3D content generation has recently attracted significant research interest, driven by its critical applications in VR/AR and embodied AI. In this work, we tackle the challenging task of synthesizing multiple 3D assets within a single scene image. Concretely, our contributions are fourfold: (i) we present SceneGen, a novel framework that takes a scene image and corresponding object masks as input, simultaneously producing multiple 3D assets with geometry and texture. Notably, SceneGen operates with no need for extra optimization or asset retrieval; (ii) we introduce a novel feature aggregation module that integrates local and global scene information from visual and geometric encoders within the feature extraction module. Coupled with a position head, this enables the generation of 3D assets and their relative spatial positions in a single feedforward pass; (iii) we demonstrate SceneGen's direct extensibility to multi-image input scenarios. Despite being trained solely on single-image inputs, our architecture yields improved generation performance when multiple images are provided; and (iv) extensive quantitative and qualitative evaluations confirm the efficiency and robustness of our approach. We believe this paradigm offers a novel solution for high-quality 3D content generation, potentially advancing its practical applications in downstream tasks. The code and model will be publicly available at: https://mengmouxu.github.io/SceneGen. “Everything you can imagine is real.” —Pablo Picasso

Representation learning serves as a foundational component of medical vision-language models (MVLMs), enabling cross-modal alignment, semantic consistency, and enhanced generalization capabilities for downstream tasks. As generalist models rapidly evolve, there is a pressing need to unify diverse downstream tasks, such as diagnosis, segmentation, report generation, and multiple choice within a cohesive framework, demanding more efficient and versatile visual representation learning. However, current MVLMs predominately follow CLIP-style vision pretraining, failing to leverage heterogeneous data resources with multi-dimensional imaging and diverse annotation forms. And there lacks systematic analysis of efficient vision encoder design across varied downstream applications, including diagnosis, segmentation, and text generation tasks, particularly for volumetric imaging like Computed Tomography (CT). Besides, current MVLMs exhibit constrained voxel-level capabilities, lacking effective multi-task instruction tuning framework capable of achieving robust performance across various downstream tasks. To address these challenges, we propose CTInstruct, a novel MVLM employing a hybrid ResNet-ViT encoder with multi-granular vision-language pretraining for efficient heterogeneous data modeling, and unified instruction tuning that jointly optimizes discriminative, generative, and voxel-level reasoning for volumetric medical imaging. CTInstruct achieves SOTA performance across 8 CT benchmarks, setting a new standard for data-efficient multimodal learning in medical imaging.

In this paper, we consider the problem of long-term point tracking, which requires consistent identification of points across video frames under significant appearance changes, motion, and occlusion. We target the online setting, i.e., tracking points frame-by-frame, making it suitable for real-time and streaming applications. We extend our prior model Track-On into Track-On2, a simple and efficient transformer-based model for online long-term tracking. Track-On2 improves both performance and efficiency through architectural refinements, more effective use of memory, and improved synthetic training strategies. Unlike prior approaches that rely on full-sequence access or iterative updates, our model processes frames causally and maintains temporal coherence via a memory mechanism, which is key to handling drift and occlusions without requiring future frames. At inference, we perform coarse patch-level classification followed by refinement. Beyond architecture, we systematically study synthetic training setups and their impact on memory behavior, showing how they shape temporal robustness over long sequences. Through comprehensive experiments, Track-On2 achieves state-of-the-art results across five synthetic and real-world benchmarks, surpassing prior online trackers and even strong offline methods that exploit bidirectional context. These results highlight the effectiveness of causal, memory-based architectures trained purely on synthetic data as scalable solutions for real-world point tracking.

Radiologists face increasing workloads that make accurate and timely report generation both critical and challenging. This paper presents a novel system for grounded automatic brain MRI report generation, with contributions in three key areas: First, we release RadGenome-Brain MRI, a benchmark dataset featuring multi-modal scans, expert-annotated abnormality masks, and radiology reports with region-level grounding to support fine-grained, explainable report generation. Second, we propose AutoRG-Brain, the first brain MRI report generation framework that combines automatic anomaly segmentation with a visual prompting-based language model to produce structured, anatomically grounded findings. Third, we conduct extensive quantitative and expert evaluations across segmentation and reporting tasks, and demonstrate in real clinical settings that our system significantly enhances junior radiologists' ability to detect subtle abnormalities and compose high-quality reports, narrowing the gap with senior doctors. All code, models, and datasets will be publicly released to facilitate future research and development.

This paper considers the problem of Multi-Hop Video Question Answering (MH-VidQA) in long-form egocentric videos. This task not only requires to answer visual questions, but also to localize multiple relevant time intervals within the video as visual evidences. We develop an automated pipeline to create multi-hop question-answering pairs with associated temporal evidence, enabling to construct a large-scale dataset for instruction-tuning. To monitor the progress of this new task, we further curate a high-quality benchmark, MULTIHOP-EGOQA, with careful manual verification and refinement. Experimental results reveal that existing multimodal systems exhibit inadequate multi-hop grounding and reasoning abilities, resulting in unsatisfactory performance. We then propose a novel architecture, termed as Grounding Scattered Evidence with Large Language Model (GeLM), that enhances multi-modal large language models by incorporating a grounding module to retrieve temporal evidence from videos using flexible grounding tokens. Trained on our visual instruction-tuning data, GeLM demonstrates improved multi-hop grounding and reasoning capabilities, setting a baseline for this new task. Furthermore, when trained on third-person view videos, the same architecture also achieves state-of-the-art performance on the single-hop VidQA benchmark, ActivityNet-RTL, demonstrating its effectiveness.

