We present a method for reconstructing a clear Neural Radiance Field (NeRF) even with fast camera motions. To address blur artifacts, we leverage both (blurry) RGB images and event camera data captured in a binocular configuration. Importantly, when reconstructing our clear NeRF, we consider the camera modeling imperfections that arise from the simple pinhole camera model as learned embeddings for each camera measurement, and further learn a mapper that connects event camera measurements with RGB data. As no previous dataset exists for our binocular setting, we introduce an event camera dataset with captures from a 3D-printed stereo configuration between RGB and event cameras. Empirically, we evaluate our introduced dataset and EVIMOv2 and show that our method leads to improved reconstructions. Our code and dataset are available at https://github.com/ubc-vision/LSENeRF.

White balance (WB) correction in scenes with multiple illuminants remains a persistent challenge in computer vision. Recent methods explored fusion-based approaches, where a neural network linearly blends multiple sRGB versions of an input image, each processed with predefined WB presets. However, we demonstrate that these methods are suboptimal for common multi-illuminant scenarios. Additionally, existing fusion-based methods rely on sRGB WB datasets lacking dedicated multi-illuminant images, limiting both training and evaluation. To address these challenges, we introduce two key contributions. First, we propose an efficient transformer-based model that effectively captures spatial dependencies across sRGB WB presets, substantially improving upon linear fusion techniques. Second, we introduce a large-scale multi-illuminant dataset comprising over 16,000 sRGB images rendered with five different WB settings, along with WB-corrected images. Our method achieves up to 100\% improvement over existing techniques on our new multi-illuminant image fusion dataset.

We present Universal Sparse Autoencoders (USAEs), a framework for uncovering and aligning interpretable concepts spanning multiple pretrained deep neural networks. Unlike existing concept-based interpretability methods, which focus on a single model, USAEs jointly learn a universal concept space that can reconstruct and interpret the internal activations of multiple models at once. Our core insight is to train a single, overcomplete sparse autoencoder (SAE) that ingests activations from any model and decodes them to approximate the activations of any other model under consideration. By optimizing a shared objective, the learned dictionary captures common factors of variation—concepts—across different tasks, architectures, and datasets. We show that USAEs discover semantically coherent and important universal concepts across vision models; ranging from low-level features (e.g., colors and textures) to higher-level structures (e.g., parts and objects). Overall, USAEs provide a powerful new method for interpretable cross-model analysis and offers novel applications—such as coordinated activation maximization—that open avenues for deeper insights in multi-model AI systems.

There is limited understanding of the information captured by deep spatiotemporal models in their intermediate representations. For example, while evidence suggests that action recognition algorithms are heavily influenced by visual appearance in single frames, no quantitative methodology exists for evaluating such static bias in the latent representation compared to bias toward dynamics. We tackle this challenge by proposing an approach for quantifying the static and dynamic biases of any spatiotemporal model, and apply our approach to three tasks, action recognition, automatic video object segmentation (AVOS) and video instance segmentation (VIS). Our key findings are: (i) Most examined models are biased toward static information. (ii) Some datasets that are assumed to be biased toward dynamics are actually biased toward static information. (iii) Individual channels in an architecture can be biased toward static, dynamic or jointly encode a combination static and dynamic information. (iv) Most models converge to their culminating biases in the first half of training. We then explore how these biases affect performance on dynamically biased datasets. For action recognition, we propose StaticDropout, a semantically guided dropout that debiases a model from static information toward dynamics. For AVOS, we design a better combination of fusion and cross connection layers compared with previous architectures.

This paper studies the problem of concept-based inter-pretability of transformer representations for videos. Con-cretely, we seek to explain the decision-making process of video transformers based on high-level, spatiotemporal concepts that are automatically discovered. Prior research on concept-based interpretability has concentrated solely on image-level tasks. Comparatively, video models deal with the added temporal dimension, increasing complexity and posing challenges in identifying dynamic concepts over time. In this work, we systematically address these challenges by introducing the first Video Transformer Con-cept Discovery (VTCD) algorithm. To this end, we propose an efficient approach for unsupervised identification of units of video transformer representations - concepts, and ranking their importance to the output of a model. The resulting concepts are highly interpretable, revealing spatiotemporal reasoning mechanisms and object-centric representations in unstructured video models. Performing this analysis jointly over a diverse set of supervised and self-supervised representations, we discover that some of these mechanism are universal in video transformers. Finally, we show that VTCD can be usedfor fine-grained action recognition and video object segmentation.

Understanding what deep network models capture in their learned representations is a fundamental challenge in computer vision. We present a new methodology to understanding such vision models, the Visual Concept Con-nectome (VCC), which discovers human interpretable concepts and their interlayer connections in a fully unsuper-vised manner. Our approach simultaneously reveals fine-grained concepts at a layer, connection weightings across all layers and is amendable to global analysis of network structure (e.g. branching pattern of hierarchical concept assemblies). Previous work yielded ways to extract inter-pretable concepts from single layers and examine their im-pact on classification, but did not afford multilayer concept analysis across an entire network architecture. Quantitative and qualitative empirical results show the effectiveness of VCCs in the domain of image classification. Also, we lever-age VCCs for the application of failure mode debugging to reveal where mistakes arise in deep networks.

