Today’s most advanced vision-language models (VLMs) remain proprietary. The strongest open-weight models rely heavily on synthetic data from proprietary VLMs to achieve good performance, effectively distilling these closed VLMs into open ones. As a result, the community has been missing foundational knowledge about how to build performant VLMs from scratch. We present Molmo, a new family of VLMs that are state-of-the-art in their class of openness. Our key contribution is a collection of new datasets called PixMo, including a dataset of highly detailed image captions for pre-training, a free-form image Q&A dataset for fine-tuning, and an innovative 2D pointing dataset, all collected without the use of external VLMs. The success of our approach relies on careful modeling choices, a well- tuned training pipeline, and, most critically, the quality of our newly collected datasets. Our best-in-class 72B model not only outperforms others in the class of open weight and data models, but also outperforms larger proprietary models including Claude 3.5 Sonnet, and Gemini 1.5 Pro and Flash, second only to GPT-4o based on both academic benchmarks and on a large human evaluation. Our model weights, new datasets, and source code are available at https://molmo.allenai.org/blog.

Despite the unprecedented progress in the field of 3D generation, current systems still often fail to produce high-quality 3D assets that are visually appealing and geometrically and semantically consistent across multiple viewpoints. To effectively assess the quality of the generated 3D data, there is a need for a reliable 3D evaluation tool. Unfortunately, existing 3D evaluation metrics often overlook the geometric quality of generated assets or merely rely on black-box multimodal large language models for coarse assessment. In this paper, we introduce Eval3D, a fine-grained, interpretable evaluation tool that can faithfully evaluate the quality of generated 3D assets based on various distinct yet complementary criteria. Our key observation is that many desired properties of 3D generation, such as semantic and geometric consistency, can be effectively captured by measuring the consistency among various foundation models and tools. We thus leverage a diverse set of models and tools as probes to evaluate the inconsistency of generated 3D assets across different aspects. Compared to prior work, Eval3D provides pixel-wise measurement, enables accurate 3D spatial feedback, and aligns more closely with human judgments. We comprehensively evaluate existing 3D generation models using Eval3D and highlight the limitations and challenges of current models. Project page: http://eval3d.github.io.

Reasoning over visual relationships—spatial, functional, interactional, social, etc.—is considered to be a fundamental component of human cognition. Yet, despite the major advances in visual comprehension in multimodal language models (MLMs), precise reasoning over relationships and their generations remains a challenge. We introduce Robin: an MLM instruction-tuned with densely annotated relationships capable of constructing high-quality dense scene graphs at scale. To train Robin, we curate SVG1, a synthetic scene graph dataset by completing the missing relations of selected objects in existing scene graphs using a teacher MLM and a carefully designed filtering process to ensure high-quality. To generate more accurate and rich scene graphs at scale for any image, we introduce SG-Edit: a self-distillation framework where GPT-4o further refines Robin’s predicted scene graphs by removing unlikely relations and/or suggesting relevant ones. In total, our dataset contains 146K images and 5.6M relationships for 2.6M objects. Results show that our Robin-3B model, despite being trained on less than 3 million instances, outperforms similar-size models trained on over 300 million instances on relationship understanding benchmarks, and even surpasses larger models up to 13B parameters. Notably, it achieves state-of-the-art performance in referring expression comprehension with a score of 88.2, surpassing the previous best of 87.4. Our results suggest that training on the refined scene graph data is crucial to maintaining high performance across diverse visual reasoning tasks2.

Creating virtual digital replicas from real-world data unlocks significant potential across domains like gaming and robotics. In this paper, we present DRAWER, a novel framework that converts a video of a static indoor scene into a photorealistic and interactive digital environment. Our approach centers on two main contributions: (i) a reconstruction module based on a dual scene representation that reconstructs the scene with fine-grained geometric details, and (ii) an articulation module that identifies articulation types and hinge positions, reconstructs simulatable shapes and appearances and integrates them into the scene. The resulting virtual environment is photorealistic, interactive, and runs in real time, with compatibility for game engines and robotic simulation platforms. We demonstrate the potential of DRAWER by using it to automatically create an interactive game in Unreal Engine and to enable real-to-sim-to-real transfer for robotics applications. Project page: here.

Unconditional flow-matching trains diffusion models to transport samples from a source distribution to a target distribution by enforcing that the flows between sample pairs are unique. However, in conditional settings (e.g., class-conditioned models), this uniqueness is no longer guaranteed--flows from different conditions may overlap, leading to more ambiguous generations. We introduce Contrastive Flow Matching, an extension to the flow matching objective that explicitly enforces uniqueness across all conditional flows, enhancing condition separation. Our approach adds a contrastive objective that maximizes dissimilarities between predicted flows from arbitrary sample pairs. We validate Contrastive Flow Matching by conducting extensive experiments across varying model architectures on both class-conditioned (ImageNet-1k) and text-to-image (CC3M) benchmarks. Notably, we find that training models with Contrastive Flow Matching (1) improves training speed by a factor of up to 9x, (2) requires up to 5x fewer de-noising steps and (3) lowers FID by up to 8.9 compared to training the same models with flow matching. We release our code at: https://github.com/gstoica27/DeltaFM.git.

We introduce OLMoE, a fully open, state-of-the-art language model leveraging sparse Mixture-of-Experts (MoE). OLMoE-1B-7B has 7 billion (B) parameters but uses only 1B per input token. We pretrain it on 5 trillion tokens and further adapt it to create OLMoE-1B-7B-Instruct. Our models outperform all available models with similar active parameters, even surpassing larger ones like Llama2-13B-Chat and DeepSeekMoE-16B. We present novel findings on MoE training, define and analyze new routing properties showing high specialization in our model, and open-source all our work: model weights, training data, code, and logs.

In recent years, the Robotics field has initiated several efforts toward building generalist robot policies through large-scale multi-task Behavior Cloning. However, direct deployments of these policies have led to unsatisfactory performance, where the policy struggles with unseen states and tasks. How can we break through the performance plateau of these models and elevate their capabilities to new heights? In this paper, we propose FLaRe, a large-scale Reinforcement Learning fine-tuning framework that integrates robust pre-trained representations, large-scale training, and gradient stabilization techniques. Our method aligns pre-trained policies towards task completion, achieving state-of-the-art (SoTA) performance both on previously demonstrated and on entirely novel tasks and embodiments. Specifically, on a set of long-horizon mobile manipulation tasks, FLaRe achieves an average success rate of 79.5% in unseen environments, with absolute improvements of$+23.6 \%$in simulation and$+30.7 \%$on real robots over prior SoTA methods. By utilizing only sparse rewards, our approach can enable generalizing to new capabilities beyond the pretraining data with minimal human effort. Moreover, we demonstrate rapid adaptation to new embodiments and behaviors with less than a day of fine-tuning. Videos, code, and appendix can be found on the project website at robot-flare.github.io

We introduce LOKI, a compute-efficient framework for co-designing morphologies and control policies that generalize across unseen tasks. Inspired by biological adaptation—where animals quickly adjust to morphological changes—our method overcomes the inefficiencies of traditional evolutionary and quality-diversity algorithms. We propose learning convergent functions: shared control policies trained across clusters of morphologically similar designs in a learned latent space, drastically reducing the training cost per design. Simultaneously, we promote divergent forms by replacing mutation with dynamic local search, enabling broader exploration and preventing premature convergence. The policy reuse allows us to explore $\sim780\times$ more designs using 78\% fewer simulation steps and 40\% less compute per design. Local competition paired with a broader search results in a diverse set of high-performing final morphologies. Using the UNIMAL design space and a flat-terrain locomotion task, LOKI discovers a rich variety of designs—ranging from quadrupeds to crabs, bipedals, and spinners—far more diverse than those produced by prior work. These morphologies also transfer better to unseen downstream tasks in agility, stability, and manipulation domains (e.g. $2 \times$ higher reward on bump and push box incline tasks). Overall, our approach produces designs that are both diverse and adaptable, with substantially greater sample efficiency than existing co-design methods.

Current image generation methods are based on a two-stage training approach. In stage 1, an auto-encoder is trained to compress an image into a latent space; in stage 2, a generative model is trained to learn a distribution over that latent space. This reveals a fundamental trade-off, do we compress more aggressively to make the latent distribution easier for the stage 2 model to learn even if it makes reconstruction worse? We study this problem in the context of discrete, auto-regressive image generation. Through the lens of scaling laws, we show that smaller stage 2 models can benefit from more compressed stage 1 latents even if reconstruction performance worsens, demonstrating that generation modeling capacity plays a role in this trade-off. Diving deeper, we rigorously study the connection between compute scaling and the stage 1 rate-distortion trade-off. Next, we introduce Causally Regularized Tokenization (CRT), which uses knowledge of the stage 2 generation modeling procedure to embed useful inductive biases in stage 1 latents. This regularization improves stage 2 generation performance better by making the tokens easier to model without affecting the stage 1 compression rate and marginally affecting distortion: we are able to improve compute efficiency 2-3$\times$ over baseline. Finally, we use CRT with further optimizations to the visual tokenizer setup to result in a generative pipeline that matches LlamaGen-3B generation performance (2.18 FID) with half the tokens per image (256 vs. 576) and a fourth the total model parameters (775M vs. 3.1B) while using the same architecture and inference procedure.

We introduce FlexOLMo, a new class of language models (LMs) that supports (1) distributed training without data sharing, where different model parameters are independently trained on private datasets, and (2) data-flexible inference, where these parameters along with their associated data can be easily included or excluded from model inferences with no further training. FlexOLMo employs a mixture-of-experts (MoE) architecture where each expert is trained independently on private datasets and later integrated through a new nonparametric routing without any joint training across datasets. FlexOLMo is trained on FLEXMIX, a corpus we curate comprising seven restricted sets, either real or realistic approximations, alongside publicly available datasets. We evaluate models with up to 37 billion parameters (20 billion active) on 31 diverse downstream tasks. We show that a general expert trained on public data can be effectively combined with independently trained experts from other data owners significantly benefiting from these restricted sets (an average 41% relative improvement) while allowing flexible opt-out at inference time (e.g., for users without appropriate licenses or permissions). Our approach also outperforms prior model merging methods by 10.1% on average and surpasses the standard MoE trained without data restrictions using the same training FLOPs. Altogether, FlexOLMo enables training on restricted data while keeping data local and supports fine-grained control of data access at inference.

Embodied AI models often employ off the shelf vision backbones like CLIP to encode their visual observations. Although such general purpose representations encode rich syntactic and semantic information about the scene, much of this information is often irrelevant to the specific task at hand. This introduces noise within the learning process and distracts the agent's focus from task-relevant visual cues. Inspired by selective attention in humans—the process through which people filter their perception based on their experiences, knowledge, and the task at hand—we introduce a parameter-efficient approach to filter visual stimuli for embodied AI. Our approach induces a task-conditioned bottleneck using a small learnable codebook module. This codebook is trained jointly to optimize task reward and acts as a task-conditioned selective filter over the visual observation. Our experiments showcase state-of-the-art performance for object goal navigation and object displacement across $5$ benchmarks, ProcTHOR, ArchitecTHOR, RoboTHOR, AI2-iTHOR, and ManipulaTHOR. The filtered representations produced by the codebook are also able generalize better and converge faster when adapted to other simulation environments such as Habitat. Our qualitative analyses show that agents explore their environments more effectively and their representations retain task-relevant information like target object recognition while ignoring superfluous information about other objects.

Foundation models are applied in a broad spectrum of settings with different inference constraints, from massive multi-accelerator clusters to resource-constrained standalone mobile devices. However, the substantial costs associated with training these models often limit the number of unique model sizes that can be offered. Consequently, practitioners are compelled to select a model that may not be optimally aligned with their specific latency and cost requirements. We present MatFormer, a novel Transformer architecture designed to provide elastic inference across diverse deployment constraints. MatFormer achieves this by incorporating a nested Feed Forward Network (FFN) block structure within a standard Transformer model. During training, we optimize the parameters of multiple nested FFN blocks with varying sizes, enabling the extraction of hundreds of accurate smaller models without incurring additional computational costs. We empirically validate the efficacy of MatFormer across different model classes (decoders and encoders) and modalities (language and vision), demonstrating its potential for real-world deployment. We show that a 850M decoder-only MatFormer language model (MatLM) allows us to extract multiple smaller models spanning from 582M to 850M parameters, each exhibiting better validation loss and one-shot downstream evaluations than independently trained counterparts. Furthermore, we observe that smaller encoders extracted from a universal MatFormer-based ViT (MatViT) encoder preserve the metric-space structure for adaptive large-scale retrieval. Finally, we showcase that speculative decoding with the accurate and consistent submodels extracted from MatFormer can lead to significant reduction in inference latency.