The objective in this paper is to improve the per-formance of text-to-image retrieval. To this end, we introduce a new framework that can boost the performance of large-scale pre-trained vision-language models, so that they can be used for text-to-image re-ranking. The approach, Enhanced Language-Image Pre-training (ELIP), uses the text query, via a simple MLP mapping network, to predict a set of visual prompts to condition the ViT image encoding. ELIP can easily be applied to the commonly used CLIP, SigLIP and BLIP-2 networks. On the evaluation side, we set up two new out-of-distribution (OOD) benchmarks, Occluded COCO and ImageNet-R, to assess the zero-shot generalisation of the models to different domains. The results demonstrate that ELIP significantly boosts CLIP/SigLIP/SigLIP-2 text-to-image retrieval performance and outperforms BLIP-2 on several benchmarks, as well as providing an easy means to adapt to OOD datasets.

With the rapid advancement of multimodal information retrieval, increasingly complex retrieval tasks have emerged. Existing methods predominately rely on task-specific fine-tuning of vision-language models, often those trained with image-text contrastive learning. In this paper, we explore the possibility of re-purposing generative Large Multimodal Models (LMMs) for retrieval. This approach enables unifying all retrieval tasks under the same formulation and, more importantly, allows for extrapolation towards unseen retrieval tasks without additional training. Our contributions can be summarised in the following aspects: (i) We introduce LamRA, a versatile framework designed to empower LMMs with sophisticated retrieval and reranking capabilities. (ii) For retrieval, we adopt a two-stage training strategy comprising language-only pre-training and multimodal instruction tuning to progressively enhance LMM’s retrieval performance. (iii) For reranking, we employ joint training of both pointwise and listwise reranking, offering two distinct ways to further boost the retrieval performance. (iv) Extensive experiments underscore the efficacy of our method in handling more than ten retrieval tasks, demonstrating robust performance in both supervised and zero-shot settings, including scenarios involving previously unseen retrieval tasks. Project page: https://code-kunkun.github.io/LamRA/.

As a globally celebrated sport, soccer has attracted widespread interest from fans all over the world. This paper aims to develop a comprehensive multi-modal framework for soccer video understanding. Specifically, we make the following contributions in this paper: (i) we introduce SoccerReplay-1988, the largest multi-modal soccer dataset to date, featuring videos and detailed annotations from 1,988 complete matches, with an automated annotation pipeline; (ii) we present an advanced soccer-specific visual encoder, MatchVision, which leverages spatiotemporal information across soccer videos and excels in various downstream tasks; (iii) we conduct extensive experiments and ablation studies on event classification, commentary generation, and multi-view foul recognition. MatchVision demonstrates state-of-the-art performance on all of them, substantially outperforming existing models, which highlights the superiority of our proposed data and model. We believe that this work will offer a standard paradigm for sports understanding research.“Football is one of the world’s best means of communication. It is impartial, apolitical, and universal.”—— Franz Beckenbauer (1945 - 2024)

This paper tackles the problem of video question answering (VideoQA), a task that often requires multi-step reasoning and a profound understanding of spatial-temporal dynamics. While large video-language models perform well on benchmarks, they often lack explainability and spatial temporal grounding. In this paper, we propose Agent-of-Thoughts Distillation (AoTD), a method that enhances models by incorporating automatically generated Chain- of-Thoughts (CoTs) into the instruction-tuning process. Specifically, we leverage an agent-based system to decompose complex questions into sub-tasks, and address them with specialized vision models, the intermediate results are then treated as reasoning chains. We also introduce a verification mechanism using a large language model (LLM) to ensure the reliability of generated CoTs. Extensive experiments demonstrate that AoTD improves the performance on multiple-choice and open-ended benchmarks.

Video Large Language Models (VideoLLMs) have recently demonstrated remarkable progress in general video understanding. However, existing models primarily focus on high-level comprehension and are limited to text-only responses, restricting the flexibility for object-centric, multiround interactions. In this paper, we make three contributions: (i) we address these limitations by introducing a VideoLLM model, capable of performing both object referring for input and grounding for output in video reasoning tasks, i.e., allowing users to interact with videos using both textual and visual prompts; (ii) we propose STOM (Spatial-Temporal Overlay Module), a novel approach that propagates arbitrary visual prompts input at any single timestamp to the remaining frames within a video; (iii) we present VideoInfer, a manually curated object-centric video instruction dataset featuring questionanswering pairs that require reasoning. We conduct comprehensive experiments on VideoInfer and other existing benchmarks across video question answering and referring object segmentation. The results on 12 benchmarks of 6 tasks show that our proposed model consistently outperforms baselines in both video question answering and segmentation, underscoring its robustness in multimodal, object-centric video and image understanding. Project page: https://qirui-chen.github.io/RGA3-release/.

Our objective is the automatic generation of Audio Descriptions (ADs) for edited video material, such as movies and TV series. To achieve this, we propose a two-stage framework that leverages "shots" as the fundamental units of video understanding. This includes extending temporal context to neighbouring shots and incorporating film grammar devices, such as shot scales and thread structures, to guide AD generation. Our method is compatible with both open-source and proprietary Visual-Language Models (VLMs), integrating expert knowledge from add-on modules without requiring additional training of the VLMs. We achieve state-of-the-art performance among all prior training-free approaches and even surpass fine-tuned methods on several benchmarks. To evaluate the quality of predicted ADs, we introduce a new evaluation measure -- an action score -- specifically targeted to assessing this important aspect of AD. Additionally, we propose a novel evaluation protocol that treats automatic frameworks as AD generation assistants and asks them to generate multiple candidate ADs for selection.

