Joao Carreira

Carl Doersch, Yi Yang, Mel Vecerik et al.

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, and can be flexibly extended to higher-resolution videos. 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 on our project webpage.

Olivia Wiles, Joao Carreira, Iain Barr et al.

Experience and reasoning occur across multiple temporal scales: milliseconds, seconds, hours or days. The vast majority of computer vision research, however, still focuses on individual images or short videos lasting only a few seconds. This is because handling longer videos require more scalable approaches even to process them. In this work, we propose a framework enabling research on hour-long videos with the same hardware that can now process second-long videos. We replace standard video compression, e.g. JPEG, with neural compression and show that we can directly feed compressed videos as inputs to regular video networks. Operating on compressed videos improves efficiency at all pipeline levels -- data transfer, speed and memory -- making it possible to train models faster and on much longer videos. Processing compressed signals has, however, the downside of precluding standard augmentation techniques if done naively. We address that by introducing a small network that can apply transformations to latent codes corresponding to commonly used augmentations in the original video space. We demonstrate that with our compressed vision pipeline, we can train video models more efficiently on popular benchmarks such as Kinetics600 and COIN. We also perform proof-of-concept experiments with new tasks defined over hour-long videos at standard frame rates. Processing such long videos is impossible without using compressed representation.

Tengda Han, Sayna Ebrahimi, Dilara Gokay et al.

We introduce the "single-life" learning paradigm, where we train a distinct vision model exclusively on egocentric videos captured by one individual. We leverage the multiple viewpoints naturally captured within a single life to learn a visual encoder in a self-supervised manner. Our experiments demonstrate three key findings. First, models trained independently on different lives develop a highly aligned geometric understanding. We demonstrate this by training visual encoders on distinct datasets each capturing a different life, both indoors and outdoors, as well as introducing a novel cross-attention-based metric to quantify the functional alignment of the internal representations developed by different models. Second, we show that single-life models learn generalizable geometric representations that effectively transfer to downstream tasks, such as depth estimation, in unseen environments. Third, we demonstrate that training on up to 30 hours from one week of the same person's life leads to comparable performance to training on 30 hours of diverse web data, highlighting the strength of single-life representation learning. Overall, our results establish that the shared structure of the world, both leads to consistency in models trained on individual lives, and provides a powerful signal for visual representation learning.

Daniel Zoran, Nikhil Parthasarathy, Yi Yang et al.

We present Recurrent Video Masked-Autoencoders (RVM): a novel video representation learning approach that uses a transformer-based recurrent neural network to aggregate dense image features over time, effectively capturing the spatio-temporal structure of natural video data. RVM learns via an asymmetric masked prediction task requiring only a standard pixel reconstruction objective. This design yields a highly efficient ``generalist'' encoder: RVM achieves competitive performance with state-of-the-art video models (e.g. VideoMAE, V-JEPA) on video-level tasks like action recognition and point/object tracking, while also performing favorably against image models (e.g. DINOv2) on tasks that test geometric and dense spatial understanding. Notably, RVM achieves strong performance in the small-model regime without requiring knowledge distillation, exhibiting up to 30x greater parameter efficiency than competing video masked autoencoders. Moreover, we demonstrate that RVM's recurrent nature allows for stable feature propagation over long temporal horizons with linear computational cost, overcoming some of the limitations of standard spatio-temporal attention-based architectures. Finally, we use qualitative visualizations to highlight that RVM learns rich representations of scene semantics, structure, and motion.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Kelsey Allen, Carl Doersch, Guangyao Zhou et al.

A current limitation of video generative video models is that they generate plausible looking frames, but poor motion -- an issue that is not well captured by FVD and other popular methods for evaluating generated videos. Here we go beyond FVD by developing a metric which better measures plausible object interactions and motion. Our novel approach is based on auto-encoding point tracks and yields motion features that can be used to not only compare distributions of videos (as few as one generated and one ground truth, or as many as two datasets), but also for evaluating motion of single videos. We show that using point tracks instead of pixel reconstruction or action recognition features results in a metric which is markedly more sensitive to temporal distortions in synthetic data, and can predict human evaluations of temporal consistency and realism in generated videos obtained from open-source models better than a wide range of alternatives. We also show that by using a point track representation, we can spatiotemporally localize generative video inconsistencies, providing extra interpretability of generated video errors relative to prior work. An overview of the results and link to the code can be found on the project page: http://trajan-paper.github.io.

Shiting Xiao, Rishabh Kabra, Yuhang Li et al.