Our world is full of varied actions and moves in specialized fields that we, as humans, seek to identify and learn about. To evaluate the effectiveness of multi-modal models in helping us recognize such fine-grained actions, we introduce ActionAtlas, a video question answering (VideoQA) benchmark on fine-grained action recognition with short videos across various sports. ActionAtlas contains 554 videos spanning 284 actions across 42 sports with 1161 actions as total potential choices. Unlike most existing action recognition benchmarks that focus on simplistic actions, often identifiable from a single frame, ActionAtlas focuses on intricate movements and tests the models' ability to discern subtle differences. Additionally, each video in ActionAtlas also includes a question, which helps to more accurately pinpoint the action's performer in scenarios where multiple individuals are involved in different activities. We evaluate proprietary and open models on this benchmark and show that the state-of-the-art models only perform at most 48.73% accurately where random chance is 20%. Furthermore, our results show that a high frame sampling rate is essential for recognizing actions in ActionAtlas, a feature that current top proprietary models like Gemini lack in their default settings.

Many applications today provide users with multiple auto-complete drafts as they type, including GitHub's code completion, Gmail's smart compose, and Apple's messaging auto-suggestions. Under the hood, language models support this by running an autoregressive inference pass to provide a draft. Consequently, providing $k$ drafts to the user requires running an expensive language model $k$ times. To alleviate the computation cost of running $k$ inference passes, we propose Superposed Decoding, a new decoding algorithm that generates $k$ drafts at the computation cost of one autoregressive inference pass. We achieve this by feeding a superposition of the most recent token embeddings from the $k$ drafts as input to the next decoding step of the language model. At every inference step we combine the $k$ drafts with the top-$k$ tokens to get $k^2$ new drafts and cache the $k$ most likely options, using an n-gram interpolation with minimal compute overhead to filter out incoherent generations. Our experiments show that $k$ drafts from Superposed Decoding are at least as coherent and factual as Nucleus Sampling and Greedy Decoding respectively, while being at least $2.44\times$ faster for $k\ge3$. In a compute-normalized setting, user evaluations demonstrably favor text generated by Superposed Decoding over Nucleus Sampling. Superposed Decoding can also be combined with other decoding strategies, resulting in universal coverage gains when scaling inference time compute. Code and more examples open-sourced at https://github.com/RAIVNLab/SuperposedDecoding.

Three-dimensional (3D) understanding of objects and scenes play a key role in humans' ability to interact with the world and has been an active area of research in computer vision, graphics, and robotics. Large scale synthetic and object-centric 3D datasets have shown to be effective in training models that have 3D understanding of objects. However, applying a similar approach to real-world objects and scenes is difficult due to a lack of large-scale data. Videos are a potential source for real-world 3D data, but finding diverse yet corresponding views of the same content have shown to be difficult at scale. Furthermore, standard videos come with fixed viewpoints, determined at the time of capture. This restricts the ability to access scenes from a variety of more diverse and potentially useful perspectives. We argue that large scale ODIN videos can address these limitations to provide scalable corresponding frames from diverse views. In this paper we introduce 360-1M, a 360° video dataset consisting of 1 million videos, and a process for efficiently finding corresponding frames from diverse viewpoints at scale. We train our diffusion-based model, ODIN, on 360-1M. Empowered by the largest real-world, multi-view dataset to date, ODIN is able to freely generate novel views of real-world scenes. Unlike previous methods, ODIN can move the camera through the environment, enabling the model to infer the geometry and layout of the scene. Additionally, we show improved performance on standard novel view synthesis and 3D reconstruction benchmarks.

Benchmarks for large multimodal language models (MLMs) now serve to simultaneously assess the general capabilities of models instead of evaluating for a specific capability. As a result, when a developer wants to identify which models to use for their application, they are overwhelmed by the number of benchmarks and remain uncertain about which benchmark's results are most reflective of their specific use case. This paper introduces Task-Me-Anything, a benchmark generation engine which produces a benchmark tailored to a user's needs. Task-Me-Anything maintains an extendable taxonomy of visual assets and can programmatically generate a vast number of task instances. Additionally, it algorithmically addresses user queries regarding MLM performance efficiently within a computational budget. It contains 113K images, 10K videos, 2K 3D object assets, over 365 object categories, 655 attributes, and 335 relationships. It can generate 500M image/video question-answering pairs, which focus on evaluating MLM perceptual capabilities. Task-Me-Anything reveals critical insights: open-source MLMs excel in object and attribute recognition but lack spatial and temporal understanding; each model exhibits unique strengths and weaknesses; larger models generally perform better, though exceptions exist; and GPT4O demonstrates challenges in recognizing rotating/moving objects and distinguishing colors.

Training embodied agents in simulation has become mainstream for the embodied AI community. However, these agents often struggle when deployed in the physical world due to their inability to generalize to real-world environments. In this paper, we present Phone2Proc, a method that uses a 10-minute phone scan and conditional procedural generation to create a distribution of training scenes that are semantically similar to the target environment. The generated scenes are conditioned on the wall layout and arrangement of large objects from the scan, while also sampling lighting, clutter, surface textures, and instances of smaller objects with randomized placement and materials. Leveraging just a simple RGB camera, training with Phone2Proc shows massive improvements from 34.7% to 70.7% success rate in sim-to-real ObjectNav performance across a test suite of over 200 trials in diverse real-world environments, including homes, offices, and RoboTHOR. Furthermore, Phone2Proc's diverse distribution of generated scenes makes agents remarkably robust to changes in the real world, such as human movement, object rearrangement, lighting changes, or clutter.

Massive data corpora like WebText, Wikipedia, Conceptual Captions, WebImageText, and LAION have propelled recent dramatic progress in AI. Large neural models trained on such datasets produce impressive results and top many of today's benchmarks. A notable omisslion within this family of large-scale datasets is 3D data. Despite considerable interest and potential applications in 3D vision, datasets of high-fidelity 3D models continue to be mid-sized with limited diversity of object categories. Addressing this gap, we present Objaverse 1.0, a large dataset of objects with 800K + (and growing) 3D models with descriptive captions, tags, and animations. Objaverse improves upon present day 3D repositories in terms of scale, number of categories, and in the visual diversity of instances within a category. We demonstrate the large potential of Objaverse via four diverse applications: training generative 3D models, improving tail category segmentation on the LVIS benchmark, training open-vocabulary object-navigation models for Embodied AI, and creating a new benchmark for robustness analysis of vision models. Objaverse can open new directions for research and enable new applications across the field of AI.

We propose Dataset Reinforcement, a strategy to improve a dataset once such that the accuracy of any model architecture trained on the reinforced dataset is improved at no additional training cost for users. We propose a Dataset Reinforcement strategy based on data augmentation and knowledge distillation. Our generic strategy is designed based on extensive analysis across CNN- and transformer-based models and performing large-scale study of distillation with state-of-the-art models with various data augmentations. We create a reinforced version of the ImageNet training dataset, called ImageNet+, as well as reinforced datasets CIFAR-100+, Flowers-102+, and Food-101+. Models trained with ImageNet+are more accurate, robust, and calibrated, and transfer well to downstream tasks (e.g., segmentation and detection). As an example, the accuracy of ResNet-50 improves by 1.7% on the ImageNet validation set, 3.5% on ImageNetV2, and 10.0% on ImageNet-R. Expected Calibration Error (ECE) on the ImageNet validation set is also reduced by 9.9%. Using this backbone with Mask-RCNN for object detection on MS-COCO, the mean average precision improves by 0.8%. We reach similar gains for MobileNets, ViTs, and Swin-Transformers. For MobileNetV3 and Swin-Tiny, we observe significant improvements on ImageNet-R/A/C of up to 20% improved robustness. Models pretrained on ImageNet+and fine-tuned on CIFAR-100+, Flowers-102+, and Food-101+, reach up to 3.4% improved accuracy. The code, datasets, and pretrained models are available at https://github.com/apple/ml-dr.

Open-vocabulary models are a promising new paradigm for image classification. Unlike traditional classification models, open-vocabulary models classify among any arbitrary set of categories specified with natural language during inference. This natural language, called "prompts", typically consists of a set of hand-written templates (e.g., "a photo of a {}") which are completed with each of the category names. This work introduces a simple method to generate higher accuracy prompts, without relying on any explicit knowledge of the task domain and with far fewer hand-constructed sentences. To achieve this, we combine open-vocabulary models with large language models (LLMs) to create Customized Prompts via Language models (CuPL, pronounced "couple"). In particular, we leverage the knowledge contained in LLMs in order to generate many descriptive sentences that contain important discriminating characteristics of the image categories. This allows the model to place a greater importance on these regions in the image when making predictions. We find that this straightforward and general approach improves accuracy on a range of zero-shot image classification benchmarks, including over one percentage point gain on ImageNet. Finally, this simple baseline requires no additional training and remains completely zero-shot. Code available at https://github.com/sarahpratt/CuPL.

A household robot should be able to navigate to target objects without requiring users to first annotate everything in their home. Most current approaches to object navigation do not test on real robots and rely solely on reconstructed scans of houses and their expensively labeled semantic 3D meshes. In this work, our goal is to build an agent that builds self-supervised models of the world via exploration, the same as a child might - thus we (1) eschew the expense of labeled 3D mesh and (2) enable self-supervised in-situ finetuning in the real world. We identify a strong source of self-supervision (Location Consistency - LocCon) that can train all components of an ObjectNav agent, using unannotated simulated houses. Our key insight is that embodied agents can leverage location consistency as a self-supervision signal - collecting images from different views/angles and applying contrastive learning. We show that our agent can perform competitively in the real world and simulation. Our results also indicate that supervised training with 3D mesh annotations causes models to learn simulation artifacts, which are not transferrable to the real world. In contrast, our LocCon shows the most robust transfer in the real world among the set of models we compare to, and that the real-world performance of all models can be further improved with self-supervised LocCon in-situ training.

Natural language processing and 2D vision models have attained remarkable proficiency on many tasks primarily by escalating the scale of training data. However, 3D vision tasks have not seen the same progress, in part due to the challenges of acquiring high-quality 3D data. In this work, we present Objaverse-XL, a dataset of over 10 million 3D objects. Our compilation comprises deduplicated 3D objects from a diverse set of sources, including manually designed objects, photogrammetry scans of landmarks and everyday items, and professional scans of historic and antique artifacts. Representing the largest scale and diversity in the realm of 3D datasets, Objaverse-XL enables significant new possibilities for 3D vision. Our experiments demonstrate the vast improvements enabled with the scale provided by Objaverse-XL. We show that by training Zero123 on novel view synthesis, utilizing over 100 million multi-view rendered images, we achieve strong zero-shot generalization abilities. We hope that releasing Objaverse-XL will enable further innovations in the field of 3D vision at scale.

Multimodal datasets are a critical component in recent breakthroughs such as CLIP, Stable Diffusion and GPT-4, yet their design does not receive the same research attention as model architectures or training algorithms. To address this shortcoming in the machine learning ecosystem, we introduce DataComp, a testbed for dataset experiments centered around a new candidate pool of 12.8 billion image-text pairs from Common Crawl. Participants in our benchmark design new filtering techniques or curate new data sources and then evaluate their new dataset by running our standardized CLIP training code and testing the resulting model on 38 downstream test sets. Our benchmark consists of multiple compute scales spanning four orders of magnitude, which enables the study of scaling trends and makes the benchmark accessible to researchers with varying resources. Our baseline experiments show that the DataComp workflow leads to better training sets. Our best baseline, DataComp-1B, enables training a CLIP ViT-L/14 from scratch to 79.2% zero-shot accuracy on ImageNet, outperforming OpenAI's CLIP ViT-L/14 by 3.7 percentage points while using the same training procedure and compute. We release \datanet and all accompanying code at www.datacomp.ai.

Instruction following vision-language (VL) models offer a flexible interface that supports a broad range of multimodal tasks in a zero-shot fashion. However, interfaces that operate on full images do not directly enable the user to “point to" and access specific regions within images. This capability is important not only to support reference-grounded VL benchmarks, but also, for practical applications that require precise within-image reasoning. We build Localized Visual Commonsense model which allows users to specify (multiple) regions- as-input. We train our model by sampling localized commonsense knowledge from a large language model (LLM): specifically, we prompt a LLM to collect commonsense knowledge given a global literal image description and a local literal region description automatically generated by a set of VL models. This pipeline is scalable and fully automatic, as no aligned or human-authored image and text pairs are required. With a separately trained critic model that selects high quality examples, we find that training on the localized commonsense corpus expanded solely from images can successfully distill existing VL models to support a reference-as-input interface. Empirical results and human evaluations in zero-shot settings demonstrate that our distillation method results in more precise VL models of reasoning compared to a baseline of passing a generated referring expression.

