Diffusion models have revolutionized generative modeling, enabling unprecedented realism in image and video synthesis. This success has sparked interest in leveraging their representations for visual understanding tasks. While recent works have explored this potential for image generation, the visual understanding capabilities of video diffusion models remain largely uncharted. To address this gap, we systematically compare the same model architecture trained for video versus image generation, analyzing the performance of their latent representations on various downstream tasks including image classification, action recognition, depth estimation, and tracking. Results show that video diffusion models consistently outperform their image counterparts, though we find a striking range in the extent of this superiority. We further analyze features extracted from different layers and with varying noise levels, as well as the effect of model size and training budget on representation and generation quality. This work marks the first direct comparison of video and image diffusion objectives for visual understanding, offering insights into the role of temporal information in representation learning.

We introduce a framework for online learning from a single continuous video stream-the way people and animals learn, without mini-batches, data augmentation or shuffling. This poses great challenges given the high correlation between consecutive video frames and there is very little prior work on it. Our framework allows us to do a first deep dive into the topic and includes a collection of streams and tasks composed from two existing video datasets, plus methodology for performance evaluation that considers both adaptation and generalization. We employ pixel-to-pixel modelling as a practical and flexible way to switch between pre-training and single-stream evaluation as well as between arbitrary tasks, without ever requiring changes to models and always using the same pixel loss. Equipped with this framework we obtained large single-stream learning gains from pre-training with a novel family of future prediction tasks, found that momentum hurts, and that the pace of weight updates matters. The combination of these insights leads to matching the performance of IID learning with batch size 1, when using the same architecture and without costly replay buffers. An overview of the pa-per is available online at https://sites.google.com/view/one-stream-video.

Many robot manipulation tasks can be framed as geometric reasoning tasks, where an agent must be able to precisely manipulate an object into a position that satisfies the task from a set of initial conditions. Often, task success is defined based on the relationship between two objects - for instance, hanging a mug on a rack. In such cases, the solution should be equivariant to the initial position of the objects as well as the agent, and invariant to the pose of the camera. This poses a challenge for learning systems which attempt to solve this task by learning directly from high-dimensional demonstrations: the agent must learn to be both equivariant as well as precise, which can be challenging without any inductive biases about the problem. In this work, we propose a method for precise relative pose prediction which is provably SE(3)-equivariant, can be learned from only a few demonstrations, and can generalize across variations in a class of objects. We accomplish this by factoring the problem into learning an SE(3) invariant task-specific representation of the scene and then interpreting this representation with novel geometric reasoning layers which are provably SE(3) equivariant. We demonstrate that our method can yield substantially more precise placement predictions in simulated placement tasks than previous methods trained with the same amount of data, and can accurately represent relative placement relationships data collected from real-world demonstrations. Supplementary information and videos can be found at https://sites.google.com/view/reldist-iclr-2023.

For robots to be useful outside labs and specialized factories we need a way to teach them new useful behaviors quickly. Current approaches lack either the generality to onboard new tasks without task-specific engineering, or else lack the data-efficiency to do so in an amount of time that enables practical use. In this work we explore dense tracking as a representational vehicle to allow faster and more general learning from demonstration. Our approach utilizes Track-Any-Point (TAP) models to isolate the relevant motion in a demonstration, and parameterize a low-level controller to reproduce this motion across changes in the scene configuration. We show this results in robust robot policies that can solve complex object-arrangement tasks such as shape-matching, stacking, and even full path-following tasks such as applying glue and sticking objects together, all from demonstrations that can be collected in minutes.

We introduce a new benchmark, TAPVid-3D, for evaluating the task of long-range Tracking Any Point in 3D (TAP-3D). While point tracking in two dimensions (TAP-2D) has many benchmarks measuring performance on real-world videos, such as TAPVid-DAVIS, three-dimensional point tracking has none. To this end, leveraging existing footage, we build a new benchmark for 3D point tracking featuring 4,000+ real-world videos, composed of three different data sources spanning a variety of object types, motion patterns, and indoor and outdoor environments. To measure performance on the TAP-3D task, we formulate a collection of metrics that extend the Jaccard-based metric used in TAP-2D to handle the complexities of ambiguous depth scales across models, occlusions, and multi-track spatio-temporal smoothness. We manually verify a large sample of trajectories to ensure correct video annotations, and assess the current state of the TAP-3D task by constructing competitive baselines using existing tracking models. We anticipate this benchmark will serve as a guidepost to improve our ability to understand precise 3D motion and surface deformation from monocular video.

