We address the challenge of representation learning from a continuous stream of video as input, in a self-supervised manner. This differs from the standard approaches to video learning where videos are chopped and shuffled during training in order to create a non-redundant batch that satisfies the independently and identically distributed (IID) sample assumption expected by conventional training paradigms. When videos are only available as a continuous stream of input, the IID assumption is evidently broken, leading to poor performance. We demonstrate the drop in performance when moving from shuffled to sequential learning on three tasks: the one-video representation learning method DoRA, standard VideoMAE on multi-video datasets, and the task of future video prediction.To address this drop, we propose a geometric modification to standard optimizers, to decorrelate batches by utilising orthogonal gradients during training. The proposed modification can be applied to any optimizer –we demonstrate it with Stochastic Gradient Descent (SGD) and AdamW. Our proposed orthogonal optimizer allows models trained from streaming videos to alleviate the drop in representation learning performance, as evaluated on downstream tasks. On three scenarios (DoRA, VideoMAE, future prediction), we show our orthogonal optimizer outperforms the strong AdamW in all three scenarios.

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.

Understanding long, real-world videos requires modeling of long-range visual dependencies. To this end, we explore video-first architectures, building on the common paradigm of transferring large-scale, image–text models to video via shallow temporal fusion. However, we expose two limitations to the approach: (1) decreased spatial capabilities, likely due to poor video–language alignment in standard video datasets, and (2) higher memory consumption, bottlenecking the number of frames that can be processed. To mitigate the memory bottleneck, we systematically analyze the memory/accuracy trade-off of various efficient methods: factorized attention, parameter-efficient image-to-video adaptation, input masking, and multi-resolution patchification. Surprisingly, simply masking large portions of the video (up to 75%) during contrastive pre-training proves to be one of the most robust ways to scale encoders to videos up to 4.3 minutes at 1 FPS. Our simple approach for training long video-to-text models, which scales to 1B parameters, does not add new architectural complexity and is able to outperform the popular paradigm of using much larger LLMs as an information aggregator over segment-based information on benchmarks with long-range temporal dependencies (YouCook2, EgoSchema).

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

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.

Most successful self-supervised learning methods are trained to align the representations of two independent views from the data. State-of-the-art methods in video are inspired by image techniques, where these two views are similarly extracted by cropping and augmenting the resulting crop. However, these methods miss a crucial element in the video domain: time. We introduce BraVe, a self-supervised learning framework for video. In BraVe, one of the views has access to a narrow temporal window of the video while the other view has a broad access to the video content. Our models learn to generalise from the narrow view to the general content of the video. Furthermore, BraVe processes the views with different backbones, enabling the use of alternative augmentations or modalities into the broad view such as optical flow, randomly convolved RGB frames, audio or their combinations. We demonstrate that BraVe achieves state-of-the-art results in self-supervised representation learning on standard video and audio classification benchmarks including UCF101, HMDB51, Kinetics, ESC-50 and AudioSet.

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.

We propose a line segment and elliptical arc detector that produces a reduced number of false detections on various types of images without any parameter tuning. For a given region of pixels in a grey-scale image, the detector decides whether a line segment or an elliptical arc is present (model validation). If both interpretations are possible for the same region, the detector chooses the one that best explains the data (model selection ). We describe a statistical criterion based on the a contrario theory, which serves for both validation and model selection. The experimental results highlight the performance of the proposed approach compared to state-of-the-art detectors, when applied on synthetic and real images.

Scene understanding is a prerequisite to many high level tasks for any automated intelligent machine operating in real world environments. Recent attempts with supervised learning have shown promise in this direction but also highlighted the need for enormous quantity of supervised data- performance increases in proportion to the amount of data used. However, this quickly becomes prohibitive when considering the manual labour needed to collect such data. In this work, we focus our attention on depth based semantic per-pixel labelling as a scene understanding problem and show the potential of computer graphics to generate virtually unlimited labelled data from synthetic 3D scenes. By carefully synthesizing training data with appropriate noise models we show comparable performance to state-of-the-art RGBD systems on NYUv2 dataset despite using only depth data as input and set a benchmark on depth-based segmentation on SUN RGB-D dataset.

We introduce SceneNet, a framework for generating high-quality annotated 3D scenes to aid indoor scene understanding. SceneNet leverages manually-annotated datasets of real world scenes such as NYUv2 to learn statistics about object co-occurrences and their spatial relationships. Using a hierarchical simulated annealing optimisation, these statistics are exploited to generate a potentially unlimited number of new annotated scenes, by sampling objects from various existing databases of 3D objects such as ModelNet, and textures such as OpenSurfaces and ArchiveTextures. Depending on the task, SceneNet can be used directly in the form of annotated 3D models for supervised training and 3D reconstruction benchmarking, or in the form of rendered annotated sequences of RGB-D frames or videos.

We describe a simple metric for image patches similarity, together with a robust criterion for unsupervised patch matching. The gradient orientations at corresponding positions in the two patches are compared and the normalized errors are accumulated. Based on the a contrario framework, the matching criterion validates a match between two patches when this cumulative error is too small to have occurred as the result of an accidental agreement. The method is illustrated in the validation of keypoint matches.

Building Information Models (BIMs) are becoming the official standard in the construction industry for encoding, reusing, and exchanging information about structural assets. Automatically generating such representations for existing assets stirs up the interest of various industrial, academic, and governmental parties, as it is expected to have a high economic impact. The purpose of this paper is to provide a general overview of the as-built modelling process, with focus on the geometric modelling side. Relevant works from the Computer Vision, Geometry Processing, and Civil Engineering communities are presented and compared in terms of their potential to lead to automatic as-built modelling.

Abstract We introduce a novel method for non‐rigid shape matching, designed to address the symmetric ambiguity problem present when matching shapes with intrinsic symmetries. Unlike the majority of existing methods which try to overcome this ambiguity by sampling a set of landmark correspondences, we address this problem directly by performing shape matching in an appropriate quotient space, where the symmetry has been identified and factored out. This allows us to both simplify the shape matching problem by matching between subspaces, and to return multiple solutions with equally good dense correspondences. Remarkably, both symmetry detection and shape matching are done without establishing any landmark correspondences between either points or parts of the shapes. This allows us to avoid an expensive combinatorial search present in most intrinsic symmetry detection and shape matching methods. We compare our technique with state‐of‐the‐art methods and show that superior performance can be achieved both when the symmetry on each shape is known and when it needs to be estimated.

We have at our disposal a large database containing images of various configurations of coplanar circles, randomly laid-out, called “Bubble Tags”. The images are taken from different viewpoints. Given a new image (query image), the goal is to find in the database the image containing the same bubble tag as the query image. We propose representing the images through projective invariant signatures which allow identifying the bubble tag without passing through an Euclidean reconstruction step. This is justified by the size of the database, which imposes the use of queries in 1D/vectorial form, i.e. not in 2D/matrix form. The experiments carried out confirm the efficiency of our approach, in terms of precision and complexity.