The ability to segment objects based on open-ended language prompts remains a critical challenge, requiring models to ground textual semantics into precise spatial masks while handling diverse and unseen categories. We present OpenWorldSAM, a framework that extends the prompt-driven Segment Anything Model v2 (SAM2) to open-vocabulary scenarios by integrating multi-modal embeddings extracted from a lightweight vision-language model (VLM). Our approach is guided by four key principles: i) Unified prompting: OpenWorldSAM supports a diverse range of prompts, including category-level and sentence-level language descriptions, providing a flexible interface for various segmentation tasks. ii) Efficiency: By freezing the pre-trained components of SAM2 and the VLM, we train only 4.5 million parameters on the COCO-stuff dataset, achieving remarkable resource efficiency. iii) Instance Awareness: We enhance the model's spatial understanding through novel positional tie-breaker embeddings and cross-attention layers, enabling effective segmentation of multiple instances. iv) Generalization: OpenWorldSAM exhibits strong zero-shot capabilities, generalizing well on unseen categories and an open vocabulary of concepts without additional training. Extensive experiments demonstrate that OpenWorldSAM achieves state-of-the-art performance in open-vocabulary semantic, instance, and panoptic segmentation across multiple benchmarks. Code is available at https://github.com/GinnyXiao/OpenWorldSAM.

Sjoerd van Steenkiste, Daniel Zoran, Yi Yang et al.

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.

Neerja Thakkar, Shiry Ginosar, Jacob Walker et al.

Visual intelligence requires anticipating the future behavior of agents, yet vision systems lack a general representation for motion and behavior. We propose dense point trajectories as visual tokens for behavior, a structured mid-level representation that disentangles motion from appearance and generalizes across diverse non-rigid agents, such as animals in-the-wild. Building on this abstraction, we design a diffusion transformer that models unordered sets of trajectories and explicitly reasons about occlusion, enabling coherent forecasts of complex motion patterns. To evaluate at scale, we curate 300 hours of unconstrained animal video with robust shot detection and camera-motion compensation. Experiments show that forecasting trajectory tokens achieves category-agnostic, data-efficient prediction, outperforms state-of-the-art baselines, and generalizes to rare species and morphologies, providing a foundation for predictive visual intelligence in the wild.

Goker Erdogan, Nikhil Parthasarathy, Catalin Ionescu et al.

We introduce LayerLock, a simple yet effective approach for self-supervised visual representation learning, that gradually transitions from pixel to latent prediction through progressive layer freezing. First, we make the observation that during training of video masked-autoencoding (MAE) models, ViT layers converge in the order of their depth: shallower layers converge early, deeper layers converge late. We then show that this observation can be exploited to accelerate standard MAE by progressively freezing the model according to an explicit schedule, throughout training. Furthermore, this same schedule can be used in a simple and scalable approach to latent prediction that does not suffer from "representation collapse". We apply our proposed approach, LayerLock, to large models of up to 4B parameters with results surpassing those of non-latent masked prediction on the 4DS perception suite.

Rishabh Kabra, Drew A. Hudson, Sjoerd van Steenkiste et al.

We introduce a hierarchical probabilistic approach to go from a 2D image to multiview 3D: a diffusion "prior" predicts the unseen 3D geometry, which then conditions a diffusion "decoder" to generate novel views of the subject. We use a pointmap-based geometric representation to coordinate the generation of multiple target views simultaneously. We construct a predictable distribution of geometric features per target view to enable learnability across examples, and generalization to arbitrary inputs images. Our modular, geometry-driven approach to novel-view synthesis (called "unPIC") beats competing baselines such as CAT3D, EscherNet, Free3D, and One-2-3-45 on held-out objects from ObjaverseXL, as well as unseen real-world objects from Google Scanned Objects, Amazon Berkeley Objects, and the Digital Twin Catalog.

Joao Carreira, Skanda Koppula, Daniel Zoran et al.

General perception systems such as Perceivers can process arbitrary modalities in any combination and are able to handle up to a few hundred thousand inputs. They achieve this generality by using exclusively global attention operations. This however hinders them from scaling up to the inputs sizes required to process raw high-resolution images or video. In this paper, we show that some degree of locality can be introduced back into these models, greatly improving their efficiency while preserving their generality. To scale them further, we introduce a self-supervised approach that enables learning dense low-dimensional positional embeddings for very large signals. We call the resulting model a Hierarchical Perceiver (HiP). In sum our contributions are: 1) scaling Perceiver-type models to raw high-resolution images and audio+video, 2) showing the feasibility of learning 1M+ positional embeddings from scratch using masked auto-encoding, 3) demonstrating competitive performance on raw data from ImageNet, AudioSet, PASCAL VOC, ModelNet40 and Kinetics datasets with the same exact, unchanged model and without specialized preprocessing or any tokenization.

Luyu Wang, Pauline Luc, Yan Wu et al.