Instructional videos are an important resource to learn procedural tasks from human demonstrations. However, the instruction steps in such videos are typically short and sparse, with most of the video being irrelevant to the procedure. This motivates the need to temporally localize the instruction steps in such videos, i.e. the task called key-step localization. Traditional methods for key-step localization require video-level human annotations and thus do not scale to large datasets. In this work, we tackle the problem with no human supervision and introduce StepFormer, a self-supervised model that discovers and localizes instruction steps in a video. StepFormer is a transformer decoder that attends to the video with learnable queries, and produces a sequence of slots capturing the key-steps in the video. We train our system on a large dataset of instructional videos, using their automatically-generated subtitles as the only source of supervision. In particular, we supervise our system with a sequence of text narrations using an order-aware loss function that filters out irrelevant phrases. We show that our model outperforms all previous unsupervised and weakly-supervised approaches on step detection and localization by a large margin on three challenging benchmarks. Moreover, our model demonstrates an emergent property to solve zero-shot multi-step localization and outperforms all relevant baselines at this task.

Neural Radiance Fields (NeRFs) have emerged as a popular approach for novel view synthesis. While NeRFs are quickly being adapted for a wider set of applications, intuitively editing NeRF scenes is still an open challenge. One important editing task is the removal of unwanted objects from a 3D scene, such that the replaced region is visually plausible and consistent with its context. We refer to this task as 3D inpainting. In 3D, solutions must be both consistent across multiple views and geometrically valid. In this paper, we propose a novel 3D inpainting method that addresses these challenges. Given a small set of posed images and sparse annotations in a single input image, our framework first rapidly obtains a 3D segmentation mask for a target object. Using the mask, a perceptual optimization-based approach is then introduced that leverages learned 2D image inpainters, distilling their information into 3D space, while ensuring view consistency. We also address the lack of a diverse benchmark for evaluating 3D scene inpainting methods by introducing a dataset comprised of challenging real-world scenes. In particular, our dataset contains views of the same scene with and without a target object, enabling more principled benchmarking of the 3D inpainting task. We first demonstrate the superiority of our approach on multiview segmentation, comparing to NeRF-based methods and 2D segmentation approaches. We then evaluate on the task of 3D inpainting, establishing state-of-the-art performance against other NeRF manipulation algorithms, as well as a strong 2D image inpainter baseline.

We study the problem of future step anticipation in procedural videos. Given a video of an ongoing procedural activity, we predict a plausible next procedure step described in rich natural language. While most previous work focuses on the problem of data scarcity in procedural video datasets, another core challenge of future anticipation is how to account for multiple plausible future realizations in natural settings. This problem has been largely overlooked in previous work. To address this challenge, we frame future step prediction as modelling the distribution of all possible candidates for the next step. Specifically, we design a generative model that takes a series of video clips as input, and generates multiple plausible and diverse candidates (in natural language) for the next step. Following previous work, we side-step the video annotation scarcity by pretraining our model on a large text-based corpus of procedural activities, and then transfer the model to the video domain. Our experiments, both in textual and video domains, show that our model captures diversity in the next step prediction and generates multiple plausible future predictions. Moreover, our model establishes new state-of-the-art results on YouCookII, where it outperforms existing baselines on the next step anticipation. Finally, we also show that our model can successfully transfer from text to the video domain zero-shot, i.e., without fine-tuning or adaptation, and produces good-quality future step predictions from video.

The popularity of Neural Radiance Fields (NeRFs) for view synthesis has led to a desire for NeRF editing tools. Here, we focus on inpainting regions in a view-consistent and controllable manner. In addition to the typical NeRF inputs and masks delineating the unwanted region in each view, we require only a single inpainted view of the scene, i.e., a reference view. We use monocular depth estimators to back-project the inpainted view to the correct 3D positions. Then, via a novel rendering technique, a bilateral solver can construct view-dependent effects in non-reference views, making the inpainted region appear consistent from any view. For non-reference disoccluded regions, which cannot be supervised by the single reference view, we devise a method based on image inpainters to guide both the geometry and appearance. Our approach shows superior performance to NeRF inpainting baselines, with the additional advantage that a user can control the output via a single inpainted image. Please visit our project page.

Novel view synthesis from a single input image is a challenging task, where the goal is to generate a new view of a scene from a desired camera pose that may be separated by a large motion. The highly uncertain nature of this synthesis task due to unobserved elements within the scene (i.e. occlusion) and outside the field-of-view makes the use of generative models appealing to capture the variety of possible outputs. In this paper, we propose a novel generative model capable of producing a sequence of photorealistic images consistent with a specified camera trajectory, and a single starting image. Our approach is centred on an autoregressive conditional diffusion-based model capable of interpolating visible scene elements, and extrapolating unobserved regions in a view, in a geometrically consistent manner. Conditioning is limited to an image capturing a single camera view and the (relative) pose of the new camera view. To measure the consistency over a sequence of generated views, we introduce a new metric, the thresholded symmetric epipolar distance (TSED), to measure the number of consistent frame pairs in a sequence. While previous methods have been shown to produce high quality images and consistent semantics across pairs of views, we show empirically with our metric that they are often inconsistent with the desired camera poses. In contrast, we demonstrate that our method produces both photorealistic and view-consistent imagery. Additional material is available on our project page: https://yorkucvil.github.io/Photoconsistent-NVS/.

