We present Segment Anything Model (SAM) 3, a unified model that detects, segments, and tracks objects in images and videos based on concept prompts, which we define as either short noun phrases (e.g., “yellow school bus”), image exemplars, or a combination of both. Promptable Concept Segmentation (PCS) takes such prompts and returns segmentation masks and unique identities for all matching object instances. To advance PCS, we build a scalable data engine that produces a high-quality dataset with 4M unique concept labels, including hard negatives, across images and videos. Our model consists of a vision backbone shared between an image-level detector and a memory-based video tracker. Recognition and localization are decoupled with a presence head, which significantly boosts detection accuracy. SAM 3 delivers a 2x gain over existing systems in both image and video PCS, and improves previous SAM capabilities in interactive visual segmentation tasks. We open source SAM 3 along with our new Segment Anything with Concepts (SA-Co) benchmark.

Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM–VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about ''what'', ''where'', ''when'', and ''how'' of a video. We make our work fully reproducible by providing data, training recipes, code & models.

We introduce ED-VTG, a method for fine-grained video temporal grounding utilizing multi-modal large language models. Our approach harnesses the capabilities of multimodal LLMs to jointly process text and video, in order to effectively localize natural language queries in videos through a two-stage process. Rather than being directly grounded, language queries are initially transformed into enriched sentences that incorporate missing details and cues to aid in grounding. In the second stage, these enriched queries are grounded, using a lightweight decoder, which specializes at predicting accurate boundaries conditioned on contextualized representations of the enriched queries. To mitigate noise and reduce the impact of hallucinations, our model is trained with a multiple-instance-learning objective that dynamically selects the optimal version of the query for each training sample. We demonstrate state-of-the-art results across various benchmarks in temporal video grounding and paragraph grounding settings. Experiments reveal that our method significantly outperforms all previously proposed LLM-based temporal grounding approaches and is either superior or comparable to specialized models, while maintaining a clear advantage against them in zero-shot evaluation scenarios.

We present Ego-Exo4D, a diverse, large-scale multimodal multiview video dataset and benchmark challenge. Ego-Exo4D centers around simultaneously-captured egocentric and exocentric video of skilled human activities (e.g., sports, music, dance, bike repair). 740 participants from 13 cities worldwide performed these activities in 123 different natural scene contexts, yielding long-form captures from 1 to 42 minutes each and 1,286 hours of video combined. The multimodal nature of the dataset is unprecedented: the video is accompanied by multichannel audio, eye gaze, 3D point clouds, camera poses, IMU, and multiple paired language descriptions—including a novel “expert commentary” done by coaches and teachers and tailored to the skilled-activity domain. To push the frontier of first-person video understanding of skilled human activity, we also present a suite of benchmark tasks and their annotations, including fine-grained activity understanding, proficiency estimation, cross-view translation, and 3D hand/body pose. All resources are open sourced to fuel new research in the community. https://ego-exo4d-data.org/

Egocentric videos provide a first-person perspective of the wearer's activities, involving simultaneous interactions with multiple objects. In this work, we propose the task of weakly-supervised Narration-based Video Object Segmentation (NVOS). Given an egocentric video clip and a narration of the wearer's activities, our aim is to segment object instances mentioned in the narration, without using any spatial annotations during training. Existing weakly-supervised video object grounding methods typically yield bounding boxes for referred objects. In contrast, we propose ROSA, a weakly-supervised pixel-level grounding framework learning alignments between referred objects and segmentation mask proposals. Our model harnesses vision-language models pre-trained on image-text pairs to embed region masks and object phrases. During training, we combine (a) a video-narration contrastive loss that implicitly supervises the alignment between regions and phrases, and (b) a region-phrase contrastive loss based on inferred latent alignments. To address the lack of annotated NVOS datasets in egocentric videos, we create a new evaluation benchmark, VISOR-NVOS, leveraging existing annotations of segmentation masks from VISOR alongside 14.6k newly-collected, object-based video clip narrations. Our approach achieves state-of-the-art zero-shot pixel-level grounding performance compared to strong baselines under similar supervision. Additionally, we demonstrate generalization capabilities for zero-shot video object grounding on YouCook2, a third-person instructional video dataset.