Understanding continuous video streams plays a fundamental role in real-time applications including embodied AI and autonomous driving. Unlike offline video understanding, streaming video understanding requires the ability to process video streams frame by frame, preserve historical information, and make low-latency decisions. To address these challenges, our main contributions are three-fold. (i) We develop a novel streaming video backbone, termed as StreamFormer, by incorporating causal temporal attention into a pre-trained vision transformer. This enables efficient streaming video processing while maintaining image representation capability. (ii) To train StreamFormer, we propose to unify diverse spatial-temporal video understanding tasks within a multitask visual-language alignment framework. Hence, StreamFormer learns global semantics, temporal dynamics, and fine-grained spatial relationships simultaneously. (iii) We conduct extensive experiments on online action detection, online video instance segmentation, and video question answering. StreamFormer achieves competitive results while maintaining efficiency, demonstrating its potential for real-time applications.

In this paper, we consider the problem of long-term point tracking, which requires consistent identification of points across multiple frames in a video, despite changes in appearance, lighting, perspective, and occlusions. We target online tracking on a frame-by-frame basis, making it suitable for real-world, streaming scenarios. Specifically, we introduce Track-On, a simple transformer-based model designed for online long-term point tracking. Unlike prior methods that depend on full temporal modeling, our model processes video frames causally without access to future frames, leveraging two memory modules —spatial memory and context memory— to capture temporal information and maintain reliable point tracking over long time horizons. At inference time, it employs patch classification and refinement to identify correspondences and track points with high accuracy. Through extensive experiments, we demonstrate that Track-On sets a new state-of-the-art for online models and delivers superior or competitive results compared to offline approaches on seven datasets, including the TAP-Vid benchmark. Our method offers a robust and scalable solution for real-time tracking in diverse applications. Project page: https://kuis-ai.github.io/track_on

In egocentric video understanding, the motion of hands and objects as well as their interactions play a significant role by nature. However, existing egocentric video representation learning methods mainly focus on aligning video representation with high-level narrations, overlooking the intricate dynamics between hands and objects. In this work, we aim to integrate the modeling of fine-grained hand-object dynamics into the video representation learning process. Since no suitable data is available, we introduce HOD, a novel pipeline employing a hand-object detector and a large language model to generate high-quality narrations with detailed descriptions of hand-object dynamics. To learn these fine-grained dynamics, we propose EgoVideo, a model with a new lightweight motion adapter to capture fine-grained hand-object motion information. Through our co-training strategy, EgoVideo effectively and efficiently leverages the fine-grained hand-object dynamics in the HOD data. Extensive experiments demonstrate that our method achieves state-of-the-art performance across multiple egocentric downstream tasks, including improvements of 6.3% in EK-100 multi-instance retrieval, 5.7% in EK-100 classification, and 16.3% in EGTEA classification in zero-shot settings. Furthermore, our model exhibits robust generalization capabilities in hand-object interaction and robot manipulation tasks.

Generating videos in the first-person perspective has broad application prospects in the field of augmented reality and embodied intelligence. In this work, we explore the cross-view video prediction task, where given an exo-centric video, the first frame of the corresponding ego-centric video, and textual instructions, the goal is to generate future frames of the ego-centric video. Inspired by the notion that hand-object interactions (HOI) in ego-centric videos represent the primary intentions and actions of the current actor, we present EgoExo-Gen that explicitly models the hand-object dynamics for cross-view video prediction. EgoExo-Gen consists of two stages. First, we design a cross-view HOI mask prediction model that anticipates the HOI masks in future ego-frames by modeling the spatio-temporal ego-exo correspondence. Next, we employ a video diffusion model to predict future ego-frames using the first ego-frame and textual instructions, while incorporating the HOI masks as structural guidance to enhance prediction quality. To facilitate training, we develop a fully automated pipeline to generate pseudo HOI masks for both ego- and exo-videos by exploiting vision foundation models. Extensive experiments demonstrate that our proposed EgoExo-Gen achieves better prediction performance compared to previous video prediction models on the public Ego-Exo4D and H2O benchmark datasets, with the HOI masks significantly improving the generation of hands and interactive objects in the ego-centric videos.

With the rapid development of generative models, discerning AI-generated content has evoked increasing attention from both industry and academia. In this paper, we conduct a sanity check on whether the task of AI-generated image detection has been solved. To start with, we present Chameleon dataset, consisting of AI-generated images that are genuinely challenging for human perception. To quantify the generalization of existing methods, we evaluate 9 off-the-shelf AI-generated image detectors on Chameleon dataset. Upon analysis, almost all models misclassify AI-generated images as real ones. Later, we propose AIDE AI-generated Image DEtector with Hybrid Features, which leverages multiple experts to simultaneously extract visual artifacts and noise patterns. Specifically, to capture the high-level semantics, we utilize CLIP to compute the visual embedding. This effectively enables the model to discern AI-generated images based on semantics and contextual information. Secondly, we select the highest and lowest frequency patches in the image, and compute the low-level patchwise features, aiming to detect AI-generated images by low-level artifacts, for example, noise patterns, anti-aliasing effects. While evaluating on existing benchmarks, for example, AIGCDetectBenchmark and GenImage, AIDE achieves +3.5% and +4.6% improvements to state-of-the-art methods, and on our proposed challenging Chameleon benchmarks, it also achieves promising results, despite the problem of detecting AI-generated images remains far from being solved.