In this paper, we study the problem of procedure planning in instructional videos. Here, an agent must produce a plausible sequence of actions that can transform the environment from a given start to a desired goal state. When learning procedure planning from instructional videos, most recent work leverages intermediate visual observations as supervision, which requires expensive annotation efforts to localize precisely all the instructional steps in training videos. In contrast, we remove the need for expensive temporal video annotations and propose a weakly supervised approach by learning from natural language instructions. Our model is based on a transformer equipped with a memory module, which maps the start and goal observations to a sequence of plausible actions. Furthermore, we augment our model with a probabilistic generative module to capture the uncertainty inherent to procedure planning, an aspect largely overlooked by previous work. We evaluate our model on three datasets and show our weakly-supervised approach outperforms previous fully supervised state-of-the-art models on multiple metrics.

Deep spatiotemporal models are used in a variety of computer vision tasks, such as action recognition and video object segmentation. Currently, there is a limited understanding of what information is captured by these models in their intermediate representations. For example, while it has been observed that action recognition algorithms are heavily influenced by visual appearance in single static frames, there is no quantitative methodology for evaluating such static bias in the latent representation compared to bias toward dynamic information (e.g. motion). We tackle this challenge by proposing a novel approach for quantifying the static and dynamic biases of any spatiotemporal model. To show the efficacy of our approach, we analyse two widely studied tasks, action recognition and video object segmentation. Our key findings are threefold: (i) Most examined spatiotemporal models are biased toward static information; although, certain two-stream architectures with cross-connections show a better balance between the static and dynamic information captured. (ii) Some datasets that are commonly assumed to be biased toward dynamics are actually biased toward static information. (iii) Individual units (channels) in an architecture can be biased toward static, dynamic or a combination of the two.11Project page and code

Capturing photographs with wrong exposures remains a major source of errors in camera-based imaging. Exposure problems are categorized as either: (i) overexposed, where the camera exposure was too long, resulting in bright and washed-out image regions, or (ii) underexposed, where the exposure was too short, resulting in dark regions. Both under- and overexposure greatly reduce the contrast and visual appeal of an image. Prior work mainly focuses on underexposed images or general image enhancement. In contrast, our proposed method targets both over- and underexposure errors in photographs. We formulate the exposure correction problem as two main sub-problems: (i) color enhancement and (ii) detail enhancement. Accordingly, we propose a coarse-to-fine deep neural network (DNN) model, trainable in an end-to-end manner, that addresses each sub-problem separately. A key aspect of our solution is a new dataset of over 24,000 images exhibiting the broadest range of exposure values to date with a corresponding properly exposed image. Our method achieves results on par with existing state-of-the-art methods on underexposed images and yields significant improvements for images suffering from overexposure errors.

Video understanding calls for a model to learn the characteristic interplay between static scene content and its dynamics: Given an image, the model must be able to predict a future progression of the portrayed scene and, conversely, a video should be explained in terms of its static image content and all the remaining characteristics not present in the initial frame. This naturally suggests a bijective mapping between the video domain and the static content as well as residual information. In contrast to common stochastic image-to-video synthesis, such a model does not merely generate arbitrary videos progressing the initial image. Given this image, it rather provides a one-to-one mapping between the residual vectors and the video with stochastic outcomes when sampling. The approach is naturally implemented using a conditional invertible neural network (cINN) that can explain videos by independently modelling static and other video characteristics, thus laying the basis for controlled video synthesis. Experiments on diverse video datasets demonstrate the effectiveness of our approach in terms of both the quality and diversity of the synthesized results. Our project page is available at https://bit.ly/3dg90fV.

We introduce a weakly supervised method for representation learning based on aligning temporal sequences (e.g., videos) of the same process (e.g., human action). The main idea is to use the global temporal ordering of latent correspondences across sequence pairs as a supervisory signal. In particular, we propose a loss based on scoring the optimal sequence alignment to train an embedding network. Our loss is based on a novel probabilistic path finding view of dynamic time warping (DTW) that contains the following three key features: (i) the local path routing decisions are contrastive and differentiable, (ii) pairwise distances are cast as probabilities that are contrastive as well, and (iii) our formulation naturally admits a global cycle-consistency loss that verifies correspondences. For evaluation, we consider the tasks of fine-grained action classification, few shot learning, and video synchronization. We report significant performance increases over previous methods. In addition, we report two applications of our temporal alignment framework, namely 3D pose reconstruction and fine-grained audio/visual retrieval.

