Motion control is crucial for generating expressive and compelling video content; however, most existing video generation models rely mainly on text prompts for control, which struggle to capture the nuances of dynamic actions and temporal compositions. To this end, we train a video generation model conditioned on spatiotemporally sparse or dense motion trajectories. In contrast to prior motion conditioning work, this flexible representation can encode any number of trajectories, object-specific or global scene motion, and temporally sparse motion; due to its flexibility we refer to this conditioning as motion prompts. While users may directly specify sparse trajectories, we also show how to translate high-level user requests into detailed, semi-dense motion prompts, a process we term motion prompt expansion. We demonstrate the versatility of our approach through various applications, including camera and object motion control, "interacting" with an image, motion transfer, and image editing. Our results showcase emergent behaviors, such as realistic physics, suggesting the potential of motion prompts for probing video models and interacting with future generative world models. Finally, we evaluate quantitatively, conduct a human study, and demonstrate strong performance.

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.

We introduce Genie, the first *generative interactive environment* trained in an unsupervised manner from unlabelled Internet videos. The model can be prompted to generate an endless variety of action-controllable virtual worlds described through text, synthetic images, photographs, and even sketches. At 11B parameters, Genie can be considered a *foundation world model*. It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model. Genie enables users to act in the generated environments on a frame-by-frame basis *despite training without any ground-truth action labels* or other domain specific requirements typically found in the world model literature. Further the resulting learned latent action space facilitates training agents to imitate behaviors from unseen videos, opening the path for training generalist agents of the future.

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 FlexCap, a vision-language model that generates region-specific descriptions of varying lengths. FlexCap is trained to produce length-conditioned captions for input boxes, enabling control over information density, with descriptions ranging from concise object labels to detailed captions. To achieve this, we create large-scale training datasets of image region descriptions with varying lengths from captioned web images. We demonstrate FlexCap’s effectiveness in several applications: first, it achieves strong performance in dense captioning tasks on the Visual Genome dataset. Second, we show how FlexCap’s localized descriptions can serve as input to a large language model to create a visual question answering (VQA) system, achieving state-of-the-art zero-shot performance on multiple VQA benchmarks. Our experiments illustrate FlexCap’s utility for tasks including image labeling, object attribute recognition, and visual dialog. Project webpage: https://flex-cap.github.io.

We address the problem of multi-object 3D pose control in image diffusion models. Instead of conditioning on a sequence of text tokens, we propose to use a set of per-object representations, *Neural Assets*, to control the 3D pose of individual objects in a scene. Neural Assets are obtained by pooling visual representations of objects from a reference image, such as a frame in a video, and are trained to reconstruct the respective objects in a different image, e.g., a later frame in the video. Importantly, we encode object visuals from the reference image while conditioning on object poses from the target frame, which enables learning disentangled appearance and position features. Combining visual and 3D pose representations in a sequence-of-tokens format allows us to keep the text-to-image interface of existing models, with Neural Assets in place of text tokens. By fine-tuning a pre-trained text-to-image diffusion model with this information, our approach enables fine-grained 3D pose and placement control of individual objects in a scene. We further demonstrate that Neural Assets can be transferred and recomposed across different scenes. Our model achieves state-of-the-art multi-object editing results on both synthetic 3D scene datasets, as well as two real-world video datasets (Objectron, Waymo Open).

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.

Self-supervised learning algorithms based on instance discrimination train encoders to be invariant to pre-defined transformations of the same instance. While most methods treat different views of the same image as positives for a contrastive loss, we are interested in using positives from other instances in the dataset. Our method, Nearest-Neighbor Contrastive Learning of visual Representations (NNCLR), samples the nearest neighbors from the dataset in the latent space, and treats them as positives. This provides more semantic variations than pre-defined transformations.We find that using the nearest-neighbor as positive in contrastive losses improves performance significantly on ImageNet classification using ResNet-50 under the linear evaluation protocol, from 71.7% to 75.6%, outperforming previous state-of-the-art methods. On semi-supervised learning benchmarks we improve performance significantly when only 1% ImageNet labels are available, from 53.8% to 56.5%. On transfer learning benchmarks our method outperforms state-of-the-art methods (including supervised learning with ImageNet) on 8 out of 12 downstream datasets. Furthermore, we demonstrate empirically that our method is less reliant on complex data augmentations. We see a relative reduction of only 2.1% ImageNet Top-1 accuracy when we train using only random crops.