Pre-training has been widely adopted in deep learning to improve model performance, especially when the training data for a target task is limited. In our work, we seek to understand the implications of this training strategy on the generalization properties of downstream models. More specifically, we ask the following question: how do properties of the pre-training distribution affect the robustness of a fine-tuned model? The properties we explore include the label space, label semantics, image diversity, data domains, and data quantity of the pre-training distribution. We find that the primary factor influencing downstream effective robustness (Taori et al., 2020) is data quantity, while other factors have limited significance. For example, reducing the number of ImageNet pre-training classes by 4x while increasing the number of images per class by 4x (that is, keeping total data quantity fixed) does not impact the robustness of fine-tuned models. We demonstrate our findings on pre-training distributions drawn from various natural and synthetic data sources, primarily using the iWildCam-WILDS distribution shift as a test for robustness.

Web-scale search systems learn an encoder to embed a given query which is then hooked into an approximate nearest neighbor search (ANNS) pipeline to retrieve similar data points. To accurately capture tail queries and data points, learned representations typically are _rigid, high-dimensional_ vectors that are generally used as-is in the entire ANNS pipeline and can lead to computationally expensive retrieval. In this paper, we argue that instead of rigid representations, different stages of ANNS can leverage _adaptive representations_ of varying capacities to achieve significantly better accuracy-compute trade-offs, i.e., stages of ANNS that can get away with more approximate computation should use a lower-capacity representation of the same data point. To this end, we introduce AdANNS, a novel ANNS design framework that explicitly leverages the flexibility of Matryoshka Representations. We demonstrate state-of-the-art accuracy-compute trade-offs using novel AdANNS-based key ANNS building blocks like search data structures (AdANNS-IVF) and quantization (AdANNS-OPQ). For example on ImageNet retrieval, AdANNS-IVF is up to $\mathbf{1.5}$% more accurate than the rigid representations-based IVF at the same compute budget; and matches accuracy while being up to $\mathbf{90}\times$ faster in _wall-clock time_. For Natural Questions, $32$-byte AdANNS-OPQ matches the accuracy of the $64$-byte OPQ baseline constructed using rigid representations -- _same accuracy at half the cost!_ We further show that the gains from AdANNS translate to modern-day composite ANNS indices that combine search structures and quantization. Finally, we demonstrate that AdANNS can enable inference-time adaptivity for compute-aware search on ANNS indices built non-adaptively on matryoshka representations. Code is open-sourced at https://github.com/RAIVNLab/AdANNS.

We propose Neural Priming, a technique for adapting large pretrained models to distribution shifts and downstream tasks given few or no labeled examples. Presented with class names or unlabeled test samples, Neural Priming enables the model to recall and conditions its parameters on relevant data seen throughout pretraining, thereby priming it for the test distribution. Neural Priming can be performed at test time in even for pretraining datasets as large as LAION-2B. Performing lightweight updates on the recalled data significantly improves accuracy across a variety of distribution shift and transfer learning benchmarks. Concretely, in the zero-shot setting, we see a 2.45% improvement in accuracy on ImageNet and 3.81% accuracy improvement on average across standard transfer learning benchmarks. Further, using our test time inference scheme, we see a 1.41% accuracy improvement on ImageNetV2. These results demonstrate the effectiveness of Neural Priming in addressing the common challenge of limited labeled data and changing distributions. Code and models are open-sourced at [https://www.github.com/RAIVNLab/neural-priming](https://www.github.com/RAIVNLab/neural-priming).

We introduce new methods for 1) accelerating and 2) stabilizing training for large language-vision models. 1) For acceleration, we introduce SwitchBack, a linear layer for int8 quantized training which provides a speed-up of 13-25% while matching the performance of bfloat16 training within 0.1 percentage points for the 1B parameter CLIP ViT-Huge---the largest int8 training to date. Our main focus is int8 as GPU support for float8 is rare, though we also analyze float8 training through simulation. While SwitchBack proves effective for float8, we show that standard techniques are also successful if the network is trained and initialized so that large feature magnitudes are discouraged, which we accomplish via layer-scale initialized with zeros. 2) For stability, we analyze loss spikes and find they consistently occur 1-8 iterations after the squared gradients become under-estimated by their AdamW second moment estimator. As a result, we recommend an AdamW-Adafactor hybrid which avoids loss spikes when training a CLIP ViT-Huge model and outperforms gradient clipping at the scales we test.

In visual retrieval systems, updating the embedding model requires recomputing features for every piece of data. This expensive process is referred to as backfilling. Re-cently, the idea of backward compatible training (BCT) was proposed. To avoid the cost of backfilling, BCT modifies training of the new model to make its representations com-patible with those of the old model. However, BCT can sig-nificantly hinder the performance of the new model. In this work, we propose a new learning paradigm for representation learning: forward compatible training (FCT). In Fct, when the old model is trained, we also prepare for a future unknown version of the model. We propose learning side- information, an auxiliary feature for each sample which fa-cilitates future updates of the model. To develop a powerful and flexible frameworkfor model compatibility, we combine side-information with a forward transformation from old to newembeddings. Training of the new model is not modified, hence, its accuracy is not degraded. We demonstrate sig-nificant retrieval accuracy improvement compared to BCT for various datasets: ImageNet-1k (+18.1%), Places-365 (+5.4%), and VGG-Face2 (+8.3%). Fctobtains model compatibility when the new and old models are trained across different datasets, losses, and architectures.11Code available at https://github.com/apple/ml-fct.

Large pre-trained models such as CLIP or ALIGN offer consistent accuracy across a range of data distributions when performing zero-shot inference (i.e., without fine-tuning on a specific dataset). Although existing fine-tuning methods substantially improve accuracy on a given target distribution, they often reduce robustness to distribution shifts. We address this tension by introducing a simple and effective method for improving robustness while fine-tuning: ensembling the weights of the zero-shot and fine-tuned models (WiSE-FT). Compared to standard fine-tuning, WiSE-FT provides large accuracy improvements under distribution shift, while preserving high accuracy on the target distribution. On ImageNet and five derived distribution shifts, WiSE-FT improves accuracy under distribution shift by 4 to 6 percentage points (pp) over prior work while increasing ImageNet accuracy by 1.6 pp. WiSE-FT achieves similarly large robustness gains (2 to 23 pp) on a diverse set of six further distribution shifts, and accuracy gains of 0.8 to 3.3 pp compared to standard fine-tuning on commonly used transfer learning datasets. These improvements come at no additional computational cost during fine-tuning or inference.

As humans, we navigate a multimodal world, building a holistic understanding from all our senses. We introduce @MERLOT RESERVE, a model that represents videos jointly over time - through a new training objective that learns from audio, subtitles, and video frames. Given a video, we replace snippets of text and audio with a MASK token; the model learns by choosing the correct masked-out snippet. Our objective learns faster than alternatives, and performs well at scale: we pretrain on 20 million YouTube videos. Empirical results show that @MERLOT RESERVE learns strong multimodal representations. When finetuned, it sets state-of-the-art on Visual Commonsense Reasoning (VCR), TVQA, and Kinetics-600; outperforming prior work by 5%, 7%, and 1.5% respectively. Ablations show that these tasks benefit from audio pretraining - even VCR, a QA task centered around images (without sound). Moreover, our objective enables out-of-the-box prediction, revealing strong multimodal commonsense understanding. In a fully zero-shot setting, our model obtains competitive results on four video tasks, even outperforming supervised approaches on the recently proposed Situated Reasoning (STAR) benchmark. We analyze why audio enables better vision-language representations, suggesting significant opportunities for future research. We conclude by discussing ethical and societal implications of multimodal pretraining.

The conventional recipe for maximizing model accuracy is to (1) train multiple models with various hyperparameters and (2) pick the individual model which performs best on a held-out validation set, discarding the remainder. In this paper, we revisit the second step of this procedure in the context of fine-tuning large pre-trained models, where fine-tuned models often appear to lie in a single low error basin. We show that averaging the weights of multiple models fine-tuned with different hyperparameter configurations often improves accuracy and robustness. Unlike a conventional ensemble, we may average many models without incurring any additional inference or memory costs—we call the results “model soups.” When fine-tuning large pre-trained models such as CLIP, ALIGN, and a ViT-G pre-trained on JFT, our soup recipe provides significant improvements over the best model in a hyperparameter sweep on ImageNet. The resulting ViT-G model, which attains 90.94% top-1 accuracy on ImageNet, achieved a new state of the art. Furthermore, we show that the model soup approach extends to multiple image classification and natural language processing tasks, improves out-of-distribution performance, and improves zero-shot performance on new downstream tasks. Finally, we analytically relate the performance similarity of weight-averaging and logit-ensembling to flatness of the loss and confidence of the predictions, and validate this relation empirically. Code is available at https://github.com/mlfoundations/model-soups.

We present a computationally efficient method for compressing a trained neural network without using real data. We break the problem of data-free network compression into independent layer-wise compressions. We show how to efficiently generate layer-wise training data using only a pretrained network. We use this data to perform independent layer-wise compressions on the pretrained network. We also show how to precondition the network to improve the accuracy of our layer-wise compression method. We present results for layer-wise compression using quantization and pruning. When quantizing, we compress with higher accuracy than related works while using orders of magnitude less compute. When compressing MobileNetV2 and evaluating on ImageNet, our method outperforms existing methods for quantization at all bit-widths, achieving a +0.34% improvement in 8-bit quantization, and a stronger improvement at lower bit-widths (up to a +28.50% improvement at 5 bits). When pruning, we outperform baselines of a similar compute envelope, achieving 1.5 times the sparsity rate at the same accuracy. We also show how to combine our efficient method with high-compute generative methods to improve upon their results.

Jae Sung Park, Sheng Shen, Ali Farhadi, Trevor Darrell, Yejin Choi, Anna Rohrbach. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Open-vocabulary models like CLIP achieve high accuracy across many image classification tasks. However, there are still settings where their zero-shot performance is far from optimal. We study model patching, where the goal is to improve accuracy on specific tasks without degrading accuracy on tasks where performance is already adequate. Towards this goal, we introduce PAINT, a patching method that uses interpolations between the weights of a model before fine-tuning and the weights after fine-tuning on a task to be patched. On nine tasks where zero-shot CLIP performs poorly, PAINT increases accuracy by 15 to 60 percentage points while preserving accuracy on ImageNet within one percentage point of the zero-shot model. PAINT also allows a single model to be patched on multiple tasks and improves with model scale. Furthermore, we identify cases of broad transfer, where patching on one task increases accuracy on other tasks even when the tasks have disjoint classes. Finally, we investigate applications beyond common benchmarks such as counting or reducing the impact of typographic attacks on CLIP. Our findings demonstrate that it is possible to expand the set of tasks on which open-vocabulary models achieve high accuracy without re-training them from scratch.

Learned representations are a central component in modern ML systems, serving a multitude of downstream tasks. When training such representations, it is often the case that computational and statistical constraints for each downstream task are unknown. In this context rigid, fixed capacity representations can be either over or under-accommodating to the task at hand. This leads us to ask: can we design a flexible representation that can adapt to multiple downstream tasks with varying computational resources? Our main contribution is Matryoshka Representation Learning (MRL) which encodes information at different granularities and allows a single embedding to adapt to the computational constraints of downstream tasks. MRL minimally modifies existing representation learning pipelines and imposes no additional cost during inference and deployment. MRL learns coarse-to-fine representations that are at least as accurate and rich as independently trained low-dimensional representations. The flexibility within the learned Matryoshka Representations offer: (a) up to $\mathbf{14}\times$ smaller embedding size for ImageNet-1K classification at the same level of accuracy; (b) up to $\mathbf{14}\times$ real-world speed-ups for large-scale retrieval on ImageNet-1K and 4K; and (c) up to $\mathbf{2}\%$ accuracy improvements for long-tail few-shot classification, all while being as robust as the original representations. Finally, we show that MRL extends seamlessly to web-scale datasets (ImageNet, JFT) across various modalities -- vision (ViT, ResNet), vision + language (ALIGN) and language (BERT). MRL code and pretrained models are open-sourced at https://github.com/RAIVNLab/MRL.

Rowan Zellers, Ari Holtzman, Matthew Peters, Roozbeh Mottaghi, Aniruddha Kembhavi, Ali Farhadi, Yejin Choi. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

We have observed significant progress in visual navigation for embodied agents. A common assumption in studying visual navigation is that the environments are static; this is a limiting assumption. Intelligent navigation may involve interacting with the environment beyond just moving forward/backward and turning left/right. Sometimes, the best way to navigate is to push something out of the way. In this paper, we study the problem of interactive navigation where agents learn to change the environment to navigate more efficiently to their goals. To this end, we introduce the Neural Interaction Engine (NIE) to explicitly predict the change in the environment caused by the agent’s actions. By modeling the changes while planning, we find that agents exhibit significant improvements in their navigational capabilities. More specifically, we consider two downstream tasks in the physics-enabled, visually rich, AI2-THOR environment: (1) reaching a target while the path to the target is blocked (2) moving an object to a target location by pushing it. For both tasks, agents equipped with an NIE significantly outperform agents without the understanding of the effect of the actions indicating the benefits of our approach. The code and dataset are available at github.com/KuoHaoZeng/Interactive_Visual_Navigation.