Current vision models typically maintain a fixed correspondence between their representation structure and image space. Each layer comprises a set of tokens arranged “on-the-grid,” which biases patches or tokens to encode information at a specific spatio(-temporal) location. In this work we present *Moving Off-the-Grid* (MooG), a self-supervised video representation model that offers an alternative approach, allowing tokens to move “off-the-grid” to better enable them to represent scene elements consistently, even as they move across the image plane through time. By using a combination of cross-attention and positional embeddings we disentangle the representation structure and image structure. We find that a simple self-supervised objective—next frame prediction—trained on video data, results in a set of latent tokens which bind to specific scene structures and track them as they move. We demonstrate the usefulness of MooG’s learned representation both qualitatively and quantitatively by training readouts on top of the learned representation on a variety of downstream tasks. We show that MooG can provide a strong foundation for different vision tasks when compared to “on-the-grid” baselines.

We present a novel model for Tracking Any Point (TAP) that effectively tracks any queried point on any physical surface throughout a video sequence. Our approach employs two stages: (1) a matching stage, which independently locates a suitable candidate point match for the query point on every other frame, and (2) a refinement stage, which updates both the trajectory and query features based on local correlations. The resulting model surpasses all baseline methods by a significant margin on the TAP-Vid benchmark, as demonstrated by an approximate 20% absolute average Jaccard (AJ) improvement on DAVIS. Our model facilitates fast inference on long and high-resolution video sequences. On a modern GPU, our implementation has the capacity to track points faster than real-time. Given the high-quality trajectories extracted from a large dataset, we demonstrate a proof-of-concept diffusion model which generates trajectories from static images, enabling plausible animations. Visualizations, source code, and pretrained models can be found at https://deepmind-tapir.github.io.

We propose a novel multimodal video benchmark - the Perception Test - to evaluate the perception and reasoning skills of pre-trained multimodal models (e.g. Flamingo, BEiT-3, or GPT-4). Compared to existing benchmarks that focus on computational tasks (e.g. classification, detection or tracking), the Perception Test focuses on skills (Memory, Abstraction, Physics, Semantics) and types of reasoning (descriptive, explanatory, predictive, counterfactual) across video, audio, and text modalities, to provide a comprehensive and efficient evaluation tool. The benchmark probes pre-trained models for their transfer capabilities, in a zero-shot / few-shot or limited finetuning regime. For these purposes, the Perception Test introduces 11.6k real-world videos, 23s average length, designed to show perceptually interesting situations, filmed by around 100 participants worldwide. The videos are densely annotated with six types of labels (multiple-choice and grounded video question-answers, object and point tracks, temporal action and sound segments), enabling both language and non-language evaluations. The fine-tuning and validation splits of the benchmark are publicly available (CC-BY license), in addition to a challenge server with a held-out test split. Human baseline results compared to state-of-the-art video QA models show a significant gap in performance (91.4% vs 45.8%), suggesting that there is significant room for improvement in multimodal video understanding.Dataset, baselines code, and challenge server are available at https://github.com/deepmind/perception_test

Generic motion understanding from video involves not only tracking objects, but also perceiving how their surfaces deform and move. This information is useful to make inferences about 3D shape, physical properties and object interactions. While the problem of tracking arbitrary physical points on surfaces over longer video clips has received some attention, no dataset or benchmark for evaluation existed, until now. In this paper, we first formalize the problem, naming it tracking any point (TAP). We introduce a companion benchmark,TAP-Vid, which is composed of both real-world videos with accurate human annotations of point tracks, and synthetic videos with perfect ground-truth point tracks. Central to the construction of our benchmark is a novel semi-automatic crowdsourced pipeline which uses optical flow estimates to compensate for easier, short-term motion like camera shake, allowing annotators to focus on harder sections of the video. We validate our pipeline on synthetic data and propose a simple end-to-end point tracking model, TAP-Net, showing that it outperforms all prior methods on our benchmark when trained on synthetic data.