The ability to learn universal audio representations that can solve diverse speech, music, and environment tasks can spur many applications that require general sound content understanding. In this work, we introduce a holistic audio representation evaluation suite (HARES) spanning 12 downstream tasks across audio domains and provide a thorough empirical study of recent sound representation learning systems on that benchmark. We discover that previous sound event classification or speech models do not generalize outside of their domains. We observe that more robust audio representations can be learned with the SimCLR objective; however, the model's transferability depends heavily on the model architecture. We find the Slowfast architecture is good at learning rich representations required by different domains, but its performance is affected by the normalization scheme. Based on these findings, we propose a novel normalizer-free Slowfast NFNet and achieve state-of-the-art performance across all domains.

Mateusz Malinowski, Dimitrios Vytiniotis, Grzegorz Swirszcz et al.

How can neural networks be trained on large-volume temporal data efficiently? To compute the gradients required to update parameters, backpropagation blocks computations until the forward and backward passes are completed. For temporal signals, this introduces high latency and hinders real-time learning. It also creates a coupling between consecutive layers, which limits model parallelism and increases memory consumption. In this paper, we build upon Sideways, which avoids blocking by propagating approximate gradients forward in time, and we propose mechanisms for temporal integration of information based on different variants of skip connections. We also show how to decouple computation and delegate individual neural modules to different devices, allowing distributed and parallel training. The proposed Skip-Sideways achieves low latency training, model parallelism, and, importantly, is capable of extracting temporal features, leading to more stable training and improved performance on real-world action recognition video datasets such as HMDB51, UCF101, and the large-scale Kinetics-600. Finally, we also show that models trained with Skip-Sideways generate better future frames than Sideways models, and hence they can better utilize motion cues.

Andrew Jaegle, Felix Gimeno, Andrew Brock et al.

Biological systems perceive the world by simultaneously processing high-dimensional inputs from modalities as diverse as vision, audition, touch, proprioception, etc. The perception models used in deep learning on the other hand are designed for individual modalities, often relying on domain-specific assumptions such as the local grid structures exploited by virtually all existing vision models. These priors introduce helpful inductive biases, but also lock models to individual modalities. In this paper we introduce the Perceiver - a model that builds upon Transformers and hence makes few architectural assumptions about the relationship between its inputs, but that also scales to hundreds of thousands of inputs, like ConvNets. The model leverages an asymmetric attention mechanism to iteratively distill inputs into a tight latent bottleneck, allowing it to scale to handle very large inputs. We show that this architecture is competitive with or outperforms strong, specialized models on classification tasks across various modalities: images, point clouds, audio, video, and video+audio. The Perceiver obtains performance comparable to ResNet-50 and ViT on ImageNet without 2D convolutions by directly attending to 50,000 pixels. It is also competitive in all modalities in AudioSet.

Mateusz Malinowski, Grzegorz Swirszcz, Joao Carreira et al.

We propose Sideways, an approximate backpropagation scheme for training video models. In standard backpropagation, the gradients and activations at every computation step through the model are temporally synchronized. The forward activations need to be stored until the backward pass is executed, preventing inter-layer (depth) parallelization. However, can we leverage smooth, redundant input streams such as videos to develop a more efficient training scheme? Here, we explore an alternative to backpropagation; we overwrite network activations whenever new ones, i.e., from new frames, become available. Such a more gradual accumulation of information from both passes breaks the precise correspondence between gradients and activations, leading to theoretically more noisy weight updates. Counter-intuitively, we show that Sideways training of deep convolutional video networks not only still converges, but can also potentially exhibit better generalization compared to standard synchronized backpropagation.

Joao Carreira, Eric Noland, Chloe Hillier et al.

We describe an extension of the DeepMind Kinetics human action dataset from 600 classes to 700 classes, where for each class there are at least 600 video clips from different YouTube videos. This paper details the changes introduced for this new release of the dataset, and includes a comprehensive set of statistics as well as baseline results using the I3D neural network architecture.

Joao Carreira, Eric Noland, Andras Banki-Horvath et al.

We describe an extension of the DeepMind Kinetics human action dataset from 400 classes, each with at least 400 video clips, to 600 classes, each with at least 600 video clips. In order to scale up the dataset we changed the data collection process so it uses multiple queries per class, with some of them in a language other than english -- portuguese. This paper details the changes between the two versions of the dataset and includes a comprehensive set of statistics of the new version as well as baseline results using the I3D neural network architecture. The paper is a companion to the release of the ground truth labels for the public test set.

Joao Carreira, Viorica Patraucean, Laurent Mazare et al.

We introduce a class of causal video understanding models that aims to improve efficiency of video processing by maximising throughput, minimising latency, and reducing the number of clock cycles. Leveraging operation pipelining and multi-rate clocks, these models perform a minimal amount of computation (e.g. as few as four convolutional layers) for each frame per timestep to produce an output. The models are still very deep, with dozens of such operations being performed but in a pipelined fashion that enables depth-parallel computation. We illustrate the proposed principles by applying them to existing image architectures and analyse their behaviour on two video tasks: action recognition and human keypoint localisation. The results show that a significant degree of parallelism, and implicitly speedup, can be achieved with little loss in performance.