Christopher Clark, Jordi Salvador, Dustin Schwenk, Derrick Bonafilia, Mark Yatskar, Eric Kolve, Alvaro Herrasti, Jonghyun Choi, Sachin Mehta, Sam Skjonsberg, Carissa Schoenick, Aaron Sarnat, Hannaneh Hajishirzi, Aniruddha Kembhavi, Oren Etzioni, Ali Farhadi. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.

Recent observations have advanced our understanding of the neural network optimization landscape, revealing the existence of (1) paths of high accuracy containing diverse solutions and (2) wider minima offering improved performance. Previous methods observing diverse paths require multiple training runs. In contrast we aim to leverage both property (1) and (2) with a single method and in a single training run. With a similar computational cost as training one model, we learn lines, curves, and simplexes of high-accuracy neural networks. These neural network subspaces contain diverse solutions that can be ensembled, approaching the ensemble performance of independently trained networks without the training cost. Moreover, using the subspace midpoint boosts accuracy, calibration, and robustness to label noise, outperforming Stochastic Weight Averaging.

Gabriel Ilharco, Rowan Zellers, Ali Farhadi, Hannaneh Hajishirzi. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Rowan Zellers, Ari Holtzman, Elizabeth Clark, Lianhui Qin, Ali Farhadi, Yejin Choi. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Learning binary representations of instances and classes is a classical problem with several high potential applications. In modern settings, the compression of high-dimensional neural representations to low-dimensional binary codes is a challenging task and often require large bit-codes to be accurate. In this work, we propose a novel method for $\textbf{L}$earning $\textbf{L}$ow-dimensional binary $\textbf{C}$odes $(\textbf{LLC})$ for instances as well as classes. Our method does ${\textit{not}}$ require any side-information, like annotated attributes or label meta-data, and learns extremely low-dimensional binary codes ($\approx 20$ bits for ImageNet-1K). The learnt codes are super-efficient while still ensuring $\textit{nearly optimal}$ classification accuracy for ResNet50 on ImageNet-1K. We demonstrate that the learnt codes capture intrinsically important features in the data, by discovering an intuitive taxonomy over classes. We further quantitatively measure the quality of our codes by applying it to the efficient image retrieval as well as out-of-distribution (OOD) detection problems. For ImageNet-100 retrieval problem, our learnt binary codes outperform $16$ bit HashNet using only $10$ bits and also are as accurate as $10$ dimensional real representations. Finally, our learnt binary codes can perform OOD detection, out-of-the-box, as accurately as a baseline that needs $\approx3000$ samples to tune its threshold, while we require ${\textit{none}}$. Code is open-sourced at https://github.com/RAIVNLab/LLC.

As humans, we understand events in the visual world contextually, performing multimodal reasoning across time to make inferences about the past, present, and future. We introduce MERLOT, a model that learns multimodal script knowledge by watching millions of YouTube videos with transcribed speech -- in an entirely label-free, self-supervised manner. By pretraining with a mix of both frame-level (spatial) and video-level (temporal) objectives, our model not only learns to match images to temporally corresponding words, but also to contextualize what is happening globally over time. As a result, MERLOT exhibits strong out-of-the-box representations of temporal commonsense, and achieves state-of-the-art performance on 12 different video QA datasets when finetuned. It also transfers well to the world of static images, allowing models to reason about the dynamic context behind visual scenes. On Visual Commonsense Reasoning, MERLOT~answers questions correctly with 80.6\% accuracy, outperforming state-of-the-art models of similar size by over 3\%, even those that make heavy use of auxiliary supervised data (like object bounding boxes).Ablation analyses demonstrate the complementary importance of: 1) training on videos versus static images; 2) scaling the magnitude and diversity of the pretraining video corpus; and 3) using diverse objectives that encourage full-stack multimodal reasoning, from the recognition to cognition level.

In this paper, we show that extending the butterfly operations from the FFT algorithm to a general Butterfly Transform (BFT) can be beneficial in building an efficient block structure for CNN designs. Pointwise convolutions, which we refer to as channel fusions, are the main computational bottleneck in the state-of-the-art efficient CNNs (e.g. MobileNets). We introduce a set of criterion for channel fusion, and prove that BFT yields an asymptotically optimal FLOP count with respect to these criteria. By replacing pointwise convolutions with BFT, we reduce the computational complexity of these layers from O(n2) to O(n log n) with respect to the number of channels. Our experimental evaluations show that our method results in significant accuracy gains across a wide range of network architectures, especially at low FLOP ranges. For example, BFT results in up to a 6.75% absolute Top-1 improvement for MobileNetV1, 4.4 % for ShuffleNet V2 and 5.4% for MobileNetV3 on ImageNet under a similar number of FLOPS. Notably, ShuffleNet-V2+BFT outperforms state-of-the-art architecture search methods MNasNet, FBNet and MobilenetV3 in the low FLOP regime.

Visual recognition ecosystems (e.g. ImageNet, Pascal, COCO) have undeniably played a prevailing role in the evolution of modern computer vision. We argue that interactive and embodied visual AI has reached a stage of development similar to visual recognition prior to the advent of these ecosystems. Recently, various synthetic environments have been introduced to facilitate research in embodied AI. Notwithstanding this progress, the crucial question of how well models trained in simulation generalize to reality has remained largely unanswered. The creation of a comparable ecosystem for simulation-to-real embodied AI presents many challenges: (1) the inherently interactive nature of the problem, (2) the need for tight alignments between real and simulated worlds, (3) the difficulty of replicating physical conditions for repeatable experiments, (4) and the associated cost. In this paper, we introduce RoboTHOR to democratize research in interactive and embodied visual AI. RoboTHOR offers a framework of simulated environments paired with physical counterparts to systematically explore and overcome the challenges of simulation-to-real transfer, and a platform where researchers across the globe can remotely test their embodied models in the physical world. As a first benchmark, our experiments show there exists a significant gap between the performance of models trained in simulation when they are tested in both simulations and their carefully constructed physical analogs. We hope that RoboTHOR will spur the next stage of evolution in embodied computer vision.

When we humans look at a video of human-object interaction, we can not only infer what is happening but we can even extract actionable information and imitate those interactions. On the other hand, current recognition or geometric approaches lack the physicality of action representation. In this paper, we take a step towards more physical understanding of actions. We address the problem of inferring contact points and the physical forces from videos of humans interacting with objects. One of the main challenges in tackling this problem is obtaining ground-truth labels for forces. We sidestep this problem by instead using a physics simulator for supervision. Specifically, we use a simulator to predict effects, and enforce that estimated forces must lead to same effect as depicted in the video. Our quantitative and qualitative results show that (a) we can predict meaningful forces from videos whose effects lead to accurate imitation of the motions observed, (b) by jointly optimizing for contact point and force prediction, we can improve the performance on both tasks in comparison to independent training, and (c) we can learn a representation from this model that generalizes to novel objects using few shot examples.

Training a neural network is synonymous with learning the values of the weights. In contrast, we demonstrate that randomly weighted neural networks contain subnetworks which achieve impressive performance without ever modifying the weight values. Hidden in a randomly weighted Wide ResNet-50 [32] we find a subnetwork (with random weights) that is smaller than, but matches the performance of a ResNet-34 [9] trained on ImageNet [4]. Not only do these “untrained subnetworks” exist, but we provide an algorithm to effectively find them. We empirically show that as randomly weighted neural networks with fixed weights grow wider and deeper, an “untrained subnetwork” approaches a network with learned weights in accuracy. Our code and pretrained models are available at: https://github.com/allenai/hidden-networks.

In this paper we address the problem of visual reaction: the task of interacting with dynamic environments where the changes in the environment are not necessarily caused by the agent itself. Visual reaction entails predicting the future changes in a visual environment and planning accordingly. We study the problem of visual reaction in the context of playing catch with a drone in visually rich synthetic environments. This is a challenging problem since the agent is required to learn (1) how objects with different physical properties and shapes move, (2) what sequence of actions should be taken according to the prediction, (3) how to adjust the actions based on the visual feedback from the dynamic environment (e.g., when objects bouncing off a wall), and (4) how to reason and act with an unexpected state change in a timely manner. We propose a new dataset for this task, which includes 30K throws of 20 types of objects in different directions with different forces. Our results show that our model that integrates a forecaster with a planner outperforms a set of strong baselines that are based on tracking as well as pure model-based and model-free RL baselines. The code and dataset are available at github.com/KuoHaoZeng/Visual_Reaction.

Sparsity in Deep Neural Networks (DNNs) is studied extensively with the focus of maximizing prediction accuracy given an overall parameter budget. Existing methods rely on uniform or heuristic non-uniform sparsity budgets which have sub-optimal layer-wise parameter allocation resulting in a) lower prediction accuracy or b) higher inference cost (FLOPs). This work proposes Soft Threshold Reparameterization (STR), a novel use of the soft-threshold operator on DNN weights. STR smoothly induces sparsity while learning pruning thresholds thereby obtaining a non-uniform sparsity budget. Our method achieves state-of-the-art accuracy for unstructured sparsity in CNNs (ResNet50 and MobileNetV1 on ImageNet-1K), and, additionally, learns non-uniform budgets that empirically reduce the FLOPs by up to 50%. Notably, STR boosts the accuracy over existing results by up to 10% in the ultra sparse (99%) regime and can also be used to induce low-rank (structured sparsity) in RNNs. In short, STR is a simple mechanism which learns effective sparsity budgets that contrast with popular heuristics. Code, pretrained models and sparsity budgets are at https://github.com/RAIVNLab/STR.

We present the Supermasks in Superposition (SupSup) model, capable of sequentially learning thousands of tasks without catastrophic forgetting. Our approach uses a randomly initialized, fixed base network and for each task finds a subnetwork (supermask) that achieves good performance. If task identity is given at test time, the correct subnetwork can be retrieved with minimal memory usage. If not provided, SupSup can infer the task using gradient-based optimization to find a linear superposition of learned supermasks which minimizes the output entropy. In practice we find that a single gradient step is often sufficient to identify the correct mask, even among 2500 tasks. We also showcase two promising extensions. First, SupSup models can be trained entirely without task identity information, as they may detect when they are uncertain about new data and allocate an additional supermask for the new training distribution. Finally the entire, growing set of supermasks can be stored in a constant-sized reservoir by implicitly storing them as attractors in a fixed-sized Hopfield network.

Existing open-domain question answering (QA) models are not suitable for real-time usage because they need to process several long documents on-demand for every input query, which is computationally prohibitive. In this paper, we introduce query-agnostic indexable representations of document phrases that can drastically speed up open-domain QA. In particular, our dense-sparse phrase encoding effectively captures syntactic, semantic, and lexical information of the phrases and eliminates the pipeline filtering of context documents. Leveraging strategies for optimizing training and inference time, our model can be trained and deployed even in a single 4-GPU server. Moreover, by representing phrases as pointers to their start and end tokens, our model indexes phrases in the entire English Wikipedia (up to 60 billion phrases) using under 2TB. Our experiments on SQuAD-Open show that our model is on par with or more accurate than previous models with 6000x reduced computational cost, which translates into at least 68x faster end-to-end inference benchmark on CPUs. Code and demo are available at nlp.cs.washington.edu/denspi

Recent work by Zellers et al. (2018) introduced a new task of commonsense natural language inference: given an event description such as "A woman sits at a piano," a machine must select the most likely followup: "She sets her fingers on the keys."With the introduction of BERT (Devlin et al., 2018), near human-level performance was reached.Does this mean that machines can perform human level commonsense inference?In this paper, we show that commonsense inference still proves di cult for even stateof-the-art models, by presenting HellaSwag, a new challenge dataset.Though its questions are trivial for humans (°95% accuracy), state-of-the-art models struggle ( †48%).We achieve this via Adversarial Filtering (AF), a data collection paradigm wherein a series of discriminators iteratively select an adversarial set of machine-generated wrong answers.AF proves to be surprisingly robust.The key insight is to scale up the length and complexity of the dataset examples towards a critical 'Goldilocks' zone wherein generated text is ridiculous to humans, yet often misclassified by state-of-the-art models.Our construction of HellaSwag, and its resulting di culty, sheds light on the inner workings of deep pretrained models.More broadly, it suggests a new path forward for NLP research, in which benchmarks co-evolve with the evolving state-of-the-art in an adversarial way, so as to present ever-harder challenges.