In this paper, we introduce Recipe1M+, a new large-scale, structured corpus of over one million cooking recipes and 13 million food images. As the largest publicly available collection of recipe data, Recipe1M+ affords the ability to train high-capacity models on aligned, multi-modal data. Using these data, we train a neural network to learn a joint embedding of recipes and images that yields impressive results on an image-recipe retrieval task. Moreover, we demonstrate that regularization via the addition of a high-level classification objective both improves retrieval performance to rival that of humans and enables semantic vector arithmetic. We postulate that these embeddings will provide a basis for further exploration of the Recipe1M+ dataset and food and cooking in general. Code, data and models are publicly available.

We present an approach for estimating the period with which an action is repeated in a video. The crux of the approach lies in constraining the period prediction module to use temporal self-similarity as an intermediate representation bottleneck that allows generalization to unseen repetitions in videos in the wild. We train this model, called RepNet, with a synthetic dataset that is generated from a large unlabeled video collection by sampling short clips of varying lengths and repeating them with different periods and counts. This combination of synthetic data and a powerful yet constrained model, allows us to predict periods in a class-agnostic fashion. Our model substantially exceeds the state of the art performance on existing periodicity (PERTUBE) and repetition counting (QUVA) benchmarks. We also collect a new challenging dataset called Countix (~90 times larger than existing datasets) which captures the challenges of repetition counting in real-world videos. Project webpage: https://sites.google.com/view/repnet .

Collecting and automatically obtaining reward signals from real robotic visual data for the purposes of training reinforcement learning algorithms can be quite challenging and time-consuming. Methods for utilizing unlabeled data can have a huge potential to further accelerate robotic learning. We consider here the problem of performing manipulation tasks from pixels. In such tasks, choosing an appropriate state representation is crucial for planning and control. This is even more relevant with real images where noise, occlusions and resolution affect the accuracy and reliability of state estimation. In this work, we learn a latent state representation implicitly with deep reinforcement learning in simulation, and then adapt it to the real domain using unlabeled real robot data. We propose to do so by optimizing sequence-based self- supervised objectives. These use the temporal nature of robot experience, and can be common in both the simulated and real domains, without assuming any alignment of underlying states in simulated and unlabeled real images. We further propose a novel such objective, the Contrastive Forward Dynamics loss, which combines dynamics model learning with time-contrastive techniques. The learned state representation that results from our methods can be used to robustly solve a manipulation task in simulation and to successfully transfer the learned skill on a real system. We demonstrate the effectiveness of our approaches by training a vision-based reinforcement learning agent for cube stacking. Agents trained with our method, using only 5 hours of unlabeled real robot data for adaptation, shows a clear improvement over domain randomization, and standard visual domain adaptation techniques for sim-to-real transfer.

We introduce a self-supervised representation learning method based on the task of temporal alignment between videos. The method trains a network using temporal cycle-consistency (TCC), a differentiable cycle-consistency loss that can be used to find correspondences across time in multiple videos. The resulting per-frame embeddings can be used to align videos by simply matching frames using nearest-neighbors in the learned embedding space. To evaluate the power of the embeddings, we densely label the Pouring and Penn Action video datasets for action phases. We show that (i) the learned embeddings enable few-shot classification of these action phases, significantly reducing the supervised training requirements; and (ii) TCC is complementary to other methods of self-supervised learning in videos, such as Shuffle and Learn and Time-Contrastive Networks. The embeddings are also used for a number of applications based on alignment (dense temporal correspondence) between video pairs, including transfer of metadata of synchronized modalities between videos (sounds, temporal semantic labels), synchronized playback of multiple videos, and anomaly detection. Project webpage: https://sites.google.com/view/temporal-cycle-consistency .