Joao Carreira, Andrew Zisserman

The paucity of videos in current action classification datasets (UCF-101 and HMDB-51) has made it difficult to identify good video architectures, as most methods obtain similar performance on existing small-scale benchmarks. This paper re-evaluates state-of-the-art architectures in light of the new Kinetics Human Action Video dataset. Kinetics has two orders of magnitude more data, with 400 human action classes and over 400 clips per class, and is collected from realistic, challenging YouTube videos. We provide an analysis on how current architectures fare on the task of action classification on this dataset and how much performance improves on the smaller benchmark datasets after pre-training on Kinetics. We also introduce a new Two-Stream Inflated 3D ConvNet (I3D) that is based on 2D ConvNet inflation: filters and pooling kernels of very deep image classification ConvNets are expanded into 3D, making it possible to learn seamless spatio-temporal feature extractors from video while leveraging successful ImageNet architecture designs and even their parameters. We show that, after pre-training on Kinetics, I3D models considerably improve upon the state-of-the-art in action classification, reaching 80.9% on HMDB-51 and 98.0% on UCF-101.

Will Kay, Joao Carreira, Karen Simonyan et al.

We describe the DeepMind Kinetics human action video dataset. The dataset contains 400 human action classes, with at least 400 video clips for each action. Each clip lasts around 10s and is taken from a different YouTube video. The actions are human focussed and cover a broad range of classes including human-object interactions such as playing instruments, as well as human-human interactions such as shaking hands. We describe the statistics of the dataset, how it was collected, and give some baseline performance figures for neural network architectures trained and tested for human action classification on this dataset. We also carry out a preliminary analysis of whether imbalance in the dataset leads to bias in the classifiers.

Joao Carreira, Pulkit Agrawal, Katerina Fragkiadaki et al.

Hierarchical feature extractors such as Convolutional Networks (ConvNets) have achieved impressive performance on a variety of classification tasks using purely feedforward processing. Feedforward architectures can learn rich representations of the input space but do not explicitly model dependencies in the output spaces, that are quite structured for tasks such as articulated human pose estimation or object segmentation. Here we propose a framework that expands the expressive power of hierarchical feature extractors to encompass both input and output spaces, by introducing top-down feedback. Instead of directly predicting the outputs in one go, we use a self-correcting model that progressively changes an initial solution by feeding back error predictions, in a process we call Iterative Error Feedback (IEF). IEF shows excellent performance on the task of articulated pose estimation in the challenging MPII and LSP benchmarks, matching the state-of-the-art without requiring ground truth scale annotation.

Pulkit Agrawal, Joao Carreira, Jitendra Malik

The dominant paradigm for feature learning in computer vision relies on training neural networks for the task of object recognition using millions of hand labelled images. Is it possible to learn useful features for a diverse set of visual tasks using any other form of supervision? In biology, living organisms developed the ability of visual perception for the purpose of moving and acting in the world. Drawing inspiration from this observation, in this work we investigate if the awareness of egomotion can be used as a supervisory signal for feature learning. As opposed to the knowledge of class labels, information about egomotion is freely available to mobile agents. We show that given the same number of training images, features learnt using egomotion as supervision compare favourably to features learnt using class-label as supervision on visual tasks of scene recognition, object recognition, visual odometry and keypoint matching.

Joao Carreira, Sara Vicente, Lourdes Agapito et al.

While data has certainly taken the center stage in computer vision in recent years, it can still be difficult to obtain in certain scenarios. In particular, acquiring ground truth 3D shapes of objects pictured in 2D images remains a challenging feat and this has hampered progress in recognition-based object reconstruction from a single image. Here we propose to bypass previous solutions such as 3D scanning or manual design, that scale poorly, and instead populate object category detection datasets semi-automatically with dense, per-object 3D reconstructions, bootstrapped from:(i) class labels, (ii) ground truth figure-ground segmentations and (iii) a small set of keypoint annotations. Our proposed algorithm first estimates camera viewpoint using rigid structure-from-motion and then reconstructs object shapes by optimizing over visual hull proposals guided by loose within-class shape similarity assumptions. The visual hull sampling process attempts to intersect an object's projection cone with the cones of minimal subsets of other similar objects among those pictured from certain vantage points. We show that our method is able to produce convincing per-object 3D reconstructions and to accurately estimate cameras viewpoints on one of the most challenging existing object-category detection datasets, PASCAL VOC. We hope that our results will re-stimulate interest on joint object recognition and 3D reconstruction from a single image.