Collaboration is a necessary skill to perform tasks that are beyond one agent's capabilities. Addressed extensively in both conventional and modern AI, multi-agent collaboration has often been studied in the context of simple grid worlds. We argue that there are inherently visual aspects to collaboration which should be studied in visually rich environments. A key element in collaboration is communication that can be either explicit, through messages, or implicit, through perception of the other agents and the visual world. Learning to collaborate in a visual environment entails learning (1) to perform the task, (2) when and what to communicate, and (3) how to act based on these communications and the perception of the visual world. In this paper we study the problem of learning to collaborate directly from pixels in AI2-THOR and demonstrate the benefits of explicit and implicit modes of communication to perform visual tasks. Refer to our project page for more details: https://prior.allenai.org/projects/two-body-problem.

Visual Question Answering (VQA) in its ideal form lets us study reasoning in the joint space of vision and language and serves as a proxy for the AI task of scene understanding. However, most VQA benchmarks to date are focused on questions such as simple counting, visual attributes, and object detection that do not require reasoning or knowledge beyond what is in the image. In this paper, we address the task of knowledge-based visual question answering and provide a benchmark, called OK-VQA, where the image content is not sufficient to answer the questions, encouraging methods that rely on external knowledge resources. Our new dataset includes more than 14,000 questions that require external knowledge to answer. We show that the performance of the state-of-the-art VQA models degrades drastically in this new setting. Our analysis shows that our knowledge-based VQA task is diverse, difficult, and large compared to previous knowledge-based VQA datasets. We hope that this dataset enables researchers to open up new avenues for research in this domain.

Visual relationship reasoning is a crucial yet challenging task for understanding rich interactions across visual concepts. For example, a relationship {man, open, door} involves a complex relation {open} between concrete entities {man, door}. While much of the existing work has studied this problem in the context of still images, understanding visual relationships in videos has received limited attention. Due to their temporal nature, videos enable us to model and reason about a more comprehensive set of visual relationships, such as those requiring multiple (temporal) observations (e.g., {man, lift up, box} vs. {man, put down, box}), as well as relationships that are often correlated through time (e.g., {woman, pay, money} followed by {woman, buy, coffee}). In this paper, we construct a Conditional Random Field on a fully-connected spatiotemporal graph that exploits the statistical dependency between relational entities spatially and temporally. We introduce a novel gated energy function parametrization that learns adaptive relations conditioned on visual observations. Our model optimization is computationally efficient, and its space computation complexity is significantly amortized through our proposed parameterization. Experimental results on benchmark video datasets (ImageNet Video and Charades) demonstrate state-of-the-art performance across three standard relationship reasoning tasks: Detection, Tagging, and Recognition.

Scale variation has been a challenge from traditional to modern approaches in computer vision. Most solutions to scale issues have a similar theme: a set of intuitive and manually designed policies that are generic and fixed (e.g. SIFT or feature pyramid). We argue that the scaling policy should be learned from data. In this paper, we introduce Elastic, a simple, efficient and yet very effective approach to learn a dynamic scale policy from data. We formulate the scaling policy as a non-linear function inside the network's structure that (a) is learned from data, (b) is instance specific, (c) does not add extra computation, and (d) can be applied on any network architecture. We applied Elastic to several state-of-the-art network architectures and showed consistent improvement without extra (sometimes even lower) computation on ImageNet classification, MSCOCO multi-label classification, and PASCAL VOC semantic segmentation. Our results show major improvement for images with scale challenges. Our code is available here: https://github.com/allenai/elastic.

Learning is an inherently continuous phenomenon. When humans learn a new task there is no explicit distinction between training and inference. As we learn a task, we keep learning about it while performing the task. What we learn and how we learn it varies during different stages of learning. Learning how to learn and adapt is a key property that enables us to generalize effortlessly to new settings. This is in contrast with conventional settings in machine learning where a trained model is frozen during inference. In this paper we study the problem of learning to learn at both training and test time in the context of visual navigation. A fundamental challenge in navigation is generalization to unseen scenes. In this paper we propose a self-adaptive visual navigation method (SAVN) which learns to adapt to new environments without any explicit supervision. Our solution is a meta-reinforcement learning approach where an agent learns a self-supervised interaction loss that encourages effective navigation. Our experiments, performed in the AI2-THOR framework, show major improvements in both success rate and SPL for visual navigation in novel scenes. Our code and data are available at: https://github.com/allenai/savn.

Visual understanding goes well beyond object recognition. With one glance at an image, we can effortlessly imagine the world beyond the pixels: for instance, we can infer people's actions, goals, and mental states. While this task is easy for humans, it is tremendously difficult for today's vision systems, requiring higher-order cognition and commonsense reasoning about the world. We formalize this task as Visual Commonsense Reasoning. Given a challenging question about an image, a machine must answer correctly and then provide a rationale justifying its answer. Next, we introduce a new dataset, VCR, consisting of 290k multiple choice QA problems derived from 110k movie scenes. The key recipe for generating non-trivial and high-quality problems at scale is Adversarial Matching, a new approach to transform rich annotations into multiple choice questions with minimal bias. Experimental results show that while humans find VCR easy (over 90% accuracy), state-of-the-art vision models struggle (~45%). To move towards cognition-level understanding, we present a new reasoning engine, Recognition to Cognition Networks (R2C), that models the necessary layered inferences for grounding, contextualization, and reasoning. R2C helps narrow the gap between humans and machines (~65%); still, the challenge is far from solved, and we provide analysis that suggests avenues for future work.

The success of neural networks has driven a shift in focus from feature engineering to architecture engineering. However, successful networks today are constructed using a small and manually defined set of building blocks. Even in methods of neural architecture search (NAS) the network connectivity patterns are largely constrained. In this work we propose a method for discovering neural wirings. We relax the typical notion of layers and instead enable channels to form connections independent of each other. This allows for a much larger space of possible networks. The wiring of our network is not fixed during training -- as we learn the network parameters we also learn the structure itself. Our experiments demonstrate that our learned connectivity outperforms hand engineered and randomly wired networks. By learning the connectivity of MobileNetV1we boost the ImageNet accuracy by 10% at ~41M FLOPs. Moreover, we show that our method generalizes to recurrent and continuous time networks. Our work may also be regarded as unifying core aspects of the neural architecture search problem with sparse neural network learning. As NAS becomes more fine grained, finding a good architecture is akin to finding a sparse subnetwork of the complete graph. Accordingly, DNW provides an effective mechanism for discovering sparse subnetworks of predefined architectures in a single training run. Though we only ever use a small percentage of the weights during the forward pass, we still play the so-called initialization lottery with a combinatorial number of subnetworks. Code and pretrained models are available at https://github.com/allenai/dnw while additional visualizations may be found at https://mitchellnw.github.io/blog/2019/dnw/.

Recent progress in natural language generation has raised dual-use concerns. While applications like summarization and translation are positive, the underlying technology also might enable adversaries to generate neural fake news: targeted propaganda that closely mimics the style of real news. Modern computer security relies on careful threat modeling: identifying potential threats and vulnerabilities from an adversary's point of view, and exploring potential mitigations to these threats. Likewise, developing robust defenses against neural fake news requires us first to carefully investigate and characterize the risks of these models. We thus present a model for controllable text generation called Grover. Given a headline like 'Link Found Between Vaccines and Autism,' Grover can generate the rest of the article; humans find these generations to be more trustworthy than human-written disinformation. Developing robust verification techniques against generators like Grover is critical. We find that best current discriminators can classify neural fake news from real, human-written, news with 73% accuracy, assuming access to a moderate level of training data. Counterintuitively, the best defense against Grover turns out to be Grover itself, with 92% accuracy, demonstrating the importance of public release of strong generators. We investigate these results further, showing that exposure bias -- and sampling strategies that alleviate its effects -- both leave artifacts that similar discriminators can pick up on. We conclude by discussing ethical issues regarding the technology, and plan to release Grover publicly, helping pave the way for better detection of neural fake news.

Developing useful interfaces between brains and machines is a grand challenge of neuroengineering. An effective interface has the capacity to not only interpret neural signals, but predict the intentions of the human to perform an action in the near future; prediction is made even more challenging outside well-controlled laboratory experiments. This paper describes our approach to detect and to predict natural human arm movements in the future, a key challenge in brain computer interfacing that has never before been attempted. We introduce the novel Annotated Joints in Long-term ECoG (AJILE) dataset; AJILE includes automatically annotated poses of 7 upper body joints for four human subjects over 670 total hours (more than 72 million frames), along with the corresponding simultaneously acquired intracranial neural recordings. The size and scope of AJILE greatly exceeds all previous datasets with movements and electrocorticography (ECoG), making it possible to take a deep learning approach to movement prediction. We propose a multimodal model that combines deep convolutional neural networks (CNN) with long short-term memory (LSTM) blocks, leveraging both ECoG and video modalities. We demonstrate that our models are able to detect movements and predict future movements up to 800 msec before movement initiation. Further, our multimodal movement prediction models exhibit resilience to simulated ablation of input neural signals. We believe a multimodal approach to natural neural decoding that takes context into account is critical in advancing bioelectronic technologies and human neuroscience.

Diagrams often depict complex phenomena and serve as a good test bed for visual and textual reasoning. However, understanding diagrams using natural image understanding approaches requires large training datasets of diagrams, which are very hard to obtain. Instead, this can be addressed as a matching problem either between labeled diagrams, images or both. This problem is very challenging since the absence of significant color and texture renders local cues ambiguous and requires global reasoning. We consider the problem of one-shot part labeling: labeling multiple parts of an object in a target image given only a single source image of that category. For this set-to-set matching problem, we introduce the Structured Set Matching Network (SSMN), a structured prediction model that incorporates convolutional neural networks. The SSMN is trained using global normalization to maximize local match scores between corresponding elements and a global consistency score among all matched elements, while also enforcing a matching constraint between the two sets. The SSMN significantly outperforms several strong baselines on three label transfer scenarios: diagram-to-diagram, evaluated on a new diagram dataset of over 200 categories; image-to-image, evaluated on a dataset built on top of the Pascal Part Dataset; and image-to-diagram, evaluated on transferring labels across these datasets.

We study the task of directly modelling a visually intelligent agent. Computer vision typically focuses on solving various subtasks related to visual intelligence. We depart from this standard approach to computer vision; instead we directly model a visually intelligent agent. Our model takes visual information as input and directly predicts the actions of the agent. Toward this end we introduce DECADE, a dataset of ego-centric videos from a dog's perspective as well as her corresponding movements. Using this data we model how the dog acts and how the dog plans her movements. We show under a variety of metrics that given just visual input we can successfully model this intelligent agent in many situations. Moreover, the representation learned by our model encodes distinct information compared to representations trained on image classification, and our learned representation can generalize to other domains. In particular, we show strong results on the task of walkable surface estimation and scene classification by using this dog modelling task as representation learning.

Objects often occlude each other in scenes; Inferring their appearance beyond their visible parts plays an important role in scene understanding, depth estimation, object interaction and manipulation. In this paper, we study the challenging problem of completing the appearance of occluded objects. Doing so requires knowing which pixels to paint (segmenting the invisible parts of objects) and what color to paint them (generating the invisible parts). Our proposed novel solution, SeGAN, jointly optimizes for both segmentation and generation of the invisible parts of objects. Our experimental results show that: (a) SeGAN can learn to generate the appearance of the occluded parts of objects; (b) SeGAN outperforms state-of-the-art segmentation baselines for the invisible parts of objects; (c) trained on synthetic photo realistic images, SeGAN can reliably segment natural images; (d) by reasoning about occluder-occludee relations, our method can infer depth layering.

We introduce Interactive Question Answering (IQA), the task of answering questions that require an autonomous agent to interact with a dynamic visual environment. IQA presents the agent with a scene and a question, like: "Are there any apples in the fridge?" The agent must navigate around the scene, acquire visual understanding of scene elements, interact with objects (e.g. open refrigerators) and plan for a series of actions conditioned on the question. Popular reinforcement learning approaches with a single controller perform poorly on IQA owing to the large and diverse state space. We propose the Hierarchical Interactive Memory Network (HIMN), consisting of a factorized set of controllers, allowing the system to operate at multiple levels of temporal abstraction. To evaluate HIMN, we introduce IQUAD V1, a new dataset built upon AI2-THOR [35], a simulated photo-realistic environment of configurable indoor scenes with interactive objects. IQUAD V1 has 75,000 questions, each paired with a unique scene configuration. Our experiments show that our proposed model outperforms popular single controller based methods on IQUAD V1. For sample questions and results, please view our video: https://youtu.be/pXd3C-1jr98.

Several theories in cognitive neuroscience suggest that when people interact with the world, or simulate interactions, they do so from a first-person egocentric perspective, and seamlessly transfer knowledge between third-person (observer) and first-person (actor). Despite this, learning such models for human action recognition has not been achievable due to the lack of data. This paper takes a step in this direction, with the introduction of Charades-Ego, a large-scale dataset of paired first-person and third-person videos, involving 112 people, with 4000 paired videos. This enables learning the link between the two, actor and observer perspectives. Thereby, we address one of the biggest bottlenecks facing egocentric vision research, providing a link from first-person to the abundant third-person data on the web. We use this data to learn a joint representation of first and third-person videos, with only weak supervision, and show its effectiveness for transferring knowledge from the third-person to the first-person domain.