In this paper, we challenge the common assumption that collapsing the spatial dimensions of a 3D (spatial-channel) tensor in a convolutional neural network (CNN) into a vector via global pooling removes all spatial information. Specifically, we demonstrate that positional information is encoded based on the ordering of the channel dimensions, while semantic information is largely not. Following this demonstration, we show the real world impact of these findings by applying them to two applications. First, we propose a simple yet effective data augmentation strategy and loss function which improves the translation invariance of a CNN’s output. Second, we propose a method to efficiently determine which channels in the latent representation are responsible for (i) encoding overall position information or (ii) region-specific positions. We first show that semantic segmentation has a significant reliance on the overall position channels to make predictions. We then show for the first time that it is possible to perform a ‘region-specific’ attack, and degrade a network’s performance in a particular part of the input. We believe our findings and demonstrated applications will benefit research areas concerned with understanding the characteristics of CNNs. Code is available at: https://github.com/islamamirul/PermuteNet.

Current video retrieval systems on mobile devices cannot process complex natural language queries, especially if they contain personalized concepts, such as proper names. To address these shortcomings, we propose an efficient and privacy-preserving video retrieval system that works well with personalized queries containing proper names, without re-training using personalized labelled data from users. Our system first computes an initial ranking of a video collection by using a generic attention-based video-text matching model (i.e., a model designed for non-personalized queries), and then uses a face detector to conduct personalized adjustments to these initial rankings. These adjustments are done by reasoning over the face information from the detector and the attention information provided by the generic model. We show that our system significantly outperforms existing keyword-based retrieval systems, and achieves comparable performance to the generic matching model fine-tuned on plenty of labelled data. Our results suggest that the proposed system can effectively capture both semantic context and personalized information in queries.

In this work, we consider the problem of sequence-to-sequence alignment for signals containing outliers. Assuming the absence of outliers, the standard Dynamic Time Warping (DTW) algorithm efficiently computes the optimal alignment between two (generally) variable-length sequences. While DTW is robust to temporal shifts and dilations of the signal, it fails to align sequences in a meaningful way in the presence of outliers that can be arbitrarily interspersed in the sequences. To address this problem, we introduce Drop-DTW, a novel algorithm that aligns the common signal between the sequences while automatically dropping the outlier elements from the matching. The entire procedure is implemented as a single dynamic program that is efficient and fully differentiable. In our experiments, we show that Drop-DTW is a robust similarity measure for sequence retrieval and demonstrate its effectiveness as a training loss on diverse applications. With Drop-DTW, we address temporal step localization on instructional videos, representation learning from noisy videos, and cross-modal representation learning for audio-visual retrieval and localization. In all applications, we take a weakly- or unsupervised approach and demonstrate state-of-the-art results under these settings.

For humans, visual understanding is inherently generative: given a 3D shape, we can postulate how it would look in the world; given a 2D image, we can infer the 3D structure that likely gave rise to it. We can thus translate between the 2D visual and 3D structural modalities of a given object. In the context of computer vision, this corresponds to a learnable module that serves two purposes: (i) generate a realistic rendering of a 3D object (shape-to-image translation) and (ii) infer a realistic 3D shape from an image (image-to-shape translation). In this paper, we learn such a module while being conscious of the difficulties in obtaining large paired 2D-3D datasets. By leveraging generative domain translation methods, we are able to define a learning algorithm that requires only weak supervision, with unpaired data. The resulting model is not only able to perform 3D shape, pose, and texture inference from 2D images, but can also generate novel textured 3D shapes and renders, similar to a graphics pipeline. More specifically, our method (i) infers an explicit 3D mesh representation, (ii) utilizes example shapes to regularize inference, (iii) requires only an image mask (no keypoints or camera extrinsics), and (iv) has generative capabilities. While prior work explores subsets of these properties, their combination is novel. We demonstrate the utility of our learned representation, as well as its performance on image generation and unpaired 3D shape inference tasks.

We introduce an information-theoretic loss function, RankMI, and an associated training algorithm for deep representation learning for image retrieval. Our proposed framework consists of alternating updates to a network that estimates the divergence between distance distributions of matching and non-matching pairs of learned embeddings, and an embedding network that maximizes this estimate via sampled negatives. In addition, under this information-theoretic lens we draw connections between RankMI and commonly-used ranking losses, e.g., triplet loss. We extensively evaluate RankMI on several standard image retrieval datasets, namely, CUB-200-2011, CARS-196, and Stanford Online Products. Our method achieves competitive results or significant improvements over previous reported results on all datasets.

Normalizing flows are a class of probabilistic generative models which allow for both fast density computation and efficient sampling and are effective at modelling complex distributions like images. A drawback among current methods is their significant training cost, sometimes requiring months of GPU training time to achieve state-of-the-art results. This paper introduces Wavelet Flow, a multi-scale, normalizing flow architecture based on wavelets. A Wavelet Flow has an explicit representation of signal scale that inherently includes models of lower resolution signals and conditional generation of higher resolution signals, i.e., super resolution. A major advantage of Wavelet Flow is the ability to construct generative models for high resolution data (e.g., 1024 × 1024 images) that are impractical with previous models. Furthermore, Wavelet Flow is competitive with previous normalizing flows in terms of bits per dimension on standard (low resolution) benchmarks while being up to 15× faster to train.