In this paper we present an approach for localizing steps of procedural activities in narrated how-to videos. To deal with the scarcity of labeled data at scale, we source the step descriptions from a language knowledge base (wikiHow) containing instructional articles for a large variety of procedural tasks. Without any form of manual supervision, our model learns to temporally ground the steps of procedural articles in how-to videos by matching three modalities: frames, narrations, and step descriptions. Specifically, our method aligns steps to video by fusing information from two distinct pathways: i) direct alignment of step descriptions to frames, ii) indirect alignment obtained by composing steps-to-narrations with narrations-to-video correspondences. Notably, our approach performs global temporal grounding of all steps in an article at once by exploiting order information, and is trained with step pseudo-labels which are iteratively refined and aggressively filtered. In order to validate our model we introduce a new benchmark – HT-Step – obtained by manually annotating a 124-hour subset of HowTo100M 1 with steps sourced from wikiHow articles. Experiments on this benchmark as well as zero-shot evaluations on CrossTask demonstrate that our multi-modality alignment yields dramatic gains over several baselines and prior works. Finally, we show that our inner module for matching narration-to-video outperforms by a large margin the state of the art on the HTM-Align narration-video alignment benchmark.

Video understanding tasks take many forms, from action detection to visual query localization and spatio-temporal grounding of sentences. These tasks differ in the type of inputs (only video, or video-query pair where query is an image region or sentence) and outputs (temporal segments or spatio-temporal tubes). However, at their core they require the same fundamental understanding of the video, i.e., the actors and objects in it, their actions and interactions. So far these tasks have been tackled in isolation with individual, highly specialized architectures, which do not exploit the interplay between tasks. In contrast, in this paper, we present a single, unified model for tackling query-based video understanding in long-form videos. In particular, our model can address all three tasks of the Ego4D Episodic Memory benchmark which entail queries of three different forms: given an egocentric video and a visual, textual or activity query, the goal is to determine when and where the answer can be seen within the video. Our model design is inspired by recent query-based approaches to spatio-temporal grounding, and contains modality-specific query encoders and task-specific sliding window inference that allow multi-task training with diverse input modalities and different structured outputs. We exhaustively analyze relationships among the tasks and illustrate that cross-task learning leads to improved performance on each individual task, as well as the ability to generalize to unseen tasks, such as zero-shot spatial localization of language queries.

We introduce HT-Step, a large-scale dataset containing temporal annotations of instructional article steps in cooking videos. It includes 116k segment-level annotations over 20k narrated videos (approximately 2.1k hours) of the HowTo100M dataset. Each annotation provides a temporal interval, and a categorical step label from a taxonomy of 4, 958 unique steps automatically mined from wikiHow articles which include rich descriptions of each step. Our dataset significantly surpasses existing labeled step datasets in terms of scale, number of tasks, and richness of natural language step descriptions. Based on these annotations, we introduce a strongly supervised benchmark for aligning instructional articles with how-to videos and present a comprehensive evaluation of baseline methods for this task. By publicly releasing these annotations and defining rigorous evaluation protocols and metrics, we hope to significantly accelerate research in the field of procedural activity understanding.

Given weak supervision from image- or video-caption pairs, we address the problem of grounding (localizing) each object word of a ground-truth or generated sentence describing a visual input. Recent weakly-supervised approaches leverage region proposals and ground words based on the region attention coefficients of captioning models. To predict each next word in the sentence they attend over regions using a summary of the previous words as a query, and then ground the word by selecting the most attended regions. However, this leads to sub-optimal grounding, since attention coefficients are computed without taking into account the word that needs to be localized. To address this shortcoming, we propose a novel Grounded Visual Description Conditional Variational Autoencoder (GVD-CVAE) and leverage its latent variables for grounding. In particular, we introduce a discrete random variable that models each word-to-region alignment, and learn its approximate posterior distribution given the full sentence. Experiments on challenging image and video datasets (Flickr30k Entities, YouCook2, ActivityNet Entities) validate the effectiveness of our conditional generative model, showing that it can substantially outperform soft-attention-based baselines in grounding.

We address the problem of detecting human and vehicle activities in long, untrimmed surveillance videos that capture a large field of view. Most existing activity detection approaches are designed for recognizing atomic human actions performed in the foreground. Therefore, they are not suitable for detecting activities in extended videos, which contain multiple actors performing co-occurring, complex activities with extreme spatio-temporal scale variations. In this paper, we propose a modular, actor-centric framework for real-time activity detection in extended videos. In particular, we decompose an extended video into a collection of smaller actor-centric tubelets of interest. Each tubelet is a video sub-volume associated with an actor and includes adaptive visual context for recognizing the actor’s activities. Once these tubelets are extracted via an object-detection-based approach, we are able to detect activities in each tubelet by focusing on the actor situated in its foreground. To accurately detect the activities of a tubelet’s actor we take into account the interactions with other detected actors and objects within the tubelet. We encode such interactions with a dynamic visual spatio-temporal graph and process it with a Graph Neural Network that yields context-aware actor representations. We validate our activity detection framework on the MEVA (Multiview Extended Video with Activities) dataset and the ActEV 2021 Sequestered Data Leaderboard and demonstrate its effectiveness in terms of speed and performance.