We formalize a new modular variant of current question answering tasks by enforcing complete independence of the document encoder from the question encoder.This formulation addresses a key challenge in machine comprehension by requiring a standalone representation of the document discourse.It additionally leads to a significant scalability advantage since the encoding of the answer candidate phrases in the document can be pre-computed and indexed offline for efficient retrieval.We experiment with baseline models for the new task, which achieve a reasonable accuracy but significantly underperform unconstrained QA models.We invite the QA research community to engage in Phrase-Indexed Question Answering (PIQA, pika) for closing the gap.The leaderboard is at: nlp.

Existing online 3D shape repositories contain thousands of 3D models but lack photorealistic appearance. We present an approach to automatically assign high-quality, realistic appearance models to large scale 3D shape collections. The key idea is to jointly leverage three types of online data - shape collections, material collections, and photo collections, using the photos as reference to guide assignment of materials to shapes. By generating a large number of synthetic renderings, we train a convolutional neural network to classify materials in real photos, and employ 3D-2D alignment techniques to transfer materials to different parts of each shape model. Our system produces photorealistic, relightable, 3D shapes (PhotoShapes).

Porting state of the art deep learning algorithms to resource constrained compute platforms (e.g. VR, AR, wearables) is extremely challenging. We propose a fast, compact, and accurate model for convolutional neural networks that enables efficient learning and inference. We introduce LCNN, a lookup-based convolutional neural network that encodes convolutions by few lookups to a dictionary that is trained to cover the space of weights in CNNs. Training LCNN involves jointly learning a dictionary and a small set of linear combinations. The size of the dictionary naturally traces a spectrum of trade-offs between efficiency and accuracy. Our experimental results on ImageNet challenge show that LCNN can offer 3.2x speedup while achieving 55.1% top-1 accuracy using AlexNet architecture. Our fastest LCNN offers 37.6x speed up over AlexNet while maintaining 44.3% top-1 accuracy. LCNN not only offers dramatic speed ups at inference, but it also enables efficient training. In this paper, we show the benefits of LCNN in few-shot learning and few-iteration learning, two crucial aspects of on-device training of deep learning models.

We introduce the task of Multi-Modal Machine Comprehension (M3C), which aims at answering multimodal questions given a context of text, diagrams and images. We present the Textbook Question Answering (TQA) dataset that includes 1,076 lessons and 26,260 multi-modal questions, taken from middle school science curricula. Our analysis shows that a significant portion of questions require complex parsing of the text and the diagrams and reasoning, indicating that our dataset is more complex compared to previous machine comprehension and visual question answering datasets. We extend state-of-the-art methods for textual machine comprehension and visual question answering to the TQA dataset. Our experiments show that these models do not perform well on TQA. The presented dataset opens new challenges for research in question answering and reasoning across multiple modalities.

We introduce YOLO9000, a state-of-the-art, real-time object detection system that can detect over 9000 object categories. First we propose various improvements to the YOLO detection method, both novel and drawn from prior work. The improved model, YOLOv2, is state-of-the-art on standard detection tasks like PASCAL VOC and COCO. Using a novel, multi-scale training method the same YOLOv2 model can run at varying sizes, offering an easy tradeoff between speed and accuracy. At 67 FPS, YOLOv2 gets 76.8 mAP on VOC 2007. At 40 FPS, YOLOv2 gets 78.6 mAP, outperforming state-of-the-art methods like Faster RCNN with ResNet and SSD while still running significantly faster. Finally we propose a method to jointly train on object detection and classification. Using this method we train YOLO9000 simultaneously on the COCO detection dataset and the ImageNet classification dataset. Our joint training allows YOLO9000 to predict detections for object classes that dont have labelled detection data. We validate our approach on the ImageNet detection task. YOLO9000 gets 19.7 mAP on the ImageNet detection validation set despite only having detection data for 44 of the 200 classes. On the 156 classes not in COCO, YOLO9000 gets 16.0 mAP. YOLO9000 predicts detections for more than 9000 different object categories, all in real-time.

Actions are more than just movements and trajectories: we cook to eat and we hold a cup to drink from it. A thorough understanding of videos requires going beyond appearance modeling and necessitates reasoning about the sequence of activities, as well as the higher-level constructs such as intentions. But how do we model and reason about these? We propose a fully-connected temporal CRF model for reasoning over various aspects of activities that includes objects, actions, and intentions, where the potentials are predicted by a deep network. End-to-end training of such structured models is a challenging endeavor: For inference and learning we need to construct mini-batches consisting of whole videos, leading to mini-batches with only a few videos. This causes high-correlation between data points leading to breakdown of the backprop algorithm. To address this challenge, we present an asynchronous variational inference method that allows efficient end-to-end training. Our method achieves a classification mAP of 22.4% on the Charades [42] benchmark, outperforming the state-of-the-art (17.2% mAP), and offers equal gains on the task of temporal localization.

Semantic sparsity is a common challenge in structured visual classification problems, when the output space is complex, the vast majority of the possible predictions are rarely, if ever, seen in the training set. This paper studies semantic sparsity in situation recognition, the task of producing structured summaries of what is happening in images, including activities, objects and the roles objects play within the activity. For this problem, we find empirically that most substructures required for prediction are rare, and current state-of-the-art model performance dramatically decreases if even one such rare substructure exists in the target output. We avoid many such errors by (1) introducing a novel tensor composition function that learns to share examples across substructures more effectively and (2) semantically augmenting our training data with automatically gathered examples of rarely observed outputs using web data. When integrated within a complete CRF-based structured prediction model, the tensor-based approach outperforms existing state of the art by a relative improvement of 2.11% and 4.40% on top-5 verb and noun-role accuracy, respectively. Adding 5 million images with our semantic augmentation techniques gives further relative improvements of 6.23% and 9.57% on top-5 verb and noun-role accuracy.

Humans have rich understanding of liquid containers and their contents; for example, we can effortlessly pour water from a pitcher to a cup. Doing so requires estimating the volume of the cup, approximating the amount of water in the pitcher, and predicting the behavior of water when we tilt the pitcher. Very little attention in computer vision has been made to liquids and their containers. In this paper, we study liquid containers and their contents, and propose methods to estimate the volume of containers, approximate the amount of liquid in them, and perform comparative volume estimations all from a single RGB image. Furthermore, we show the results of the proposed model for predicting the behavior of liquids inside containers when one tilts the containers. We also introduce a new dataset of Containers Of liQuid contEnt (COQE) that contains more than 5,000 images of 10,000 liquid containers in context labelled with volume, amount of content, bounding box annotation, and corresponding similar 3D CAD models.

A crucial capability of real-world intelligent agents is their ability to plan a sequence of actions to achieve their goals in the visual world. In this work, we address the problem of visual semantic planning: the task of predicting a sequence of actions from visual observations that transform a dynamic environment from an initial state to a goal state. Doing so entails knowledge about objects and their affordances, as well as actions and their preconditions and effects. We propose learning these through interacting with a visual and dynamic environment. Our proposed solution involves bootstrapping reinforcement learning with imitation learning. To ensure cross task generalization, we develop a deep predictive model based on successor representations. Our experimental results show near optimal results across a wide range of tasks in the challenging THOR environment.

Two less addressed issues of deep reinforcement learning are (1) lack of generalization capability to new goals, and (2) data inefficiency, i.e., the model requires several (and often costly) episodes of trial and error to converge, which makes it impractical to be applied to real-world scenarios. In this paper, we address these two issues and apply our model to target-driven visual navigation. To address the first issue, we propose an actor-critic model whose policy is a function of the goal as well as the current state, which allows better generalization. To address the second issue, we propose the AI2-THOR framework, which provides an environment with high-quality 3D scenes and a physics engine. Our framework enables agents to take actions and interact with objects. Hence, we can collect a huge number of training samples efficiently. We show that our proposed method (1) converges faster than the state-of-the-art deep reinforcement learning methods, (2) generalizes across targets and scenes, (3) generalizes to a real robot scenario with a small amount of fine-tuning (although the model is trained in simulation), (4) is end-to-end trainable and does not need feature engineering, feature matching between frames or 3D reconstruction of the environment.

We present a novel method to summarize unconstrained videos using salient montages (i.e., a "melange" of frames in the video as shown in Fig. 1, by finding "montageable moments" and identifying the salient people and actions to depict in each montage. Our method aims at addressing the increasing need for generating concise visualizations from the large number of videos being captured from portable devices. Our main contributions are (1) the process of finding salient people and moments to form a montage, and (2) the application of this method to videos taken "in the wild" where the camera moves freely. As such, we demonstrate results on head-mounted cameras, where the camera moves constantly, as well as on videos downloaded from YouTube. In our experiments, we show that our method can reliably detect and track humans under significant action and camera motion. Moreover, the predicted salient people are more accurate than results from state-of-the-art video salieny method [1] . Finally, we demonstrate that a novel "montageability" score can be used to retrieve results with relatively high precision which allows us to present high quality montages to users.We present a novel method to summarize unconstrained videos using salient montages (i.e., a "melange" of frames in the video as shown in Fig. 1, by finding "montageable moments" and identifying the salient people and actions to depict in each montage. Our method aims at addressing the increasing need for generating concise visualizations from the large number of videos being captured from portable devices. Our main contributions are (1) the process of finding salient people and moments to form a montage, and (2) the application of this method to videos taken "in the wild" where the camera moves freely. As such, we demonstrate results on head-mounted cameras, where the camera moves constantly, as well as on videos downloaded from YouTube. In our experiments, we show that our method can reliably detect and track humans under significant action and camera motion. Moreover, the predicted salient people are more accurate than results from state-of-the-art video salieny method [1] . Finally, we demonstrate that a novel "montageability" score can be used to retrieve results with relatively high precision which allows us to present high quality montages to users.

Due to the unprecedented growth of unedited videos, finding highlights relevant to a text query in a set of unedited videos has become increasingly important. We refer this task as semantic highlight retrieval and propose a query-dependent video representation for retrieving a variety of highlights. Our method consists of two parts: 1) "viralets", a mid-level representation bridging between semantic [Fig. 1(a)] and visual [Fig. 1(c)] spaces and 2) a novel Semantic-MODulation (SMOD) procedure to make viralets query-dependent (referred to as SMOD viralets). Given SMOD viralets, we train a single highlight ranker to predict the highlightness of clips with respect to a variety of queries (two examples in Fig. 1), whereas existing approaches can be applied only in a few predefined domains. Other than semantic highlight retrieval, viralets can also be used to associate relevant terms to each video. We utilize this property and propose a simple term prediction method based on nearest neighbor search. To conduct experiments, we collect a viral video dataset including users' comments, highlights, and/or original videos. Among a testing database with 1189 clips (13% highlights and 87% non-highlights), our highlight ranker achieves 41.2% recall at top-10 retrieved clips. It is significantly higher than the state-of-the-art domain-specific highlight ranker and its extension. Similarly, our method also outperforms all baseline methods on the publicly available video highlight dataset. Finally, our simple term prediction method utilizing viralets outperforms the state-of-the-art matrix factorization method (adapted from Kalayeh et al.). Viral videos refer to popular online videos. We focus on user-generated viral videos, which typically contain short highlight marked by users.

Human vision greatly benefits from the information about sizes of objects. The role of size in several visual reasoning tasks has been thoroughly explored in human perception and cognition. However, the impact of the information about sizes of objects is yet to be determined in AI. We postulate that this is mainly attributed to the lack of a comprehensive repository of size information. In this paper, we introduce a method to automatically infer object sizes, leveraging visual and textual information from web. By maximizing the joint likelihood of textual and visual observations, our method learns reliable relative size estimates, with no explicit human supervision. We introduce the relative size dataset and show that our method outperforms competitive textual and visual baselines in reasoning about size comparisons.

The human visual system can spot an abnormal image, and reason about what makes it strange. This task has not received enough attention in computer vision. In this paper we study various types of atypicalities in images in a more comprehensive way than has been done before. We propose a new dataset of abnormal images showing a wide range of atypicalities. We design human subject experiments to discover a coarse taxonomy of the reasons for abnormality. Our experiments reveal three major categories of abnormality: object-centric, scene-centric, and contextual. Based on this taxonomy, we propose a comprehensive computational model that can predict all different types of abnormality in images and outperform prior arts in abnormality recognition.

In this paper, we study the challenging problem of predicting the dynamics of objects in static images. Given a query object in an image, our goal is to provide a physical understanding of the object in terms of the forces acting upon it and its long term motion as response to those forces. Direct and explicit estimation of the forces and the motion of objects from a single image is extremely challenging. We define intermediate physical abstractions called Newtonian scenarios and introduce Newtonian Neural Network (N3) that learns to map a single image to a state in a Newtonian scenario. Our evaluations show that our method can reliably predict dynamics of a query object from a single image. In addition, our approach can provide physical reasoning that supports the predicted dynamics in terms of velocity and force vectors. To spur research in this direction we compiled Visual Newtonian Dynamics (VIND) dataset that includes more than 6000 videos aligned with Newtonian scenarios represented using game engines, and more than 4500 still images with their ground truth dynamics.