In this paper, we propose UnOS, an unified system for unsupervised optical flow and stereo depth estimation using convolutional neural network (CNN) by taking advantages of their inherent geometrical consistency based on the rigid-scene assumption. UnOS significantly outperforms other state-of-the-art (SOTA) unsupervised approaches that treated the two tasks independently. Specifically, given two consecutive stereo image pairs from a video, UnOS estimates per-pixel stereo depth images, camera ego-motion and optical flow with three parallel CNNs. Based on these quantities, UnOS computes rigid optical flow and compares it against the optical flow estimated from the FlowNet, yielding pixels satisfying the rigid-scene assumption. Then, we encourage geometrical consistency between the two estimated flows within rigid regions, from which we derive a rigid-aware direct visual odometry (RDVO) module. We also propose rigid and occlusion-aware flow-consistency losses for the learning of UnOS. We evaluated our results on the popular KITTI dataset over 4 related tasks, \ie stereo depth, optical flow, visual odometry and motion segmentation.

Recognizing the attributes of objects and their parts is central to many computer vision applications. Although great progress has been made to apply object-level recognition, recognizing the attributes of parts remains less applicable since the training data for part attributes recognition is usually scarce especially for internet-scale applications. Furthermore, most existing part attribute recognition methods rely on the part annotations which are more expensive to obtain. In order to solve the data insufficiency problem and get rid of dependence on the part annotation, we introduce a novel Concept Sharing Network (CSN) for part attribute recognition. A great advantage of CSN is its capability of recognizing the part attribute (a combination of part location and appearance pattern) that has insufficient or zero training data, by learning the part location and appearance pattern respectively from the training data that usually mix them in a single label. Extensive experiments on CUB, Celeb A, and a newly proposed human attribute dataset demonstrate the effectiveness of CSN and its advantages over other methods, especially for the attributes with few training samples. Further experiments show that CSN can also perform zero-shot part attribute recognition.

Feedback is a fundamental mechanism existing in the human visual system, but has not been explored deeply in designing computer vision algorithms. In this paper, we claim that feedback plays a critical role in understanding convolutional neural networks (CNNs), e.g., how a neuron in CNNs describes an object's pattern, and how a collection of neurons form comprehensive perception to an object. To model the feedback in CNNs, we propose a novel model named Feedback CNN and develop two new processing algorithms, i.e., neural pathway pruning and pattern recovering. We mathematically prove that the proposed method can reach local optimum. Note that Feedback CNN belongs to weakly supervised methods and can be trained only using category-level labels. But it possesses a powerful capability to accurately localize and segment category-specific objects. We conduct extensive visualization analysis, and the results reveal the close relationship between neurons and object parts in Feedback CNN. Finally, we evaluate the proposed Feedback CNN over the tasks of weakly supervised object localization and segmentation, and the experimental results on ImageNet and Pascal VOC show that our method remarkably outperforms the state-of-the-art ones.

It has been recently shown that a convolutional neural network can learn optical flow estimation with unsupervised learning. However, the performance of the unsupervised methods still has a relatively large gap compared to its supervised counterpart. Occlusion and large motion are some of the major factors that limit the current unsupervised learning of optical flow methods. In this work we introduce a new method which models occlusion explicitly and a new warping way that facilitates the learning of large motion. Our method shows promising results on Flying Chairs, MPI-Sintel and KITTI benchmark datasets. Especially on KITTI dataset where abundant unlabeled samples exist, our unsupervised method outperforms its counterpart trained with supervised learning.

Incorporating multi-scale features in fully convolutional neural networks (FCNs) has been a key element to achieving state-of-the-art performance on semantic image segmentation. One common way to extract multi-scale features is to feed multiple resized input images to a shared deep network and then merge the resulting features for pixelwise classification. In this work, we propose an attention mechanism that learns to softly weight the multi-scale features at each pixel location. We adapt a state-of-the-art semantic image segmentation model, which we jointly train with multi-scale input images and the attention model. The proposed attention model not only outperforms averageand max-pooling, but allows us to diagnostically visualize the importance of features at different positions and scales. Moreover, we show that adding extra supervision to the output at each scale is essential to achieving excellent performance when merging multi-scale features. We demonstrate the effectiveness of our model with extensive experiments on three challenging datasets, including PASCAL-Person-Part, PASCAL VOC 2012 and a subset of MS-COCO 2014.