Activity recognition in children and infants is important in applications such as safety monitoring, behavior assessment, and child-robot interaction, among others. However, it differs from activity recognition in adults not only because body poses and proportions are different, but also because of the way in which actions are performed. This paper addresses the problem of infant action classification in challenging conditions. The actions are performed in a pediatric rehabilitation environment in which not only infants but also robots and adults are present, with the infant being one of the smallest actors in the scene. We propose a multiview action classification system based on Faster R-CNN and LSTM networks, which fuses information from different views by using learnable fusion coefficients derived from detection confidence scores. The proposed system is view-independent, learns features that are close to view-invariant, and can handle new or missing views at test time. Our approach outperforms the state-of-the-art baseline model for a small dataset (2 subjects, 10-24 months old) by 11.4% in terms of average classification accuracy in four classes (crawl, sit, stand and walk). Moreover, experiments in an extended dataset (6 subjects, 8-24 months old) show that the proposed fusion strategy outperforms all the alternative fusion methods studied.

Background There is a lack of early (infant) mobility rehabilitation approaches that incorporate natural and complex environments and have the potential to concurrently advance motor, cognitive, and social development. The Grounded Early Adaptive Rehabilitation (GEAR) system is a pediatric learning environment designed to provide motor interventions that are grounded in social theory and can be applied in early life. Within a perceptively complex and behaviorally natural setting, GEAR utilizes novel body-weight support technology and socially-assistive robots to both ease and encourage mobility in young children through play-based, child-robot interaction. This methodology article reports on the development and integration of the different system components and presents preliminary evidence on the feasibility of the system. Methods GEAR consists of the physical and cyber components. The physical component includes the playground equipment to enrich the environment, an open-area body weight support (BWS) device to assist children by partially counter-acting gravity, two mobile robots to engage children into motor activity through social interaction, and a synchronized camera network to monitor the sessions. The cyber component consists of the interface to collect human movement and video data, the algorithms to identify the children’s actions from the video stream, and the behavioral models for the child-robot interaction that suggest the most appropriate robot action in support of given motor training goals for the child. The feasibility of both components was assessed via preliminary testing. Three very young children (with and without Down syndrome) used the system in eight sessions within a 4-week period. Results All subjects completed the 8-session protocol, participated in all tasks involving the selected objects of the enriched environment, used the BWS device and interacted with the robots in all eight sessions. Action classification algorithms to identify early child behaviors in a complex naturalistic setting were tested and validated using the video data. Decision making algorithms specific to the type of interactions seen in the GEAR system were developed to be used for robot automation. Conclusions Preliminary results from this study support the feasibility of both the physical and cyber components of the GEAR system and demonstrate its potential for use in future studies to assess the effects on the co-development of the motor, cognitive, and social systems of very young children with mobility challenges.

Events in natural videos typically arise from spatio-temporal interactions between actors and objects and involve multiple co-occurring activities and object classes. To capture this rich visual and semantic context, we propose using two graphs: (1) an attributed spatio-temporal visual graph whose nodes correspond to actors and objects and whose edges encode different types of interactions, and (2) a symbolic graph that models semantic relationships. We further propose a graph neural network for refining the representations of actors, objects and their interactions on the resulting hybrid graph. Our model goes beyond current approaches that assume nodes and edges are of the same type, operate on graphs with fixed edge weights and do not use a symbolic graph. In particular, our framework: a) has specialized attention-based message functions for different node and edge types; b) uses visual edge features; c) integrates visual evidence with label relationships; and d) performs global reasoning in the semantic space. Experiments on challenging video understanding tasks, such as temporal action localization on the Charades dataset, show that the proposed method leads to state-of-the-art performance.

Fine-grained action segmentation and recognition is an important yet challenging task. Given a long, untrimmed sequence of kinematic data, the task is to classify the action at each time frame and segment the time series into the correct sequence of actions. In this paper, we propose a novel framework that combines a temporal Conditional Random Field (CRF) model with a powerful frame-level representation based on discriminative sparse coding. We introduce an end-to-end algorithm for jointly learning the weights of the CRF model, which include action classification and action transition costs, as well as an overcomplete dictionary of mid-level action primitives. This results in a CRF model that is driven by sparse coding features obtained using a discriminative dictionary that is shared among different actions and adapted to the task of structured output learning. We evaluate our method on three surgical tasks using kinematic data from the JIGSAWS dataset, as well as on a food preparation task using accelerometer data from the 50 Salads dataset. Our results show that the proposed method performs on par or better than state-of-the-art methods.