With the recent progress in visual recognition, we have already started to see a surge of vision related real-world applications. These applications, unlike general scene understanding, are task oriented and require specific information from visual data. Considering the current growth in new sensory devices, feature designs, feature learning methods, and algorithms, the search in the space of features and models becomes combinatorial. In this paper, we propose a novel cost-sensitive task-oriented recognition method that is based on a combination of linguistic semantics and visual cues. Our task-oriented framework is able to generalize to unseen tasks for which there is no training data and outperforms state-of-the-art cost-based recognition baselines on our new task-based dataset.

We present YOLO, a new approach to object detection. Prior work on object detection repurposes classifiers to perform detection. Instead, we frame object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance. Our unified architecture is extremely fast. Our base YOLO model processes images in real-time at 45 frames per second. A smaller version of the network, Fast YOLO, processes an astounding 155 frames per second while still achieving double the mAP of other real-time detectors. Compared to state-of-the-art detection systems, YOLO makes more localization errors but is less likely to predict false positives on background. Finally, YOLO learns very general representations of objects. It outperforms other detection methods, including DPM and R-CNN, when generalizing from natural images to other domains like artwork.

What defines an action like "kicking ball"? We argue that the true meaning of an action lies in the change or transformation an action brings to the environment. In this paper, we propose a novel representation for actions by modeling an action as a transformation which changes the state of the environment before the action happens (precondition) to the state after the action (effect). Motivated by recent advancements of video representation using deep learning, we design a Siamese network which models the action as a transformation on a high-level feature space. We show that our model gives improvements on standard action recognition datasets including UCF101 and HMDB51. More importantly, our approach is able to generalize beyond learned action categories and shows significant performance improvement on cross-category generalization on our new ACT dataset.

This paper introduces situation recognition, the problem of producing a concise summary of the situation an image depicts including: (1) the main activity (e.g., clipping), (2) the participating actors, objects, substances, and locations (e.g., man, shears, sheep, wool, and field) and most importantly (3) the roles these participants play in the activity (e.g., the man is clipping, the shears are his tool, the wool is being clipped from the sheep, and the clipping is in a field). We use FrameNet, a verb and role lexicon developed by linguists, to define a large space of possible situations and collect a large-scale dataset containing over 500 activities, 1,700 roles, 11,000 objects, 125,000 images, and 200,000 unique situations. We also introduce structured prediction baselines and show that, in activity-centric images, situation-driven prediction of objects and activities outperforms independent object and activity recognition.

Large-scale annotated datasets allow AI systems to learn from and build upon the knowledge of the crowd. Many crowdsourcing techniques have been developed for collecting image annotations. These techniques often implicitly rely on the fact that a new input image takes a negligible amount of time to perceive. In contrast, we investigate and determine the most cost-effective way of obtaining high-quality multi-label annotations for temporal data such as videos. Watching even a short 30-second video clip requires a significant time investment from a crowd worker; thus, requesting multiple annotations following a single viewing is an important cost-saving strategy. But how many questions should we ask per video? We conclude that the optimal strategy is to ask as many questions as possible in a HIT (up to 52 binary questions after watching a 30-second video clip in our experiments).We demonstrate that while workers may not correctly answer all questions, the cost-benefit analysis nevertheless favors consensus from multiple such cheap-yet-imperfect iterations over more complex alternatives. When compared with a one-question-per-video baseline, our method is able to achieve a 10% improvement in recall (76.7% ours versus 66.7% baseline) at comparable precision (83.8% ours versus 83.0% baseline) in about half the annotation time (3.8 minutes ours compared to 7.1 minutes baseline). We demonstrate the effectiveness of our method by collecting multi-label annotations of 157 human activities on 1,815 videos.

Finding highlights relevant to a text query in unedited videos has become increasingly important due to their unprecedented growth. We refer this task as semantic highlight retrieval and propose a query-dependent video representation for retrieving a variety of highlights. Our method consist of two parts: (1) “viralets”, a mid-level representation bridging between visual and semantic spaces; (2) a novel Semantically-Modulation (SM) procedure to make viralets query-dependent (referred to as SM viralets). Given SM viralets, we train a single highlight ranker to predict the highlightness of clips with respect to a variety of queries, whereas existing approaches can be applied only in a few predefined domains. We collect a viral video dataset1including users' comments, highlights, and/or original videos. Among a database with 1189 (13% highlights) clips, our highlight ranker achieves 41.2% recall at top-10 retrieved clips. It is significantly higher than a state-of-the-art domain-specific highlight ranker and its extension. Similarly, our method also outperforms all baseline methods on the video highlight dataset.

Clustering is central to many data-driven application domains and has been studied extensively in terms of distance functions and grouping algorithms. Relatively little work has focused on learning representations for clustering. In this paper, we propose Deep Embedded Clustering (DEC), a method that simultaneously learns feature representations and cluster assignments using deep neural networks. DEC learns a mapping from the data space to a lower-dimensional feature space in which it iteratively optimizes a clustering objective. Our experimental evaluations on image and text corpora show significant improvement over state-of-the-art methods.

Obtaining common sense knowledge using current information extraction techniques is extremely challenging.In this work, we instead propose to derive simple common sense statements from fully annotated object detection corpora such as the Microsoft Common Objects in Context dataset.We show that many thousands of common sense facts can be extracted from such corpora at high quality.Furthermore, using WordNet and a novel submodular k-coverage formulation, we are able to generalize our initial set of common sense assertions to unseen objects and uncover over 400k potentially useful facts.

We present a fully automatic system for ranking domain-specific highlights in unconstrained personal videos by analyzing online edited videos. A novel latent linear ranking model is proposed to handle noisy training data harvested online. Specifically, given a targeted domain such as "surfing," our system mines the YouTube database to find pairs of raw and their corresponding edited videos. Leveraging the assumption that an edited video is more likely to contain highlights than the trimmed parts of the raw video, we obtain pair-wise ranking constraints to train our model. The learning task is challenging due to the amount of noise and variation in the mined data. Hence, a latent loss function is incorporated to mitigate the issues caused by the noise. We efficiently learn the latent model on a large number of videos (about 870 min in total) using a novel EM-like procedure. Our latent ranking model outperforms its classification counterpart and is fairly competitive compared with a fully supervised ranking system that requires labels from Amazon Mechanical Turk. We further show that a state-of-the-art audio feature mel-frequency cepstral coefficients is inferior to a state-of-the-art visual feature. By combining both audio-visual features, we obtain the best performance in dog activity, surfing, skating, and viral video domains. Finally, we show that impressive highlights can be detected without additional human supervision for seven domains (i.e., skating, surfing, skiing, gymnastics, parkour, dog activity, and viral video) in unconstrained personal videos.

In this paper we present a bottom-up method to instance-level Multiple Instance Learning (MIL) that learns to discover positive instances with globally constrained reasoning about local pairwise similarities. We discover positive instances by optimizing for a ranking such that positive (top rank) instances are highly and consistently similar to each other and dissimilar to negative instances. Our approach takes advantage of a discriminative notion of pairwise similarity coupled with a structural cue in the form of a consistency metric that measures the quality of each similarity. We learn a similarity function for every pair of instances in positive bags by how similarly they differ from instances in negative bags, the only certain labels in MIL. Our experiments demonstrate that our method consistently outperforms state-of-the-art MIL methods both at bag-level and instance-level predictions in standard benchmarks, image category recognition, and text categorization datasets.

How can we know whether a statement about our world is valid. For example, given a relationship between a pair of entities e.g., `eat(horse, hay)', how can we know whether this relationship is true or false in general. Gathering such knowledge about entities and their relationships is one of the fundamental challenges in knowledge extraction. Most previous works on knowledge extraction have focused purely on text-driven reasoning for verifying relation phrases. In this work, we introduce the problem of visual verification of relation phrases and developed a Visual Knowledge Extraction system called VisKE. Given a verb-based relation phrase between common nouns, our approach assess its validity by jointly analyzing over text and images and reasoning about the spatial consistency of the relative configurations of the entities and the relation involved. Our approach involves no explicit human supervision thereby enabling large-scale analysis. Using our approach, we have already verified over 12000 relation phrases. Our approach has been used to not only enrich existing textual knowledge bases by improving their recall, but also augment open-domain question-answer reasoning.

This paper introduces GEOS, the first automated system to solve unaltered SAT geometry questions by combining text understanding and diagram interpretation.We model the problem of understanding geometry questions as submodular optimization, and identify a formal problem description likely to be compatible with both the question text and diagram.GEOS then feeds the description to a geometric solver that attempts to determine the correct answer.In our experiments, GEOS achieves a 49% score on official SAT questions, and a score of 61% on practice questions. 1 Finally, we show that by integrating textual and visual information, GEOS boosts the accuracy of dependency and semantic parsing of the question text.

We introduce Segment-Phrase Table (SPT), a large collection of bijective associations between textual phrases and their corresponding segmentations. Leveraging recent progress in object recognition and natural language semantics, we show how we can successfully build a high-quality segment-phrase table using minimal human supervision. More importantly, we demonstrate the unique value unleashed by this rich bimodal resource, for both vision as well as natural language understanding. First, we show that fine-grained textual labels facilitate contextual reasoning that helps in satisfying semantic constraints across image segments. This feature enables us to achieve state-of-the-art segmentation results on benchmark datasets. Next, we show that the association of high-quality segmentations to textual phrases aids in richer semantic understanding and reasoning of these textual phrases. Leveraging this feature, we motivate the problem of visual entailment and visual paraphrasing, and demonstrate its utility on a large dataset.

We all have experienced forgetting habitual actions among our daily activities. For example, we probably have forgotten to turn the lights off before leaving a room or turn the stove off after cooking. In this paper, we propose a solution to the problem of issuing notifications on actions that may be missed. This involves learning about interdependencies between actions and being able to predict an ongoing action while segmenting the input video stream. In order to show a proof of concept, we collected a new egocentric dataset, in which people wear a camera while making lattes. We show promising results on the extremely challenging task of issuing correct and timely reminders. We also show that our model reliably segments the actions, while predicting the ongoing one when only a few frames from the beginning of the action are observed. The overall prediction accuracy is 46.2% when only 10 frames of an action are seen (2/3 of a sec). Moreover, the overall recognition and segmentation accuracy is shown to be 72.7% when the whole activity sequence is observed. Finally, the online prediction and segmentation accuracy is 68.3% when the prediction is made at every time step.

In this paper, we study the problem of answering visual analogy questions. These questions take the form of image A is to image B as image C is to what. Answering these questions entails discovering the mapping from image A to image B and then extending the mapping to image C and searching for the image D such that the relation from A to B holds for C to D. We pose this problem as learning an embedding that encourages pairs of analogous images with similar transformations to be close together using convolutional neural networks with a quadruple Siamese architecture. We introduce a dataset of visual analogy questions in natural images, and show first results of its kind on solving analogy questions on natural images.

We propose the problem of automated photo album creation from an unordered image collection. The problem is difficult as it involves a number of complex perceptual tasks that facilitate selection and ordering of photos to create a compelling visual narrative. To help solve this problem, we collect (and will make available) a new benchmark dataset based on Flickr images. Flickr Album Dataset and provides a variety of annotations useful for the task, including manually created albums of various lengths. We analyze the problem and provide experimental evidence, through user studies, that both selection and ordering of photos within an album is important for human observers. To capture and learn rules of album composition, we propose a discriminative structured model capable of encoding simple preferences for contextual layout of the scene (e.g., spatial layout of faces, global scene context, and presence/absence of attributes) and ordering between photos (e.g., exclusion principles or correlations). The parameters of the model are learned using a structured SVM framework. Once learned, the model allows automatic composition of photo albums from unordered and untagged collections of images. We quantitatively evaluate the results obtained using our model against manually created albums and baselines on a dataset of 63 personal photo collections from 5 different topics.

Automatically solving geometry questions is a long-standing AI problem. A geometry question typically includes a textual description accompanied by a diagram. The first step in solving geometry questions is diagram understanding, which consists of identifying visual elements in the diagram, their locations, their geometric properties, and aligning them to corresponding textual descriptions. In this paper, we present a method for diagram understanding that identifies visual elements in a diagram while maximizing agreement between textual and visual data. We show that the method's objective function is submodular; thus we are able to introduce an efficient method for diagram understanding that is close to optimal. To empirically evaluate our method, we compile a new dataset of geometry questions (textual descriptions and diagrams) and compare with baselines that utilize standard vision techniques. Our experimental evaluation shows an F1 boost of more than 17% in identifying visual elements and 25% in aligning visual elements with their textual descriptions.