Animated movies are captivating for their unique character designs and imaginative storytelling, yet they pose significant challenges for existing recognition systems. Unlike the consistent visual patterns detected by conventional face recognition methods, animated characters exhibit extreme diversity in their appearance, motion, and deformation. In this work, we propose an audio-visual pipeline to enable automatic and robust animated character recognition, and thereby enhance character-centric understanding of animated movies. Central to our approach is the automatic construction of an audio-visual character bank from online sources. This bank contains both visual exemplars and voice (audio) samples for each character, enabling subsequent multi-modal character recognition despite long-tailed appearance distributions. Building on accurate character recognition, we explore two downstream applications: Audio Description (AD) generation for visually impaired audiences, and character-aware subtitling for the hearing impaired. To support research in this domain, we introduce CMD-AM, a new dataset of 75 animated movies with comprehensive annotations. Our character-centric pipeline demonstrates significant improvements in both accessibility and narrative comprehension for animated content over prior face-detection-based approaches. For the code and dataset, visit https://www.robots.ox.ac.uk/~vgg/research/animated_ad/.

This paper presents a computational model for universal video temporal grounding, which accurately localizes temporal moments in videos based on natural language queries (e.g., questions or descriptions). Unlike existing methods that are often limited to specific video domains or durations, we propose **UniTime**, a robust and universal video grounding model leveraging the strong vision-language understanding capabilities of generative Multi-modal Large Language Models (MLLMs). Our model effectively handles videos of diverse views, genres, and lengths while comprehending complex language queries. The key contributions include: (i) We consider steering strong MLLMs for temporal grounding in videos. To enable precise timestamp outputs, we incorporate temporal information by interleaving timestamp tokens with video tokens. (ii) By training the model to handle videos with different input granularities through adaptive frame scaling, our approach achieves robust temporal grounding for both short and long videos. (iii) Comprehensive experiments show that UniTime outperforms state-of-the-art approaches in both zero-shot and dataset-specific finetuned settings across five public temporal grounding benchmarks. (iv) When employed as a preliminary moment retriever for long-form video question-answering (VideoQA), UniTime significantly improves VideoQA accuracy, highlighting its value for complex video understanding tasks.

Abstract We have witnessed significant progress in human-object interaction (HOI) detection. However, relying solely on mAP (mean Average Precision) scores as a summary metric does not provide sufficient insight into the nuances of model performance ( e.g. , why one model outperforms another), which can hinder further innovation in this field. To address this issue, we introduce a diagnosis toolbox in this paper to offer a detailed quantitative breakdown of HOI detection models, inspired by the success of object detection diagnosis tools. We first conduct a holistic investigation into the HOI detection pipeline. By defining a set of errors and using oracles to fix each one, we quantitatively analyze the significance of different errors based on the mAP improvement gained from fixing them. Next, we explore the two key sub-tasks of HOI detection: human-object pair localization and interaction classification. For the pair localization task, we compute the coverage of ground-truth human-object pairs and assess the noisiness of the localization results. For the classification task, we measure a model’s ability to distinguish between positive and negative detection results and to classify actual interactions when human-object pairs are correctly localized. We analyze eight state-of-the-art HOI detection models, providing valuable diagnostic insights to guide future research. For instance, our diagnosis reveals that the state-of-the-art model RLIPv2 outperforms others primarily due to its significant improvement in multi-label interaction classification accuracy. Our toolbox is applicable across various methods and datasets and is available at https://neu-vi.github.io/Diag-HOI/ .

The current variants of the Segment Anything Model (SAM), which include the original SAM and Medical SAM, still lack the capability to produce sufficiently accurate segmentation for medical images. In medical imaging contexts, it is not uncommon for human experts to rectify segmentations of specific test samples after SAM generates its segmentation predictions. These rectifications typically entail manual or semi-manual corrections employing state-of-the-art annotation tools. Motivated by this process, we introduce a novel approach that leverages the advantages of online machine learning to enhance Segment Anything (SA) during test time. We employ rectified annotations to perform online learning, with the aim of improving the segmentation quality of SA on medical images. To ensure the effectiveness and efficiency of online learning when integrated with large-scale vision models like SAM, we propose a new method called Auxiliary Online Learning (AuxOL), which entails adaptive online-batch and adaptive segmentation fusion. Experiments conducted on eight datasets covering four medical imaging modalities validate the effectiveness of the proposed method. Our work proposes and validates a new, practical, and effective approach for enhancing SA on downstream segmentation tasks (e.g., medical image segmentation). The code is publicly available at https://sam-auxol.github.io/AuxOL/.

Existing approaches to video understanding, mainly designed for short videos from a third-person perspective, are limited in their applicability in certain fields, such as robotics. In this paper, we delve into open-ended question-answering (QA) in long, egocentric videos, which allows individuals or robots to inquire about their own past visual ex-periences. This task presents unique challenges, including the complexity of temporally grounding queries within extensive video content, the high resource demands for precise data annotation, and the inherent difficulty of evaluating open-ended answers due to their ambiguous nature. Our proposed approach tackles these challenges by (i) integrating query grounding and answering within a unified model to reduce error propagation; (ii) employing large language models for efficient and scalable data synthesis; and (iii) introducing a close-ended QA task for evaluation, to man-age answer ambiguity. Extensive experiments demonstrate the effectiveness of our method, which also achieves state-of-the-art performance on the QAEgo4D and Ego4D-NLQ benchmarks. Code, data, and models are open-sourced1.