Beginning with R-CNN, there has been a rapid advancement in two-stage object detection approaches. While two-stage approaches remain the state-of-the-art in object detection, anchor-free single-stage methods have been gaining momentum. We believe that the strength of the former is in their region of interest (ROI) pooling stage, while the latter simplifies the learning problem by converting object detection into dense per-pixel prediction tasks. In this paper, we propose to combine the strengths of each approach in a new architecture. In particular, we first define several auxiliary tasks related to object detection and generate dense per-pixel predictions using a shared feature extraction backbone. As a consequence of this architecture, the shared backbone is trained using both the standard object detection losses and these per-pixel ones. Moreover, by combining the features from dense predictions with those from the backbone, we realize a more discriminative representation for subsequent downstream processing. In addition, we feed the fused features into a novel multi-scale ROI pooling layer, followed by per-ROI predictions. We refer to our architecture as the Dense Auxiliary Tasks Network (DATNet). We present an extensive set of evaluations of our method on the Pascal VOC and COCO datasets and show considerable accuracy improvements over comparable baselines.

A wide range of computer vision algorithms rely on identifying sparse interest points in images and establishing correspondences between them. However, only a subset of the initially identified interest points results in true correspondences (inliers). In this paper, we seek a detector that finds the minimum number of points that are likely to result in an application-dependent "sufficient" number of inliers k. To quantify this goal, we introduce the "k-succinctness" metric. Extracting a minimum number of interest points is attractive for many applications, because it can reduce computational load, memory, and data transmission. Alongside succinctness, we introduce an unsupervised training methodology for interest point detectors that is based on predicting the probability of a given pixel being an inlier. In comparison to previous learned detectors, our method requires the least amount of data pre-processing. Our detector and other state-of-the-art detectors are extensively evaluated with respect to succinctness on popular public datasets covering both indoor and outdoor scenes, and both wide and narrow baselines. In certain cases, our detector is able to obtain an equivalent amount of inliers with as little as 60% of the amount of points of other detectors. The code and trained networks are provided at https://github.com/uzh-rpg/sips2_open.

Event cameras are vision sensors that record asynchronous streams of per-pixel brightness changes, referred to as "events”. They have appealing advantages over frame based cameras for computer vision, including high temporal resolution, high dynamic range, and no motion blur. Due to the sparse, non-uniform spatio-temporal layout of the event signal, pattern recognition algorithms typically aggregate events into a grid-based representation and subsequently process it by a standard vision pipeline, e.g., Convolutional Neural Network (CNN). In this work, we introduce a general framework to convert event streams into grid-based representations by means of strictly differentiable operations. Our framework comes with two main advantages: (i) allows learning the input event representation together with the task dedicated network in an end to end manner, and (ii) lays out a taxonomy that unifies the majority of extant event representations in the literature and identifies novel ones. Empirically, we show that our approach to learning the event representation end-to-end yields an improvement of approximately 12% on optical flow estimation and object recognition over state-of-the-art methods.

Recovering 3D full-body human pose is a challenging problem with many applications. It has been successfully addressed by motion capture systems with body worn markers and multiple cameras. In this paper, we address the more challenging case of not only using a single camera but also not leveraging markers: going directly from 2D appearance to 3D geometry. Deep learning approaches have shown remarkable abilities to discriminatively learn 2D appearance features. The missing piece is how to integrate 2D, 3D, and temporal information to recover 3D geometry and account for the uncertainties arising from the discriminative model. We introduce a novel approach that treats 2D joint locations as latent variables whose uncertainty distributions are given by a deep fully convolutional neural network. The unknown 3D poses are modeled by a sparse representation and the 3D parameter estimates are realized via an Expectation-Maximization algorithm, where it is shown that the 2D joint location uncertainties can be conveniently marginalized out during inference. Extensive evaluation on benchmark datasets shows that the proposed approach achieves greater accuracy over state-of-the-art baselines. Notably, the proposed approach does not require synchronized 2D-3D data for training and is applicable to "in-the-wild" images, which is demonstrated with the MPII dataset.

We introduce a two-stream model for dynamic texture synthesis. Our model is based on pre-trained convolutional networks (ConvNets) that target two independent tasks: (i) object recognition, and (ii) optical flow prediction. Given an input dynamic texture, statistics of filter responses from the object recognition ConvNet encapsulate the per-frame appearance of the input texture, while statistics of filter responses from the optical flow ConvNet model its dynamics. To generate a novel texture, a randomly initialized input sequence is optimized to match the feature statistics from each stream of an example texture. Inspired by recent work on image style transfer and enabled by the two-stream model, we also apply the synthesis approach to combine the texture appearance from one texture with the dynamics of another to generate entirely novel dynamic textures. We show that our approach generates novel, high quality samples that match both the framewise appearance and temporal evolution of input texture. Finally, we quantitatively evaluate our texture synthesis approach with a thorough user study.