While deep convolutional neural networks (CNNs) have shown a great success in single-label image classification, it is important to note that real world images generally contain multiple labels, which could correspond to different objects, scenes, actions and attributes in an image. Traditional approaches to multi-label image classification learn independent classifiers for each category and employ ranking or thresholding on the classification results. These techniques, although working well, fail to explicitly exploit the label dependencies in an image. In this paper, we utilize recurrent neural networks (RNNs) to address this problem. Combined with CNNs, the proposed CNN-RNN framework learns a joint image-label embedding to characterize the semantic label dependency as well as the image-label relevance, and it can be trained end-to-end from scratch to integrate both information in a unified framework. Experimental results on public benchmark datasets demonstrate that the proposed architecture achieves better performance than the state-of-the-art multi-label classification models.

We present an approach that exploits hierarchical Recurrent Neural Networks (RNNs) to tackle the video captioning problem, i.e., generating one or multiple sentences to describe a realistic video. Our hierarchical framework contains a sentence generator and a paragraph generator. The sentence generator produces one simple short sentence that describes a specific short video interval. It exploits both temporal-and spatial-attention mechanisms to selectively focus on visual elements during generation. The paragraph generator captures the inter-sentence dependency by taking as input the sentential embedding produced by the sentence generator, combining it with the paragraph history, and outputting the new initial state for the sentence generator. We evaluate our approach on two large-scale benchmark datasets: YouTubeClips and TACoS-MultiLevel. The experiments demonstrate that our approach significantly outperforms the current state-of-the-art methods with BLEU@4 scores 0.499 and 0.305 respectively.

While feedforward deep convolutional neural networks (CNNs) have been a great success in computer vision, it is important to note that the human visual cortex generally contains more feedback than feedforward connections. In this paper, we will briefly introduce the background of feedbacks in the human visual cortex, which motivates us to develop a computational feedback mechanism in deep neural networks. In addition to the feedforward inference in traditional neural networks, a feedback loop is introduced to infer the activation status of hidden layer neurons according to the "goal" of the network, e.g., high-level semantic labels. We analogize this mechanism as "Look and Think Twice." The feedback networks help better visualize and understand how deep neural networks work, and capture visual attention on expected objects, even in images with cluttered background and multiple objects. Experiments on ImageNet dataset demonstrate its effectiveness in solving tasks such as image classification and object localization.

In this paper, we address the task of learning novel visual concepts, and their interactions with other concepts, from a few images with sentence descriptions. Using linguistic context and visual features, our method is able to efficiently hypothesize the semantic meaning of new words and add them to its word dictionary so that they can be used to describe images which contain these novel concepts. Our method has an image captioning module based on [38] with several improvements. In particular, we propose a transposed weight sharing scheme, which not only improves performance on image captioning, but also makes the model more suitable for the novel concept learning task. We propose methods to prevent overfitting the new concepts. In addition, three novel concept datasets are constructed for this new task, and are publicly available on the project page. In the experiments, we show that our method effectively learns novel visual concepts from a few examples without disturbing the previously learned concepts. The project page is: www.stat.ucla.edu/junhua. mao/projects/child_learning.html.

Hand pose estimation has matured rapidly in recent years. The introduction of commodity depth sensors and a multitude of practical applications have spurred new advances. We provide an extensive analysis of the state-of-the-art, focusing on hand pose estimation from a single depth frame. To do so, we have implemented a considerable number of systems, and will release all software and evaluation code. We summarize important conclusions here: (1) Pose estimation appears roughly solved for scenes with isolated hands. However, methods still struggle to analyze cluttered scenes where hands may be interacting with nearby objects and surfaces. To spur further progress we introduce a challenging new dataset with diverse, cluttered scenes. (2) Many methods evaluate themselves with disparate criteria, making comparisons difficult. We define a consistent evaluation criteria, rigorously motivated by human experiments. (3) We introduce a simple nearest-neighbor baseline that outperforms most existing systems. This implies that most systems do not generalize beyond their training sets. This also reinforces the under-appreciated point that training data is as important as the model itself. We conclude with directions for future progress.