We propose a new approach to action classification in video, which uses deep appearance and motion features extracted from spatio-temporal volumes defined along body part trajectories to learn mid-level classifiers called deep moving poselets. A deep moving poselet is a classifier that captures a characteristic body part configuration, with a specific appearance and undergoing a specific movement. By having this mid-level representation of a body part be shared across action classes and by learning it jointly with action classifiers, we obtain a representation that is interpretable, shared and discriminative. In addition, by using sparsity-inducing norms to regularize action classifiers, we can reduce the number of deep moving poselets used by each class without hurting performance. Experiments show that the proposed method achieves state-of-the-art performance on the popular and challenging sub-JHMDB and MSR Daily Activity datasets.

Action segmentation (segmenting a continuous sequence of motion data into a set of actions) has a wide range of applications and plays a role in many problems in computer vision. We look at subspace clustering as an unsupervised approach for this task. Classical subspace clustering methods uncover relationships within the data by learning codes for the samples (i.e. frames), but in this process these methods do not consider the temporal dependency of nearby samples in motion data. In this paper, we propose two subspace clustering methods with temporal regularization terms: Temporal Sparse Subspace Clustering Laplacian Regularization (TSSC-LR) and Temporal Sparse Subspace Clustering Linear Combination (TSSC-LC). TSSC-LR encourages similar codings of samples within a window and TSSC-LC enforces codings within a window to be linear combinations of each other. Efficient ADMM algorithms are proposed for each method. Experiments against state-of-the-art methods on three action datasets demonstrate the effectiveness of the two proposed methods.

Within the context of assistive robotics we develop an intelligent interface that provides multimodal sensory processing capabilities for human action recognition. Human action is considered in multimodal terms, containing inputs such as audio from microphone arrays, and visual inputs from high definition and depth cameras. Exploring state-of-the-art approaches from automatic speech recognition, and visual action recognition, we multimodally recognize actions and commands. By fusing the unimodal information streams, we obtain the optimum multimodal hypothesis which is to be further exploited by the active mobility assistance robot in the framework of the MOBOT EU research project. Evidence from recognition experiments shows that by integrating multiple sensors and modalities, we increase multimodal recognition performance in the newly acquired challenging dataset involving elderly people while interacting with the assistive robot.

We present a novel video representation for human action recognition by considering temporal sequences of visual words. Based on state-of-the-art dense trajectories, we introduce temporal bundles of dominant, that is most frequent, visual words. These are employed to construct a complementary action representation of ordered dominant visual word sequences, that additionally incorporates fine grained temporal information. We exploit the introduced temporal information by applying local sub-sequence alignment that quantifies the similarity between sequences. This facilitates the fusion of our representation with the bag-of-visual-words (BoVW) representation. Our approach incorporates sequential temporal structure and results in a low-dimensional representation compared to the BoVW, while still yielding a descent result when combined with it. Experiments on the KTH, Hollywood2 and the challenging HMDB51 datasets show that the proposed framework is complementary to the BoVW representation, which discards temporal order.

Query processing systems typically rely on histograms, data structures that approximate data distribution, in order to optimize query execution. Histograms can be constructed by scanning the database tables and aggregating the values of the attributes in the table, or, more efficiently, progressively refined by analysing query results. Most of the relevant literature focuses on histograms of numerical data, exploiting the natural concept of a numerical range as an estimator of the volume of data that falls within the range. This, however, leaves Semantic Web data outside the scope of the histograms literature, as its most prominent datatype, the URI, does not offer itself to defining such ranges. This article first establishes a framework that formalises histograms over arbitrary data types and provides a formalism for specifying value ranges for different datatypes. This makes explicit the properties that ranges are required to have, so that histogram refinement algorithms are applicable. We demonstrate that our framework subsumes histograms over numerical data as a special case by using to formulate the state-of-the-art in numerical histograms. We then proceed to use the Jaro-Winkler metric to define URI ranges by exploiting the hierarchical nature of URI strings. This greatly extends the state of the art, where strings are treated as categorical data that can only be described by enumeration. We then present the open-source STRHist system that implements these ideas. We finally present empirical evaluation results using STRHist over a real dataset and query workload extracted from AGRIS, the most popular and widely used bibliographic database on agricultural research and technology.