Recent advances with Convolutional Networks (ConvNets) have shifted the bottleneck for many computer vision tasks to annotated data collection. In this paper, we present a geometry-driven approach to automatically collect annotations for human pose prediction tasks. Starting from a generic ConvNet for 2D human pose, and assuming a multi-view setup, we describe an automatic way to collect accurate 3D human pose annotations. We capitalize on constraints offered by the 3D geometry of the camera setup and the 3D structure of the human body to probabilistically combine per view 2D ConvNet predictions into a globally optimal 3D pose. This 3D pose is used as the basis for harvesting annotations. The benefit of the annotations produced automatically with our approach is demonstrated in two challenging settings: (i) fine-tuning a generic ConvNet-based 2D pose predictor to capture the discriminative aspects of a subjects appearance (i.e.,personalization), and (ii) training a ConvNet from scratch for single view 3D human pose prediction without leveraging 3D pose groundtruth. The proposed multi-view pose estimator achieves state-of-the-art results on standard benchmarks, demonstrating the effectiveness of our method in exploiting the available multi-view information.

This paper addresses the challenge of 3D human pose estimation from a single color image. Despite the general success of the end-to-end learning paradigm, top performing approaches employ a two-step solution consisting of a Convolutional Network (ConvNet) for 2D joint localization and a subsequent optimization step to recover 3D pose. In this paper, we identify the representation of 3D pose as a critical issue with current ConvNet approaches and make two important contributions towards validating the value of end-to-end learning for this task. First, we propose a fine discretization of the 3D space around the subject and train a ConvNet to predict per voxel likelihoods for each joint. This creates a natural representation for 3D pose and greatly improves performance over the direct regression of joint coordinates. Second, to further improve upon initial estimates, we employ a coarse-to-fine prediction scheme. This step addresses the large dimensionality increase and enables iterative refinement and repeated processing of the image features. The proposed approach outperforms all state-of-the-art methods on standard benchmarks achieving a relative error reduction greater than 30% on average. Additionally, we investigate using our volumetric representation in a related architecture which is suboptimal compared to our end-to-end approach, but is of practical interest, since it enables training when no image with corresponding 3D groundtruth is available, and allows us to present compelling results for in-the-wild images.

We introduce an approach to integrate segmentation information within a convolutional neural network (CNN). This counter-acts the tendency of CNNs to smooth information across regions and increases their spatial precision. To obtain segmentation information, we set up a CNN to provide an embedding space where region co-membership can be estimated based on Euclidean distance. We use these embeddings to compute a local attention mask relative to every neuron position. We incorporate such masks in CNNs and replace the convolution operation with a “segmentation-aware” variant that allows a neuron to selectively attend to inputs coming from its own region. We call the resulting network a segmentation-aware CNN because it adapts its filters at each image point according to local segmentation cues, while at the same time remaining fully-convolutional. We demonstrate the merit of our method on two widely different dense prediction tasks, that involve classification (semantic segmentation) and regression (optical flow). Our results show that in semantic segmentation we can replace DenseCRF inference with a cascade of segmentation-aware filters, and in optical flow we obtain clearly sharper responses than the ones obtained with comparable networks that do not use segmentation. In both cases segmentation-aware convolution yields systematic improvements over strong baselines.

This paper presents a novel approach to estimating the continuous six degree of freedom (6-DoF) pose (3D translation and rotation) of an object from a single RGB image. The approach combines semantic keypoints predicted by a convolutional network (convnet) with a deformable shape model. Unlike prior work, we are agnostic to whether the object is textured or textureless, as the convnet learns the optimal representation from the available training image data. Furthermore, the approach can be applied to instance- and class-based pose recovery. Empirically, we show that the proposed approach can accurately recover the 6-DoF object pose for both instance- and class-based scenarios with a cluttered background. For class-based object pose estimation, state-of-the-art accuracy is shown on the large-scale PASCAL3D+ dataset.

This paper addresses the challenge of 3D full-body human pose estimation from a monocular image sequence. Here, two cases are considered: (i) the image locations of the human joints are provided and (ii) the image locations of joints are unknown. In the former case, a novel approach is introduced that integrates a sparsity-driven 3D geometric prior and temporal smoothness. In the latter case, the former case is extended by treating the image locations of the joints as latent variables to take into account considerable uncertainties in 2D joint locations. A deep fully convolutional network is trained to predict the uncertainty maps of the 2D joint locations. The 3D pose estimates are realized via an Expectation-Maximization algorithm over the entire sequence, where it is shown that the 2D joint location uncertainties can be conveniently marginalized out during inference. Empirical evaluation on the Human3.6M dataset shows that the proposed approaches achieve greater 3D pose estimation accuracy over state-of-the-art baselines. Further, the proposed approach outperforms a publicly available 2D pose estimation baseline on the challenging PennAction dataset.

This paper presents a new state-of-the-art for document image classification and retrieval, using features learned by deep convolutional neural networks (CNNs). In object and scene analysis, deep neural nets are capable of learning a hierarchical chain of abstraction from pixel inputs to concise and descriptive representations. The current work explores this capacity in the realm of document analysis, and confirms that this representation strategy is superior to a variety of popular handcrafted alternatives. Extensive experiments show that (i) features extracted from CNNs are robust to compression, (ii) CNNs trained on non-document images transfer well to document analysis tasks, and (iii) enforcing region-specific feature-learning is unnecessary given sufficient training data. This work also makes available a new labelled subset of the IIT-CDIP collection, containing 400,000 document images across 16 categories.

