalities caused by irregular interactions between local activities.

Can we discover common object shapes within unlabeled multi-category collections of images? While often a critical cue at the category-level, contour matches can be difficult to isolate reliably from edge clutter—even within labeled images from a known class, let alone unlabeled examples. We propose a shape discovery method in which local appearance (patch) matches serve to anchor the surrounding edge fragments, yielding a more reliable affinity function for images that accounts for both shape and appearance. Spectral clustering from the initial affinities provides candidate object clusters. Then, we compute the within-cluster match patterns to discern foreground edges from clutter, attributing higher weight to edges more likely to belong to a common object. In addition to discovering the object contours in each image, we show how to summarize what is found with prototypical shapes. Our results on benchmark datasets demonstrate the approach can successfully discover shapes from unlabeled images.

Active learning strategies can be useful when manual labeling effort is scarce, as they select the most informative examples to be annotated first. However, for visual category learning, the active selection problem is particularly complex: a single image will typically contain multiple object labels, and an annotator could provide multiple types of annotation (e.g., class labels, bounding boxes, segmentations), any of which would incur a variable amount of manual effort. We present an active learning framework that predicts the tradeoff between the effort and information gain associated with a candidate image annotation, thereby ranking unlabeled and partially labeled images according to their expected “net worth” to an object recognition system. We develop a multi-label multiple-instance approach that accommodates multi-object images and a mixture of strong and weak labels. Since the annotation cost can vary depending on an image's complexity, we show how to improve the active selection by directly predicting the time required to segment an unlabeled image. Given a small initial pool of labeled data, the proposed method actively improves the category models with minimal manual intervention.

Fast retrieval methods are critical for large-scale and data-driven vision applications. Recent work has explored ways to embed high-dimensional features or complex distance functions into a low-dimensional Hamming space where items can be efficiently searched. However, existing methods do not apply for high-dimensional kernelized data when the underlying feature embedding for the kernel is unknown. We show how to generalize locality-sensitive hashing to accommodate arbitrary kernel functions, making it possible to preserve the algorithm's sub-linear time similarity search guarantees for a wide class of useful similarity functions. Since a number of successful image-based kernels have unknown or incomputable embeddings, this is especially valuable for image retrieval tasks. We validate our technique on several large-scale datasets, and show that it enables accurate and fast performance for example-based object classification, feature matching, and content-based retrieval.

We introduce a method that enables scalable similarity search for learned metrics. Given pairwise similarity and dissimilarity constraints between some examples, we learn a Mahalanobis distance function that captures the examples' underlying relationships well. To allow sublinear time similarity search under the learned metric, we show how to encode the learned metric parameterization into randomized locality-sensitive hash functions. We further formulate an indirect solution that enables metric learning and hashing for vector spaces whose high dimensionality makes it infeasible to learn an explicit transformation over the feature dimensions. We demonstrate the approach applied to a variety of image data sets, as well as a systems data set. The learned metrics improve accuracy relative to commonly used metric baselines, while our hashing construction enables efficient indexing with learned distances and very large databases.

We present a method to automatically discover meaningful features in unlabeled image collections. Each image is decomposed into semi-local features that describe neighborhood appearance and geometry. The goal is to determine for each image which of these parts are most relevant, given the image content in the remainder of the collection. Our method first computes an initial image-level grouping based on feature correspondences, and then iteratively refines cluster assignments based on the evolving intra-cluster pattern of local matches. As a result, the significance attributed to each feature influences an image’s cluster membership, while related images in a cluster affect the estimated significance of their features. We show that this mutual reinforcement of object-level and feature-level similarity improves unsupervised image clustering, and apply the technique to automatically discover categories and foreground regions in images from benchmark datasets. 1

We introduce a method that enables scalable image search for learned metrics. Given pairwise similarity and dissimilarity constraints between some images, we learn a Mahalanobis distance function that captures the imagespsila underlying relationships well. To allow sub-linear time similarity search under the learned metric, we show how to encode the learned metric parameterization into randomized locality-sensitive hash functions. We further formulate an indirect solution that enables metric learning and hashing for vector spaces whose high dimensionality make it infeasible to learn an explicit weighting over the feature dimensions. We demonstrate the approach applied to a variety of image datasets. Our learned metrics improve accuracy relative to commonly-used metric baselines, while our hashing construction enables efficient indexing with learned distances and very large databases.