In this paper, we present a novel paradigm to enhance the ability of object detector, e.g., expanding categories or improving detection performance, by training on syn-thetic dataset generated from diffusion models. Specifically, we integrate an instance-level grounding head into a pre-trained, generative diffusion model, to augment it with the ability of localising instances in the generated images. The grounding head is trained to align the text embedding of category names with the regional visual feature of the diffusion model, using supervision from an off-the-shelf object detector, and a novel self-training scheme on (novel) categories not covered by the detector. We conduct thorough experiments to show that, this enhanced version of diffusion model, termed as InstaGen, can serve as a data synthe-sizer, to enhance object detectors by training on its generated samples, demonstrating superior performance over existing state-of-the-art methods in open-vocabulary (+4.5 AP) and data-sparse (+ 1. 2 ~ 5.2 AP) scenarios.

Generating Audio Description (AD) for movies is a challenging task that requires fine-grained visual understanding and an awareness of the characters and their names. Currently, visual language models for AD generation are limited by a lack of suitable training data, and also their evaluation is hampered by using performance measures not specialized to the AD domain. In this paper, we make three contributions: (i) We propose two approaches for constructing AD datasets with aligned video data, and build training and evaluation datasets using these. These datasets will be publicly released; (ii) We develop a Q-former-based architecture which ingests raw video and generates AD, using frozen pre-trained visual encoders and large language models; and (iii) We provide new evaluation metrics to benchmark AD quality that are well matched to human performance. Taken together, we improve the state of the art on AD generation.

Generative models have recently exhibited exceptional capabilities in text-to-image generation, but still struggle to generate image sequences coherently. In this work, we focus on a novel, yet challenging task of generating a co-herent image sequence based on a given storyline, denoted as open-ended visual storytelling. We make the following three contributions: (i) to fulfill the task of visual sto-rytelling, we propose a learning-based auto-regressive im-age generation model, termed as Story Gen, with a novel vision-language context module, that enables to generate the current frame by conditioning on the corresponding text prompt and preceding image-caption pairs; (ii) to ad-dress the data shortage of visual storytelling, we collect paired image-text sequences by sourcing from online videos and open-source E-books, establishing processing pipeline for constructing a large-scale dataset with diverse characters, storylines, and artistic styles, named StorySalon; (iii) Quantitative experiments and human evaluations have vali-dated the superiority of our StoryGen, where we show it can generalize to unseen characters without any optimization, and generate image sequences with coherent content and consistent character. Code, dataset, and models are avail-able at https://haoningwu3639.github.io/StoryGen_Webpage/. “Mirror mirror on the wall, who's the fairest of them all?” -Grimms' Fairy Tales

Understanding human actions from videos offirst-person view poses significant challenges. Most prior approaches explore representation learning on egocentric videos only, while overlooking the potential benefit of exploiting existing large-scale third-person videos. In this paper, (1) we develop EgoInstructor, a retrieval-augmented multimodal captioning model that automatically retrieves semantically relevant third-person instructional videos to enhance the video captioning of egocentric videos, (2) for training the cross-view retrieval module, we devise an au-tomatic pipeline to discover ego-exo video pairs from distinct large-scale egocentric and exocentric datasets, (3) we train the cross-view retrieval module with a novel EgoEx-oNCE loss that pulls egocentric and exocentric video features closer, by aligning them to shared text features that describe similar actions, (4) through extensive experiments, our cross-view retrieval module demonstrates superior performance across seven benchmarks. Regarding egocen-tric video captioning, EgoInstructor exhibits significant improvements by leveraging third-person videos as references.

This paper studies amodal image segmentation: predicting entire object segmentation masks including both visible and invisible (occluded) parts. In previous work, the amodal segmentation ground truth on real images is usually pre-dicted by manual annotaton and thus is subjective. In contrast, we use 3D data to establish an automatic pipeline to determine authentic ground truth amodal masks for partially occluded objects in real images. This pipeline is used to construct an amodal completion evaluation benchmark, MP3D-Amodal, consisting of a variety of object categories and la-bels. To better handle the amodal completion task in the wild, we explore two architecture variants: a two-stage model that first infers the occluder, followed by amodal mask completion; and a one-stage model that exploits the representation power of Stable Diffusion for amodal segmentation across many categories. Without bells and whistles, our method achieves a new state-of-the-art performance on Amodal segmentation datasets that cover a large variety of objects, in-cluding COCOA and our new MP3D-Amodal dataset. The dataset, model, and code are available at https://www.robots.ox.ac.uk/~vgg/research/amodal/

Soccer is a globally popular sport with a vast audience, in this paper, we consider constructing an automatic soccer game commentary model to improve the audiences’ viewing experience. In general, we make the following contributions: First, observing the prevalent video-text misalignment in existing datasets, we manually annotate timestamps for 49 matches, establishing a more robust benchmark for soccer game commentary generation, termed as SN-Caption-test-align; Second, we propose a multi-modal temporal alignment pipeline to automatically correct and filter the existing dataset at scale, creating a higher-quality soccer game commentary dataset for training, denoted as MatchTime; Third, based on our curated dataset, we train an automatic commentary generation model, named MatchVoice. Extensive experiments and ablation studies have demonstrated the effectiveness of our alignment pipeline, and training model on the curated datasets achieves state-of-the-art performance for commentary generation, showcasing that better alignment can lead to significant performance improvements in downstream tasks.