Occlusion poses a significant difficulty for object recognition due to the combinatorial diversity of possible occlusion patterns. We take a strongly supervised, non-parametric approach to modeling occlusion by learning deformable models with many local part mixture templates using large quantities of synthetically generated training data. This allows the model to learn the appearance of different occlusion patterns including figure-ground cues such as the shapes of occluding contours as well as the co-occurrence statistics of occlusion between neighboring parts. The underlying part mixture-structure also allows the model to capture coherence of object support masks between neighboring parts and make compelling predictions of figure-ground-occluder segmentations. We test the resulting model on human pose estimation under heavy occlusion and find it produces improved localization accuracy.

AutoCaption is a system that helps a smartphone user generate a caption for their photos. It operates by uploading the photo to a cloud service where a number of parallel modules are applied to recognize a variety of entities and relations. The outputs of the modules are combined to generate a large set of candidate captions, which are returned to the phone. The phone client includes a convenient user interface that allows users to select their favorite caption, reorder, add, or delete words to obtain the grammatical style they prefer. The user can also select from multiple candidates returned by the recognition modules.

We describe a method for articulated human detection and human pose estimation in static images based on a new representation of deformable part models. Rather than modeling articulation using a family of warped (rotated and foreshortened) templates, we use a mixture of small, nonoriented parts. We describe a general, flexible mixture model that jointly captures spatial relations between part locations and co-occurrence relations between part mixtures, augmenting standard pictorial structure models that encode just spatial relations. Our models have several notable properties: 1) They efficiently model articulation by sharing computation across similar warps, 2) they efficiently model an exponentially large set of global mixtures through composition of local mixtures, and 3) they capture the dependency of global geometry on local appearance (parts look different at different locations). When relations are tree structured, our models can be efficiently optimized with dynamic programming. We learn all parameters, including local appearances, spatial relations, and co-occurrence relations (which encode local rigidity) with a structured SVM solver. Because our model is efficient enough to be used as a detector that searches over scales and image locations, we introduce novel criteria for evaluating pose estimation and human detection, both separately and jointly. We show that currently used evaluation criteria may conflate these two issues. Most previous approaches model limbs with rigid and articulated templates that are trained independently of each other, while we present an extensive diagnostic evaluation that suggests that flexible structure and joint training are crucial for strong performance. We present experimental results on standard benchmarks that suggest our approach is the state-of-the-art system for pose estimation, improving past work on the challenging Parse and Buffy datasets while being orders of magnitude faster.

Proxemics is the study of how people interact. We present a computational formulation of visual proxemics by attempting to label each pair of people in an image with a subset of physically based “touch codes.” A baseline approach would be to first perform pose estimation and then detect the touch codes based on the estimated joint locations. We found that this sequential approach does not perform well because pose estimation step is too unreliable for images of interacting people, due to difficulties with occlusion and limb ambiguities. Instead, we propose a direct approach where we build an articulated model tuned for each touch code. Each such model contains two people, connected in an appropriate manner for the touch code in question. We fit this model to the image and then base classification on the fitting error. Experiments show that this approach significantly outperforms the sequential baseline as well as other related approches.

We formulate a layered model for object detection and image segmentation. We describe a generative probabilistic model that composites the output of a bank of object detectors in order to define shape masks and explain the appearance, depth ordering, and labels of all pixels in an image. Notably, our system estimates both class labels and object instance labels. Building on previous benchmark criteria for object detection and image segmentation, we define a novel score that evaluates both class and instance segmentation. We evaluate our system on the PASCAL 2009 and 2010 segmentation challenge data sets and show good test results with state-of-the-art performance in several categories, including segmenting humans.

We describe a method for human pose estimation in static images based on a novel representation of part models. Notably, we do not use articulated limb parts, but rather capture orientation with a mixture of templates for each part. We describe a general, flexible mixture model for capturing contextual co-occurrence relations between parts, augmenting standard spring models that encode spatial relations. We show that such relations can capture notions of local rigidity. When co-occurrence and spatial relations are tree-structured, our model can be efficiently optimized with dynamic programming. We present experimental results on standard benchmarks for pose estimation that indicate our approach is the state-of-the-art system for pose estimation, outperforming past work by 50% while being orders of magnitude faster.

We formulate a layered model for object detection and multi-class segmentation. Our system uses the output of a bank of object detectors in order to define shape priors for support masks and then estimates appearance, depth ordering and labeling of pixels in the image. We train our system on the PASCAL segmentation challenge dataset and show good test results with state of the art performance in several categories including segmenting humans.