Conventional supervised methods for image categorization rely on manually annotated (labeled) examples to learn good object models, which means their generality and scalability depends heavily on the amount of human effort available to help train them. We propose an unsupervised approach to construct discriminative models for categories specified simply by their names. We show that multiple-instance learning enables the recovery of robust category models from images returned by keyword-based search engines. By incorporating constraints that reflect the expected sparsity of true positive examples into a large-margin objective function, our approach remains accurate even when the available text annotations are imperfect and ambiguous. In addition, we show how to iteratively improve the learned classifier by automatically refining the representation of the ambiguously labeled examples. We demonstrate our method with benchmark datasets, and show that it performs well relative to both state-of-the-art unsupervised approaches and traditional fully supervised techniques.

Metric learning algorithms can provide useful distance functions for a variety of domains, and recent work has shown good accuracy for problems where the learner can access all distance constraints at once. However, in many real applications, constraints are only available incrementally, thus necessitating methods that can perform online updates to the learned metric. Existing online algorithms offer bounds on worst-case performance, but typically do not perform well in practice as compared to their offline counterparts. We present a new online metric learning algorithm that updates a learned Mahalanobis metric based on LogDet regularization and gradient descent. We prove theoretical worst-case performance bounds, and empirically compare the proposed method against existing online metric learning algorithms. To further boost the practicality of our approach, we develop an online locality-sensitive hashing scheme which leads to efficient updates for approximate similarity search data structures. We demonstrate our algorithm on multiple datasets and show that it outperforms relevant baselines.

We introduce a framework for actively learning visual categories from a mixture of weakly and strongly labeled image examples. We propose to allow the category-learner to strategically choose what annotations it receives---based on both the expected reduction in uncertainty as well as the relative costs of obtaining each annotation. We construct a multiple-instance discriminative classifier based on the initial training data. Then all remaining unlabeled and weakly labeled examples are surveyed to actively determine which annotation ought to be requested next. After each request, the current classifier is incrementally updated. Unlike previous work, our approach accounts for the fact that the optimal use of manual annotation may call for a combination of labels at multiple levels of granularity (e.g., a full segmentation on some images and a present/absent flag on others). As a result, it is possible to learn more accurate category models with a lower total expenditure of manual annotation effort.

Matching local features across images is often useful when comparing or recognizing objects or scenes, and efficient techniques for obtaining image-to-image correspondences have been developed [4, 11, 16]. However, given a query image, searching a very large image database with such measures remains impractical. We introduce a sub-linear time randomized hashing algorithm for indexing sets of feature vectors under their partial correspondences. We develop an efficient embedding function for the normalized partial matching similarity between sets, and show how to exploit random hyperplane properties to construct hash functions that satisfy locality-sensitive constraints. The result is a bounded approximate similarity search algorithm that finds (1 + epsiv)-approximate nearest neighbor images in O(N1/1+epsiv) time for a database containing N images represented by (varying numbers of) local features. We demonstrate our approach applied to image retrieval for images represented by sets of local appearance features, and show that searching over correspondences is now scalable to large image databases.

Discriminative methods for visual object category recognition are typically non-probabilistic, predicting class labels but not directly providing an estimate of uncertainty. Gaussian Processes (GPs) are powerful regression techniques with explicit uncertainty models; we show here how Gaussian Processes with covariance functions defined based on a Pyramid Match Kernel (PMK) can be used for probabilistic object category recognition. The uncertainty model provided by GPs offers confidence estimates at test points, and naturally allows for an active learning paradigm in which points are optimally selected for interactive labeling. We derive a novel active category learning method based on our probabilistic regression model, and show that a significant boost in classification performance is possible, especially when the amount of training data for a category is ultimately very small.

We present a method to automatically learn object categories from unlabeled images. Each image is represented by an unordered set of local features, and all sets are embedded into a space where they cluster according to their partial-match feature correspondences. After efficiently computing the pairwise affinities between the input images in this space, a spectral clustering technique is used to recover the primary groupings among the images. We introduce an efficient means of refining these groupings according to intra-cluster statistics over the subsets of features selected by the partial matches between the images, and based on an optional, variable amount of user supervision. We compute the consistent subsets of feature correspondences within a grouping to infer category feature masks. The output of the algorithm is a partition of the data into a set of learned categories, and a set of classifiers trained from these ranked partitions that can recognize the categories in novel images.

Pyramid intersection is an efficient method for computing an approximate partial matching between two sets of feature vectors. We introduce a novel pyramid em- bedding based on a hierarchy of non-uniformly shaped bins that takes advantage of the underlying structure of the feature space and remains accurate even for sets with high-dimensional feature vectors. The matching similarity is computed in linear time and forms a Mercer kernel. Whereas previous matching approxima- tion algorithms suffer from distortion factors that increase linearly with the fea- ture dimension, we demonstrate that our approach can maintain constant accuracy even as the feature dimension increases. When used as a kernel in a discrimina- tive classifier, our approach achieves improved object recognition results over a state-of-the-art set kernel.