This paper introduces a novel, entity-aware metric, termed as Radiological Report (Text) Evaluation (RaTEScore), to assess the quality of medical reports generated by AI models.RaTEScore emphasizes crucial medical entities, such as diagnostic outcomes and anatomical details.Moreover, it is robust against medical synonyms and sensitive to negation expressions.Technically, we developed a comprehensive medical NER dataset, RaTE-NER, and trained an NER model specifically for this purpose.This model enables the decomposition of complex radiological reports into constituent medical entities.The metric itself is derived by comparing the similarity of entity embeddings, obtained from a language model, based on their types and relevance to clinical significance.Our evaluations demonstrate that RaTEScore aligns more closely with human preference than existing metrics, validated both on established public benchmarks and our newly proposed RaTE-Eval benchmark.

Our objective is audio-visual synchronization with a focus on ‘in-the-wild’ videos, such as those on YouTube, where synchronization cues can be sparse. Our contributions include a novel audio-visual synchronization model, and training that decouples feature extraction from synchronization modelling through multi-modal segment-level contrastive pre-training. This approach achieves state-of-the-art performance in both dense and sparse settings. We also extend synchronization model training to AudioSet a million-scale ‘in-the-wild’ dataset, investigate evidence attribution techniques for interpretability, and explore a new capability for synchronization models: audio-visual synchronizability. robots.ox.ac.uk/~vgg/research/synchformer

Recently, the AI community has made significant strides in developing powerful foundation models, driven by large-scale multimodal datasets. However, for audio representation learning, existing datasets suffer from limitations in the following aspects: insufficient volume, simplistic content, and arduous collection procedures. To establish an audio dataset with high-quality captions, we propose an innovative, automatic approach leveraging multimodal inputs, such as video frames, audio streams. Specifically, we construct a large-scale, high-quality, audio-language dataset, named as Auto-ACD, comprising over 1.5M audio-text pairs. We exploit a series of pre-trained models or APIs, to determine audio-visual synchronisation, generate image captions, object detection, or audio tags for specific videos. Subsequently, we employ LLM to paraphrase a congruent caption for each audio, guided by the extracted multi-modality clues. To demonstrate the effectiveness of the proposed dataset, we train widely used models on our dataset and show performance improvement on various downstream tasks, for example, audio-language retrieval, audio captioning, zero-shot classification. In addition, we establish a novel benchmark with environmental information and provide a benchmark for audio-text tasks.

Our objective in this paper is to probe large vision models to determine to what extent they ‘understand’ different physical properties of the 3D scene depicted in an image. To this end, we make the following contributions: (i) We introduce a general and lightweight protocol to evaluate whether features of an off-the-shelf large vision model encode a number of physical ‘properties’ of the 3D scene, by training discriminative classifiers on the features for these properties. The probes are applied on datasets of real images with annotations for the property. (ii) We apply this protocol to properties covering scene geometry, scene material, support relations, lighting, and view-dependent measures, and large vision models including CLIP, DINOv1, DINOv2, VQGAN, Stable Diffusion. (iii) We find that features from Stable Diffusion and DINOv2 are good for discriminative learning of a number of properties, including scene geometry, support relations, shadows and depth, but less performant for occlusion and material, while outperforming DINOv1, CLIP and VQGAN for all properties. (iv) It is observed that different time steps of Stable Diffusion features, as well as different transformer layers of DINO/CLIP/VQGAN, are good at different properties, unlocking potential applications of 3D physical understanding.

The objective of Audio-Visual Segmentation (AVS) is to localise the sounding objects within visual scenes by accurately predicting pixel-wise segmentation masks. To tackle the task, it involves a comprehensive consideration of both the data and model aspects. In this paper, first, we initiate a novel pipeline for generating artificial data for the AVS task without extra manual annotations. We leverage existing image segmentation and audio datasets and match the image-mask pairs with its corresponding audio samples using category labels in segmentation datasets, that allows us to effortlessly compose (image, audio, mask) triplets for training AVS models. The pipeline is annotation-free and scalable to cover a large number of categories. Additionally, we introduce a lightweight model SAMA-AVS which adapts the pre-trained segment anything model (SAM) to the AVS task. By introducing only a small number of trainable parameters with adapters, the proposed model can effectively achieve adequate audio-visual fusion and interaction in the encoding stage with vast majority of parameters fixed. We conduct extensive experiments, and the results show our proposed model remarkably surpasses other competing methods. Moreover, by using the proposed model pretrained with our synthetic data, the performance on real AVSBench data is further improved, achieving 83.17 mIoU on S4 subset and 66.95 mIoU on MS3 set. The project page is https://jinxiang-liu.github.io/anno-free-AVS/.

OBJECTIVE: Recently, large language models (LLMs) have showcased remarkable capabilities in natural language understanding. While demonstrating proficiency in everyday conversations and question-answering (QA) situations, these models frequently struggle in domains that require precision, such as medical applications, due to their lack of domain-specific knowledge. In this article, we describe the procedure for building a powerful, open-source language model specifically designed for medicine applications, termed as PMC-LLaMA. MATERIALS AND METHODS: We adapt a general-purpose LLM toward the medical domain, involving data-centric knowledge injection through the integration of 4.8M biomedical academic papers and 30K medical textbooks, as well as comprehensive domain-specific instruction fine-tuning, encompassing medical QA, rationale for reasoning, and conversational dialogues with 202M tokens. RESULTS: While evaluating various public medical QA benchmarks and manual rating, our lightweight PMC-LLaMA, which consists of only 13B parameters, exhibits superior performance, even surpassing ChatGPT. All models, codes, and datasets for instruction tuning will be released to the research community. DISCUSSION: Our contributions are 3-fold: (1) we build up an open-source LLM toward the medical domain. We believe the proposed PMC-LLaMA model can promote further development of foundation models in medicine, serving as a medical trainable basic generative language backbone; (2) we conduct thorough ablation studies to demonstrate the effectiveness of each proposed component, demonstrating how different training data and model scales affect medical LLMs; (3) we contribute a large-scale, comprehensive dataset for instruction tuning. CONCLUSION: In this article, we systematically investigate the process of building up an open-source medical-specific LLM, PMC-LLaMA.