We present a novel approach for detecting objects and estimating their 3D pose in single images of cluttered scenes. Objects are given in terms of 3D models without accompanying texture cues. A deformable parts-based model is trained on clusters of silhouettes of similar poses and produces hypotheses about possible object locations at test time. Objects are simultaneously segmented and verified inside each hypothesis bounding region by selecting the set of superpixels whose collective shape matches the model silhouette. A final iteration on the 6-DOF object pose minimizes the distance between the selected image contours and the actual projection of the 3D model. We demonstrate successful grasps using our detection and pose estimate with a PR2 robot. Extensive evaluation with a novel ground truth dataset shows the considerable benefit of using shape-driven cues for detecting objects in heavily cluttered scenes.

This paper presents a novel approach for analyzing human actions in non-scripted, unconstrained video settings based on volumetric, x-y-t, patch classifiers, termed actemes. Unlike previous action-related work, the discovery of patch classifiers is posed as a strongly-supervised process. Specifically, key point labels (e.g., position) across space time are used in a data-driven training process to discover patches that are highly clustered in the space time key point configuration space. To support this process, a new human action dataset consisting of challenging consumer videos is introduced, where notably the action label, the 2D position of a set of key points and their visibilities are provided for each video frame. On a novel input video, each acteme is used in a sliding volume scheme to yield a set of sparse, non-overlapping detections. These detections provide the intermediate substrate for segmenting out the action. For action classification, the proposed representation shows significant improvement over state-of-the-art low-level features, while providing spatiotemporal localization as additional output, which sheds further light into detailed action understanding.

This paper provides a unified framework for the interrelated topics of action spotting, the spatiotemporal detection and localization of human actions in video, and action recognition, the classification of a given video into one of several predefined categories. A novel compact local descriptor of video dynamics in the context of action spotting and recognition is introduced based on visual spacetime oriented energy measurements. This descriptor is efficiently computed directly from raw image intensity data and thereby forgoes the problems typically associated with flow-based features. Importantly, the descriptor allows for the comparison of the underlying dynamics of two spacetime video segments irrespective of spatial appearance, such as differences induced by clothing, and with robustness to clutter. An associated similarity measure is introduced that admits efficient exhaustive search for an action template, derived from a single exemplar video, across candidate video sequences. The general approach presented for action spotting and recognition is amenable to efficient implementation, which is deemed critical for many important applications. For action spotting, details of a real-time GPU-based instantiation of the proposed approach are provided. Empirical evaluation of both action spotting and action recognition on challenging datasets suggests the efficacy of the proposed approach, with state-of-the-art performance documented on standard datasets.

Natural scene classification is a fundamental challenge in computer vision. By far, the majority of studies have limited their scope to scenes from single image stills and thereby ignore potentially informative temporal cues. The current paper is concerned with determining the degree of performance gain in considering short videos for recognizing natural scenes. Towards this end, the impact of multiscale orientation measurements on scene classification is systematically investigated, as related to: (i) spatial appearance, (ii) temporal dynamics and (iii) joint spatial appearance and dynamics. These measurements in visual space, x-y, and spacetime, x-y-t, are recovered by a bank of spatiotemporal oriented energy filters. In addition, a new data set is introduced that contains 420 image sequences spanning fourteen scene categories, with temporal scene information due to objects and surfaces decoupled from camera-induced ones. This data set is used to evaluate classification performance of the various orientation-related representations, as well as state-of-the-art alternatives. It is shown that a notable performance increase is realized by spatiotemporal approaches in comparison to purely spatial or purely temporal methods.

In this paper, we give a new double twist to the robot localization problem. We solve the problem for the case of prior maps which are semantically annotated perhaps even sketched by hand. Data association is achieved not through the detection of visual features but the detection of object classes used in the annotation of the prior maps. To avoid the caveats of general object recognition, we propose a new representation of the query images that consists of a vector of the detection scores for each object class. Given such soft object detections we are able to create hypotheses about pose and to refine them through particle filtering. As opposed to small confined office and kitchen spaces, our experiment takes place in a large open urban rail station with multiple semantically ambiguous places. The success of our approach shows that our new representation is a robust way to exploit the plethora of existing prior maps for GPS-denied environments avoiding the data association problems when matching point clouds or visual features.

This paper is concerned with the representation and recognition of the observed dynamics (i.e., excluding purely spatial appearance cues) of spacetime texture based on a spatiotemporal orientation analysis. The term "spacetime texture" is taken to refer to patterns in visual spacetime, (x,y,t), that primarily are characterized by the aggregate dynamic properties of elements or local measurements accumulated over a region of spatiotemporal support, rather than in terms of the dynamics of individual constituents. Examples include image sequences of natural processes that exhibit stochastic dynamics (e.g., fire, water, and windblown vegetation) as well as images of simpler dynamics when analyzed in terms of aggregate region properties (e.g., uniform motion of elements in imagery, such as pedestrians and vehicular traffic). Spacetime texture representation and recognition is important as it provides an early means of capturing the structure of an ensuing image stream in a meaningful fashion. Toward such ends, a novel approach to spacetime texture representation and an associated recognition method are described based on distributions (histograms) of spacetime orientation structure. Empirical evaluation on both standard and original image data sets shows the promise of the approach, including significant improvement over alternative state-of-the-art approaches in recognizing the same pattern from different viewpoints.