Recognition is graduating from labs to real-world applications. While it is encouraging to see its potential being tapped, it brings forth a fundamental challenge to the vision researcher: scalability. How can we learn a model for any concept that exhaustively covers all its appearance variations, while requiring minimal or no human supervision for compiling the vocabulary of visual variance, gathering the training images and annotations, and learning the models? In this paper, we introduce a fully-automated approach for learning extensive models for a wide range of variations (e.g. actions, interactions, attributes and beyond) within any concept. Our approach leverages vast resources of online books to discover the vocabulary of variance, and intertwines the data collection and modeling steps to alleviate the need for explicit human supervision in training the models. Our approach organizes the visual knowledge about a concept in a convenient and useful way, enabling a variety of applications across vision and NLP. Our online system has been queried by users to learn models for several interesting concepts including breakfast, Gandhi, beautiful, etc. To date, our system has models available for over 50, 000 variations within 150 concepts, and has annotated more than 10 million images with bounding boxes.

A scene category imposes tight distributions over the kind of objects that might appear in the scene, the appearance of those objects and their layout. In this paper, we propose a method to learn scene structures that can encode three main interlacing components of a scene: the scene category, the context-specific appearance of objects, and their layout. Our experimental evaluations show that our learned scene structures outperform state-of-the-art method of Deformable Part Models in detecting objects in a scene. Our scene structure provides a level of scene understanding that is amenable to deep visual inferences. The scene structures can also generate features that can later be used for scene categorization. Using these features, we also show promising results on scene categorization.

Computer vision systems today fail frequently. They also fail abruptly without warning or explanation. Alleviating the former has been the primary focus of the community. In this work, we hope to draw the community's attention to the latter, which is arguably equally problematic for real applications. We promote two metrics to evaluate failure prediction. We show that a surprisingly straightforward and general approach, that we call ALERT, can predict the likely accuracy (or failure) of a variety of computer vision systems - semantic segmentation, vanishing point and camera parameter estimation, and image memorability prediction - on individual input images. We also explore attribute prediction, where classifiers are typically meant to generalize to new unseen categories. We show that ALERT can be useful in predicting failures of this transfer. Finally, we leverage ALERT to improve the performance of a downstream application of attribute prediction: zero-shot learning. We show that ALERT can outperform several strong baselines for zero-shot learning on four datasets.

Language is given meaning through its correspondence with a world representation.This correspondence can be at multiple levels of granularity or resolutions.In this paper, we introduce an approach to multi-resolution language grounding in the extremely challenging domain of professional soccer commentaries.We define and optimize a factored objective function that allows us to leverage discourse structure and the compositional nature of both language and game events.We show that finer resolution grounding helps coarser resolution grounding, and vice versa.Our method results in an F1 improvement of more than 48% versus the previous state of the art for fine-resolution grounding 1 .

We propose a method to expand the visual coverage of training sets that consist of a small number of labeled examples using learned attributes. Our optimization formulation discovers category specific attributes as well as the images that have high confidence in terms of the attributes. In addition, we propose a method to stably capture example-specific attributes for a small sized training set. Our method adds images to a category from a large unlabeled image pool, and leads to significant improvement in category recognition accuracy evaluated on a large-scale dataset, Image Net.

Users often have very specific visual content in mind that they are searching for. The most natural way to communicate this content to an image search engine is to use key-words that specify various properties or attributes of the content. A naive way of dealing with such multi-attribute queries is the following: train a classifier for each attribute independently, and then combine their scores on images to judge their fit to the query. We argue that this may not be the most effective or efficient approach. Conjunctions of attribute often correspond to very characteristic appearances. It would thus be beneficial to train classifiers that detect these conjunctions as a whole. But not all conjunctions result in such tight appearance clusters. So given a multi-attribute query, which conjunctions should we model? An exhaustive evaluation of all possible conjunctions would be time consuming. Hence we propose an optimization approach that identifies beneficial conjunctions without explicitly training the corresponding classifier. It reasons about geometric quantities that capture notions similar to intra- and inter-class variances. We exploit a discriminative binary space to compute these geometric quantities efficiently. Experimental results on two challenging datasets of objects and birds show that our proposed approach can improve performance significantly over several strong base-lines, while being an order of magnitude faster than exhaustively searching through all possible conjunctions.

When describing images, humans tend not to talk about the obvious, but rather mention what they find interesting. We argue that abnormalities and deviations from typicalities are among the most important components that form what is worth mentioning. In this paper we introduce the abnormality detection as a recognition problem and show how to model typicalities and, consequently, meaningful deviations from prototypical properties of categories. Our model can recognize abnormalities and report the main reasons of any recognized abnormality. We also show that abnormality predictions can help image categorization. We introduce the abnormality detection dataset and show interesting results on how to reason about abnormalities.

In this paper, we introduce visual phrases, complex visual composites like "a person riding a horse." Visual phrases often display significantly reduced visual complexity compared to their component objects because the appearance of those objects can change profoundly when they participate in relations. We introduce a dataset suitable for phrasal recognition that uses familiar PASCAL object categories, and demonstrate significant experimental gains resulting from exploiting visual phrases. We show that a visual phrase detector significantly outperforms a baseline which detects component objects and reasons about relations, even though visual phrase training sets tend to be smaller than those for objects. We argue that any multiclass detection system must decode detector outputs to produce final results; this is usually done with nonmaximum suppression. We describe a novel decoding procedure that can account accurately for local context without solving difficult inference problems. We show this decoding procedure outperforms the state of the art. Finally, we show that decoding a combination of phrasal and object detectors produces real improvements in detector results.

We show how to build large dictionaries of meaningful image fragments. These fragments could represent objects, objects in a local context, or parts of scenes. Our fragments operate as region-based exemplars, and we show how they can be used for image classification, to localize objects, and to compose new images. While each of these activities has been demonstrated before, each has required manually extracted fragments. Because our method for fragment extraction is automatic it can operate at a large scale. Our method uses recent advances in generic object detection techniques, together with discriminative tests to obtain good, clean fragment sets with extensive diversity. Our fragments are organized by the tags of the source images to build a semantically organized fragment table. A good set of fragment exemplars describes only the object, rather than object+context. Context could help identify an object; but it could also contribute noise, because other objects might appear in the same context. We show a slight improvement in classification performance by two standard exemplar matching methods using our fragment dictionary over such methods using image exemplars. This suggests that knowing the support of an exemplar is valuable. Furthermore, we demonstrate our automatically built fragment dictionary is capable of good localization. Finally, our fragment dictionary supports a keyword based fragment search system, which allows artists to get the fragments they need to make image collages.

In this paper we introduce visual phrases, complex visual composites like “a person riding a horse”. Visual phrases often display significantly reduced visual complexity compared to their component objects, because the appearance of those objects can change profoundly when they participate in relations. We introduce a dataset suitable for phrasal recognition that uses familiar PASCAL object categories, and demonstrate significant experimental gains resulting from exploiting visual phrases. We show that a visual phrase detector significantly outperforms a baseline which detects component objects and reasons about relations, even though visual phrase training sets tend to be smaller than those for objects. We argue that any multi-class detection system must decode detector outputs to produce final results; this is usually done with non-maximum suppression. We describe a novel decoding procedure that can account accurately for local context without solving difficult inference problems. We show this decoding procedure outperforms the state of the art. Finally, we show that decoding a combination of phrasal and object detectors produces real improvements in detector results.

We present a method to analyze daily activities, such as meal preparation, using video from an egocentric camera. Our method performs inference about activities, actions, hands, and objects. Daily activities are a challenging domain for activity recognition which are well-suited to an egocentric approach. In contrast to previous activity recognition methods, our approach does not require pre-trained detectors for objects and hands. Instead we demonstrate the ability to learn a hierarchical model of an activity by exploiting the consistent appearance of objects, hands, and actions that results from the egocentric context. We show that joint modeling of activities, actions, and objects leads to superior performance in comparison to the case where they are considered independently. We introduce a novel representation of actions based on object-hand interactions and experimentally demonstrate the superior performance of our representation in comparison to standard activity representations such as bag of words.

We propose an approach to find and describe objects within broad domains. We introduce a new dataset that provides annotation for sharing models of appearance and correlation across categories. We use it to learn part and category detectors. These serve as the visual basis for an integrated model of objects. We describe objects by the spatial arrangement of their attributes and the interactions between them. Using this model, our system can find animals and vehicles that it has not seen and infer attributes, such as function and pose. Our experiments demonstrate that we can more reliably locate and describe both familiar and unfamiliar objects, compared to a baseline that relies purely on basic category detectors.

Modern computer vision research consumes labelled data in quantity, and building datasets has become an important activity. The Internet has become a tremendous resource for computer vision researchers. By seeing the Internet as a vast, slightly disorganized collection of visual data, we can build datasets. The key point is that visual data are surrounded by contextual information like text and HTML tags, which is a strong, if noisy, cue to what the visual data means. In a series of case studies, we illustrate how useful this contextual information is. It can be used to build a large and challenging labelled face dataset with no manual intervention. With very small amounts of manual labor, contextual data can be used together with image data to identify pictures of animals. In fact, these contextual data are sufficiently reliable that a very large pool of noisily tagged images can be used as a resource to build image features, which reliably improve on conventional visual features. By seeing the Internet as a marketplace that can connect sellers of annotation services to researchers, we can obtain accurately annotated datasets quickly and cheaply. We describe methods to prepare data, check quality, and set prices for work for this annotation process. The problems posed by attempting to collect very big research datasets are fertile for researchers because collecting datasets requires us to focus on two important questions: What makes a good picture? What is the meaning of a picture?

We propose to shift the goal of recognition from naming to describing. Doing so allows us not only to name familiar objects, but also: to report unusual aspects of a familiar object (“spotty dog”, not just “dog”); to say something about unfamiliar objects (“hairy and four-legged”, not just “unknown”); and to learn how to recognize new objects with few or no visual examples. Rather than focusing on identity assignment, we make inferring attributes the core problem of recognition. These attributes can be semantic (“spotty”) or discriminative (“dogs have it but sheep do not”). Learning attributes presents a major new challenge: generalization across object categories, not just across instances within a category. In this paper, we also introduce a novel feature selection method for learning attributes that generalize well across categories. We support our claims by thorough evaluation that provides insights into the limitations of the standard recognition paradigm of naming and demonstrates the new abilities provided by our attribute-based framework.

Recognition using appearance features is confounded by phenomena that cause images of the same object to look different, or images of different objects to look the same. This may occur because the same object looks different from different viewing directions, or because two generally different objects have views from which they look similar. In this paper, we introduce the idea of discriminative aspect, a set of latent variables that encode these phenomena. Changes in view direction are one cause of changes in discriminative aspect, but others include changes in texture or lighting. However, images are not labelled with relevant discriminative aspect parameters. We describe a method to improve discrimination by inferring and then using latent discriminative aspect parameters. We apply our method to two parallel problems: object category recognition and human activity recognition. In each case, appearance features are powerful given appropriate training data, but traditionally fail badly under large changes in view. Our method can recognize an object quite reliably in a view for which it possesses no training example. Our method also reweights features to discount accidental similarities in appearance. We demonstrate that our method produces a significant improvement on the state of the art for both object and activity recognition.

Labeling image collections is a tedious task, especially when multiple labels have to be chosen for each image. In this paper we introduce a new framework that extends state of the art models in word prediction to incorporate information from unlabeled examples, using manifold regularization. To the best of our knowledge this is the first semi-supervised multi-task model used in vision problems. The new model can be solved using gradient descent and is fast and efficient. We show remarkable improvements for cases with few labeled examples for challenging multi-task learning problems in vision (predicting words for images and attributes for objects).

We build word models for American Sign Language (ASL) that transfer between different signers and different aspects. This is advantageous because one could use large amounts of labelled avatar data in combination with a smaller amount of labelled human data to spot a large number of words in human data. Transfer learning is possible because we represent blocks of video with novel intermediate discriminative features based on splits of the data. By constructing the same splits in avatar and human data and clustering appropriately, our features are both discriminative and semantically similar: across signers similar features imply similar words. We demonstrate transfer learning in two scenarios: from avatar to a frontally viewed human signer and from an avatar to human signer in a 3/4 view.

We describe a method to align ASL video subtitles with a closed-caption transcript. Our alignments are partial, based on spotting words within the video sequence, which consists of joined (rather than isolated) signs with unknown word boundaries. We start with windows known to contain an example of a word, but not limited to it. We estimate the start and end of the word in these examples using a voting method. This provides a small number of training examples (typically three per word). Since there is no shared structure, we use a discriminative rather than a generative word model. While our word spotters are not perfect, they are sufficient to establish an alignment. We demonstrate that quite small numbers of good word spotters results in an alignment good enough to produce simple English-ASL translations, both by phrase matching and using word substitution.