Three-dimensional (3D) ultrasound imaging has contributed to our understanding of fetal developmental processes by providing rich contextual information of the inherently 3D anatomies. However, its use is limited in clinical settings, due to the high purchasing costs and limited diagnostic practicality. Freehand 2D ultrasound imaging, in contrast, is routinely used in standard obstetric exams, but inherently lacks a 3D representation of the anatomies, which limits its potential for more advanced assessment. Such full representations are challenging to recover even with external tracking devices due to internal fetal movement which is independent from the operator-led trajectory of the probe. Capitalizing on the flexibility offered by freehand 2D ultrasound acquisition, we propose ImplicitVol to reconstruct 3D volumes from non-sensor-tracked 2D ultrasound sweeps. Conventionally, reconstructions are performed on a discrete voxel grid. We, however, employ a deep neural network to represent, for the first time, the reconstructed volume as an implicit function. Specifically, ImplicitVol takes a set of 2D images as input, predicts their locations in 3D space, jointly refines the inferred locations, and learns a full volumetric reconstruction. When testing natively-acquired and volume-sampled 2D ultrasound video sequences collected from different manufacturers, the 3D volumes reconstructed by ImplicitVol show significantly better visual and semantic quality than the existing interpolation-based reconstruction approaches. The inherent continuity of implicit representation also enables ImplicitVol to reconstruct the volume to arbitrarily high resolutions. As formulated, ImplicitVol has the potential to integrate seamlessly into the clinical workflow, while providing richer information for diagnosis and evaluation of the developing brain.

In this paper, we consider the problem of simultaneously detecting objects and inferring their visual attributes in an image, even for those with no manual annotations provided at the training stage, resembling an open-vocabulary scenario. To achieve this goal, we make the following contributions: (i) we start with a naive two-stage approach for open-vocabulary object detection and attribute classification, termed CLIP-Attr. The candidate objects are first proposed with an offline RPN and later classified for semantic category and attributes; (ii) we combine all available datasets and train with a federated strategy to finetune the CLIP model, aligning the visual representation with attributes, additionally, we investigate the efficacy of leveraging freely available online image-caption pairs under weakly supervised learning; (iii) in pursuit of efficiency, we train a Faster-RCNN type model end-to-end with knowledge distillation, that performs class-agnostic object proposals and classification on semantic categories and attributes with classifiers generated from a text encoder; Finally, (iv) we conduct extensive experiments on VAW, MS-COCO, LSA, and OVAD datasets, and show that recognition of semantic category and attributes is complementary for visual scene understanding, i.e., jointly training object detection and attributes prediction largely outperform existing approaches that treat the two tasks independently, demonstrating strong generalization ability to novel attributes and categories.

The objective of this paper is an automatic Audio Description (AD) model that ingests movies and outputs AD in text form. Generating high-quality movie AD is challenging due to the dependency of the descriptions on context, and the limited amount of training data available. In this work, we leverage the power of pretrained foundation models, such as GPT and CLIP, and only train a mapping network that bridges the two models for visually-conditioned text generation. In order to obtain high-quality AD, we make the following four contributions: (i) we incorporate context from the movie clip, AD from previous clips, as well as the subtitles; (ii) we address the lack of training data by pretraining on large-scale datasets, where visual or contextual information is unavailable, e.g. text-only AD without movies or visual captioning datasets without context; (iii) we improve on the currently available AD datasets, by removing label noise in the MAD dataset, and adding character naming information; and (iv) we obtain strong results on the movie AD task compared with previous methods.

Camera-only 3D detection provides an economical solution with a simple configuration for localizing objects in 3D space compared to LiDAR-based detection systems. However, a major challenge lies in precise depth estimation due to the lack of direct 3D measurements in the input. Many previous methods attempt to improve depth estimation through network designs, e.g., deformable layers and larger receptive fields. This work proposes an orthogonal direction, improving the camera-only 3D detection by introducing multi-agent collaborations. Our proposed collaborative camera-only 3D detection (CoCa3D) enables agents to share complementary information with each other through communication. Meanwhile, we optimize communication efficiency by selecting the most informative cues. The shared messages from multiple view-points disambiguate the single-agent estimated depth and complement the occluded and long-range regions in the single-agent view. We evaluate CoCa3D in one real-world dataset and two new simulation datasets. Results show that CoCa3D improves previous SOTA performances by 44.21% on DAIR-V2X, 30.60% on OPV2V+, 12.59% on CoPerception-UAVs+ for AP@70. Our preliminary results show a potential that with sufficient collaboration, the camera might overtake LiDAR in some practical scenarios. We released the dataset and code.