Deep reinforcement learning methods traditionally struggle with tasks where environment rewards are particularly sparse. One successful method of guiding exploration in these domains is to imitate trajectories provided by a human demonstrator. However, these demonstrations are typically collected under artificial conditions, i.e. with access to the agent’s exact environment setup and the demonstrator’s action and reward trajectories. Here we propose a method that overcomes these limitations in two stages. First, we learn to map unaligned videos from multiple sources to a common representation using self-supervised objectives constructed over both time and modality (i.e. vision and sound). Second, we embed a single YouTube video in this representation to learn a reward function that encourages an agent to imitate human gameplay. This method of one-shot imitation allows our agent to convincingly exceed human-level performance on the infamously hard exploration games Montezuma’s Revenge, Pitfall! and Private Eye for the first time, even if the agent is not presented with any environment rewards.

People can recognize scenes across many different modalities beyond natural images. In this paper, we investigate how to learn cross-modal scene representations that transfer across modalities. To study this problem, we introduce a new cross-modal scene dataset. While convolutional neural networks can categorize scenes well, they also learn an intermediate representation not aligned across modalities, which is undesirable for cross-modal transfer applications. We present methods to regularize cross-modal convolutional neural networks so that they have a shared representation that is agnostic of the modality. Our experiments suggest that our scene representation can help transfer representations across modalities for retrieval. Moreover, our visualizations suggest that units emerge in the shared representation that tend to activate on consistent concepts independently of the modality.

In this paper, we introduce Recipe1M, a new large-scale, structured corpus of over 1m cooking recipes and 800k food images. As the largest publicly available collection of recipe data, Recipe1M affords the ability to train high-capacity models on aligned, multi-modal data. Using these data, we train a neural network to find a joint embedding of recipes and images that yields impressive results on an image-recipe retrieval task. Additionally, we demonstrate that regularization via the addition of a high-level classification objective both improves retrieval performance to rival that of humans and enables semantic vector arithmetic. We postulate that these embeddings will provide a basis for further exploration of the Recipe1M dataset and food and cooking in general. Code, data and models are publicly available.

Food is an integral part of our life and what and how much we eat crucially affects our health. Our food choices largely depend on how we perceive certain characteristics of food, such as whether it is healthy, delicious or if it qualifies as a salad. But these perceptions differ from person to person and one person's "single lettuce leaf" might be another person's "side salad". Studying how food is perceived in relation to what it actually is typically involves a laboratory setup. Here we propose to use recent advances in image recognition to tackle this problem. Concretely, we use data for 1.9 million images from Instagram from the US to look at systematic differences in how a machine would objectively label an image compared to how a human subjectively does. We show that this difference, which we call the "perception gap", relates to a number of health outcomes observed at the county level. To the best of our knowledge, this is the first time that image recognition is being used to study the "misalignment" of how people describe food images vs. what they actually depict.

People can recognize scenes across many different modalities beyond natural images. In this paper, we investigate how to learn cross-modal scene representations that transfer across modalities. To study this problem, we introduce a new cross-modal scene dataset. While convolutional neural networks can categorize cross-modal scenes well, they also learn an intermediate representation not aligned across modalities, which is undesirable for crossmodal transfer applications. We present methods to regularize cross-modal convolutional neural networks so that they have a shared representation that is agnostic of the modality. Our experiments suggest that our scene representation can help transfer representations across modalities for retrieval. Moreover, our visualizations suggest that units emerge in the shared representation that tend to activate on consistent concepts independently of the modality.

We learn rich natural sound representations by capitalizing on large amounts of unlabeled sound data collected in the wild. We leverage the natural synchronization between vision and sound to learn an acoustic representation using two-million unlabeled videos. Unlabeled video has the advantage that it can be economically acquired at massive scales, yet contains useful signals about natural sound. We propose a student-teacher training procedure which transfers discriminative visual knowledge from well established visual recognition models into the sound modality using unlabeled video as a bridge. Our sound representation yields significant performance improvements over the state-of-the-art results on standard benchmarks for acoustic scene/object classification. Visualizations suggest some high-level semantics automatically emerge in the sound network, even though it is trained without ground truth labels.