Sets of local features that are invariant to common image transformations are an effective representation to use when comparing images; current methods typically judge feature sets' similarity via a voting scheme (which ignores co-occurrence statistics) or by comparing histograms over a set of prototypes (which must be found by clustering). We present a method for efficiently comparing images based on their discrete distributions (bags) of distinctive local invariant features, without clustering descriptors. Similarity between images is measured with an approximation of the Earth Mover's Distance (EMD), which quickly computes minimal-cost correspondences between two bags of features. Each image's feature distribution is mapped into a normed space with a low-distortion embedding of EMD. Examples most similar to a novel query image are retrieved in time sublinear in the number of examples via approximate nearest neighbor search in the embedded space. We evaluate our method with scene, object, and texture recognition tasks.

Classic methods for Bayesian inference effectively constrain search to lie within regions of significant probability of the temporal prior. This is efficient with an accurate dynamics model, but otherwise is prone to ignore significant peaks in the true posterior. A more accurate posterior estimate can be obtained by explicitly finding modes of the likelihood function and combining them with a weak temporal prior. In our approach, modes are found using efficient example-based matching followed by local refinement to find peaks and estimate peak bandwidth. By reweighting these peaks according to the temporal prior we obtain an estimate of the full posterior model. We show comparative results on real and synthetic images in a high degree of freedom articulated tracking task.

Discriminative learning is challenging when examples are sets of features, and the sets vary in cardinality and lack any sort of meaningful ordering. Kernel-based classification methods can learn complex decision boundaries, but a kernel over unordered set inputs must somehow solve for correspondences epsivnerally a computationally expensive task that becomes impractical for large set sizes. We present a new fast kernel function which maps unordered feature sets to multi-resolution histograms and computes a weighted histogram intersection in this space. This "pyramid match" computation is linear in the number of features, and it implicitly finds correspondences based on the finest resolution histogram cell where a matched pair first appears. Since the kernel does not penalize the presence of extra features, it is robust to clutter. We show the kernel function is positive-definite, making it valid for use in learning algorithms whose optimal solutions are guaranteed only for Mercer kernels. We demonstrate our algorithm on object recognition tasks and show it to be accurate and dramatically faster than current approaches

We present a Bayesian approach to image-based visual hull reconstruction. The 3D (three-dimensional) shape of an object of a known class is represented by sets of silhouette views simultaneously observed from multiple cameras. We show how the use of a class-specific prior in a visual hull reconstruction can reduce the effect of segmentation errors from the silhouette extraction process. In our representation, 3D information is implicit in the joint observations of multiple contours from known viewpoints. We model the prior density using a probabilistic principal components analysis-based technique and estimate a maximum a posteriori reconstruction of multi-view contours. The proposed method is applied to a dataset of pedestrian images, and improvements in the approximate 3D models under various noise conditions are shown.

We present an image-based approach to infer 3D structure parameters using a probabilistic "shape+structure" model. The 3D shape of an object class is represented by sets of contours from silhouette views simultaneously observed from multiple calibrated cameras, while structural features of interest on the object are denoted by a number of 3D locations. A prior density over the multiview shape and corresponding structure is constructed with a mixture of probabilistic principal components analyzers. Given a novel set of contours, we infer the unknown structure parameters from the new shape's Bayesian reconstruction. Model matching and parameter inference are done entirely in the image domain and require no explicit 3D construction. Our shape model enables accurate estimation of structure despite segmentation errors or missing views in the input silhouettes, and it works even with only a single input view. Using a training set of thousands of pedestrian images generated from a synthetic model, we can accurately infer the 3D locations of 19 joints on the body based on observed silhouette contours from real images.

A method for a real-time vision system that automatically detects a user's eye blinks and accurately measures their durations is introduced. The system is intended to provide an alternate input modality to allow people with severe disabilities to access a computer. Voluntary long blinks trigger mouse clicks, while involuntary short blinks are ignored. The system enables communication using "blink patterns:" sequences of long and short blinks which are interpreted as semiotic messages. The location of the eyes is determined automatically through the motion of the user's initial blinks. Subsequently, the eye is tracked by correlation across time, and appearance changes are automatically analyzed in order to classify the eye as either open or closed at each frame. No manual initialization, special lighting, or prior face detection is required. The system has been tested with interactive games and a spelling program. Results demonstrate overall detection accuracy of 95.61% and an average rate of 28 frames per second.