This paper describes a system for classifying traffic congestion videos based on their observed visual dynamics. Central to the proposed system is treating traffic flow identification as an instance of dynamic texture classification. More specifically, a recent discriminative model of dynamic textures is adapted for the special case of traffic flows. This approach avoids the need for segmentation, tracking and motion estimation that typify extant approaches. Classification is based on matching distributions (or histograms) of spacetime orientation structure. Empirical evaluation on a publicly available data set shows high classification performance and robustness to typical environmental conditions (e.g., variable lighting).

This paper addresses action spotting, the spatiotemporal detection and localization of human actions in video. A novel compact local descriptor of video dynamics in the context of action spotting is introduced based on visual spacetime oriented energy measurements. This descriptor is efficiently computed directly from raw image intensity data and thereby forgoes the problems typically associated with flow-based features. An important aspect of the descriptor is that it allows for the comparison of the underlying dynamics of two spacetime video segments irrespective of spatial appearance, such as differences induced by clothing, and with robustness to clutter. An associated similarity measure is introduced that admits efficient exhaustive search for an action template across candidate video sequences. Empirical evaluation of the approach on a set of challenging natural videos suggests its efficacy.

This paper addresses the challenge of recognizing dynamic textures based on their observed visual dynamics. Typically, the term dynamic texture is used with reference to image sequences of various natural processes that exhibit stochastic dynamics (e.g., smoke, water and windblown vegetation); although, it applies equally well to images of simpler dynamics when analyzed in terms of aggregate region properties (e.g., uniform motion of elements in traffic video). In this paper, a novel approach to dynamic texture representation and an associated recognition method are proposed. The approach pursued here recognizes dynamic textures based on matching distributions (histograms) of spacetime orientation structure. Empirical evaluation on a standard database with controls to remove the effects of identical viewpoint demonstrates that the proposed approach achieves superior performance over alternative state-of-the-art methods.

A theoretical investigation of the frequency structure of multiplicative image motion signals is presented, e.g., as associated with translucency phenomena. Previous work has claimed that the multiplicative composition of visual signals generally results in the annihilation of oriented structure in the spectral domain. As a result, research has focused on multiplicative signals in highly specialized scenarios where highly structured spectral signatures are prevalent, or introduced a nonlinearity to transform the multiplicative image signal to an additive one. In contrast, in this paper, it is shown that oriented structure is present in multiplicative cases when natural domain constraints are taken into account. This analysis suggests that the various instances of naturally occurring multiple motion structures can be treated in a unified manner. As an example application of the developed theory, a multiple motion estimator previously proposed for translation, additive transparency, and occlusion is adapted to multiplicative image motions. This estimator is shown to yield superior performance over the alternative practice of introducing a nonlinear preprocessing step.

Spatiotemporal data is associated with vast amounts of raw samples. Given the limited computational resources typically available, an initial organization of this data supporting semantically meaningful lines of inquiry would facilitate efficient processing. In this paper, a new representation for grouping raw image data into a set of coherent spacetime regions is proposed. Unique in this proposal is that coherency is related to a richer description of local spacetime structure than generally considered. In particular, the representation describes the presence of particular oriented spacetime structures in a distributed manner. A key advantage of this representation is its ability to signal the presence of multiple oriented structures at a given spacetime location. More generally, the abstraction allows for the description and grouping of motion and non-motion-related patterns in a uniform manner. Empirical evaluation of the grouping method on synthetic and challenging natural imagery suggests its efficacy.

Histograms represent a popular means for feature representation. This paper is concerned with the problem of exhaustive histogram-based image search. Several standard histogram construction methods are explored, including the conventional approach, Huangpsilas method, and the state-of-the-art integral histogram. In addition, we present a novel multiscale histogram-based search algorithm, termed the distributive histogram, that can be evaluated exhaustively in a fast and memory efficient manner. An extensive systematic empirical evaluation is presented that explores the computational and storage consequences of altering the search image and histogram bin sizes. Experiments reveal up to an eight-fold decrease in computation time and hundreds- to thousands-fold decrease of memory use of the proposed distributive histogram in comparison to the integral histogram. Finally, we conclude with a discussion on the relative merits between the various approaches considered in the paper.

In this paper, we place the integral image-based approach for multi-scale feature construction, popularized by Viola and Jones, into a common framework of understanding. The integral image within this framework represents space variant image filtering with the zero-order B-spline. Given this framework, we propose efficiently computable higher-order B-spline image features based on generalized integral images that have the potential to be more accurate, yet efficient as compared to previous integral image-based efforts.

This paper details the construction of three-dimensional separable steerable filters. The approach presented is an extension of the construction of two-dimensional separable steerable filters outlined in W.T. Freeman and E.H. Adelson (1991). Additionally, three-dimensional separable steerable filters, both continuous and discrete versions, for the second derivative of the Gaussian and its Hilbert transform are reported. Experimental evaluation demonstrates that the errors in the constructed separable filters are negligible.