The objective of this work is object category detection in large scale image datasets in the manner of Video Google - an object category is specified by a HOG classifier template, and retrieval is immediate at run time. We make the following three contributions: (i) a new image representation based on mid-level discriminative patches, that is designed to be suited to immediate object category detection and inverted file indexing, (ii) a sparse representation of a HOG classifier using a set of mid-level discriminative classifier patches, and (iii) a fast method for spatial reranking images on their detections. We evaluate the detection method on the standard PASCAL VOC 2007 dataset, together with a 100K image subset of ImageNet, and demonstrate near state of the art detection performance at low ranks whilst maintaining immediate retrieval speeds. Applications are also demonstrated using an exemplar-SVM for pose matched retrieval.

Content based image retrieval (CBIR), the problem of searching digital images in large databases according to their visual content, is a well established research area in computer vision. In this work we are particularly interested in retrieving subwindows of images which are similar to the given query image, i.e. the goal is detection rather than image level classification. The notion of similarity is defined as being the same object class but also having similar viewpoint (e.g. frontal, left-facing, rear etc.). A query image can be a part of an object (e.g. head of a side facing horse), a complete object (e.g. frontal car image), or a composition of objects (visual phrases, e.g. person riding a horse). For instance, given a query of a horse facing left, the aim is to retrieve any left facing horse (intra-class variation) which might be walking or running with different feet formations (exemplar deformation).

Our objective is transfer training of a discriminatively trained object category detector, in order to reduce the number of training images required. To this end we propose three transfer learning formulations where a template learnt previously for other categories is used to regularize the training of a new category. All the formulations result in convex optimization problems. Experiments (on PASCAL VOC) demonstrate significant performance gains by transfer learning from one class to another (e.g. motorbike to bicycle), including one-shot learning, specialization from class to a subordinate class (e.g. from quadruped to horse) and transfer using multiple components. In the case of multiple training samples it is shown that a detection performance approaching that of the state of the art can be achieved with substantially fewer training samples.

This is a high level computer vision paper, which employs concepts from Natural Language Understanding in solving the video retrieval problem. Our main contribution is the utilization of the semantic word similarity measures (Lin and PMI-IR similarities) for video retrieval. In our approach, we use trained concept detectors, and the visual co-occurrence relations between such concepts. We propose two methods for content-based retrieval of videos: (1) A method for retrieving anewconcept(a concept which is not known to the system, and no annotation is available) using semantic word similarity and visual co-occurrence. (2) A method for retrieval of videos based on their relevance to a user defined text query using the semantic word similarity and visual content of videos. For evaluation purposes, we have mainly used the automatic search and the high level feature extraction test set of TRECVIDpsila06 benchmark, and the automatic search test set of TRECVIDpsila07. These two data sets consist of 250 hours of multilingual news video captured from American, Arabic, German and Chinese TV channels. Although our method for retrieving a new concept is an unsupervised method, it outperforms the trained concept detectors (which are supervised) on 7 out of 20 test concepts, and overall it performs very close to the trained detectors. On the other hand, our visual content based semantic retrieval method performs 81% better than the text-based retrieval method. This shows that using visual content alone we can obtain significantly good retrieval results.

In this paper we introduce a novel contextual fusion method to improve the detection scores of semantic concepts in images and videos. Our method consists of three phases. For each individual concept, the prior probability of the concept is incorporated with detection score of an individual SVM detector. Then probabilistic estimates of the target concept are computed using all of the individual SVM detectors. Finally, these estimates are linearly combined using weights learned from the training set. This procedure is applied to each target concept individually. We show significant improvements to our detection scores on the TRECVID 2005 development set and LSCOM-Lite annotation set. We achieved on average +3.9% improvements in 29 out of 39 concepts.