This paper considers the problem of open-vocabulary semantic segmentation (OVS), that aims to segment objects of arbitrary classes beyond a pre-defined, closed-set categories. The main contributions are as follows: First, we propose a transformer-based model for OVS, termed as OVSegmentor, which only exploits web-crawled imagetext pairs for pre-training without using any mask annotations. OVSegmentor assembles the image pixels into a set of learnable group tokens via a slotattention based binding module, then aligns the group tokens to corresponding caption embeddings. Second, we propose two proxy tasks for training, namely masked entity completion and cross-image mask consistency. The former aims to infer all masked entities in the caption given group tokens, that enables the model to learn fine-grained alignment between visual groups and text entities. The latter enforces consistent mask predictions between images that contain shared entities, encouraging the model to learn visual invariance. Third, we construct CC4M dataset for pre-training by filtering CC12M with frequently appeared entities, which significantly improves training efficiency. Fourth, we perform zero-shot transfer on four benchmark datasets, PASCAL VOC, PASCAL Context, COCO Object, and ADE20K. OVSegmentor achieves superior results over state-of-the-art approaches on PASCAL VOC using only 3% data (4M vs 134M) for pre-training.

Audio Description (AD) is the task of generating descriptions of visual content, at suitable time intervals, for the benefit of visually impaired audiences. For movies, this presents notable challenges – AD must occur only during existing pauses in dialogue, should refer to characters by name, and ought to aid understanding of the storyline as a whole.To this end, we develop a new model for automatically generating movie AD, given CLIP visual features of the frames, the cast list, and the temporal locations of the speech; addressing all three of the ‘who’, ‘when’, and ‘what’ questions: (i) who – we introduce a character bank consisting of the character’s name, the actor that played the part, and a CLIP feature of their face, for the principal cast of each movie, and demonstrate how this can be used to improve naming in the generated AD; (ii) when – we investigate several models for determining whether an AD should be generated for a time interval or not, based on the visual content of the interval and its neighbours; and (iii) what – we implement a new vision-language model for this task, that can ingest the proposals from the character bank, whilst conditioning on the visual features using cross-attention, and demonstrate how this improves over previous architectures for AD text generation in an apples-to-apples comparison.

Not all camouflages are equally effective, as even a partially visible contour or a slight color difference can make the animal stand out and break its camouflage. In this paper, we address the question of what makes a camouflage successful, by proposing three scores for automatically assessing its effectiveness. In particular, we show that camouflage can be measured by the similarity between background and foreground features and boundary visibility. We use these camouflage scores to assess and compare all available camouflage datasets. We also incorporate the proposed camouflage score into a generative model as an auxiliary loss and show that effective camouflage images or videos can be synthesised in a scalable manner. The generated synthetic dataset is used to train a transformer-based model for segmenting camouflaged animals in videos. Experimentally, we demonstrate state-of-the-art camouflage breaking performance on the public MoCA-Mask benchmark.

The goal of this paper is to extract the visual-language correspondence from a pre-trained text-to-image diffusion model, in the form of segmentation map, i.e., simultaneously generating images and segmentation masks for the corresponding visual entities described in the text prompt. We make the following contributions: (i) we pair the existing Stable Diffusion model with a novel grounding module, that can be trained to align the visual and textual embedding space of the diffusion model with only a small number of object categories; (ii) we establish an automatic pipeline for constructing a dataset, that consists of {image, segmentation mask, text prompt} triplets, to train the proposed grounding module; (iii) we evaluate the performance of open-vocabulary grounding on images generated from the text-to-image diffusion model and show that the module can well segment the objects of categories beyond seen ones at training time, as shown in Fig. 1; (iv) we adopt the augmented diffusion model to build a synthetic semantic segmentation dataset, and show that, training a standard segmentation model on such dataset demonstrates competitive performance on the zero-shot segmentation (ZS3) benchmark, which opens up new opportunities for adopting the powerful diffusion model for discriminative tasks.

Video Instance Segmentation (VIS) aims at segmenting and categorizing objects in videos from a closed set of training categories, lacking the generalization ability to handle novel categories in real-world videos. To address this limitation, we make the following three contributions. First, we introduce the novel task of Open-Vocabulary Video Instance Segmentation, which aims to simultaneously segment, track, and classify objects in videos from open-set categories, including novel categories unseen during training. Second, to benchmark Open-Vocabulary VIS, we collect a Large-Vocabulary Video Instance Segmentation dataset (LV-VIS), that contains well-annotated objects from 1,196 diverse categories, significantly surpassing the category size of existing datasets by more than one order of magnitude. Third, we propose an efficient Memory-Induced Transformer architecture, OV2Seg, to first achieve Open-Vocabulary VIS in an end-to-end manner with near real-time inference speed. Extensive experiments on LV-VIS and four existing VIS datasets demonstrate the strong zero-shot generalization ability of OV2Seg on novel categories. The dataset and code are released here https://github.com/haochenheheda/LVVIS.

In this paper, we consider enhancing medical visual-language pre-training (VLP) with domain-specific knowledge, by exploiting the paired image-text reports from the radiological daily practice. In particular, we make the following contributions: First, unlike existing works that directly process the raw reports, we adopt a novel triplet extraction module to extract the medical-related information, avoiding unnecessary complexity from language grammar and enhancing the supervision signals; Second, we propose a novel triplet encoding module with entity translation by querying a knowledge base, to exploit the rich domain knowledge in medical field, and implicitly build relationships between medical entities in the language embedding space; Third, we propose to use a Transformer-based fusion model for spatially aligning the entity description with visual signals at the image patch level, enabling the ability for medical diagnosis; Fourth, we conduct thorough experiments to validate the effectiveness of our architecture, and benchmark on numerous public benchmarks e.g., ChestX-ray14, RSNA Pneumonia, SIIM-ACR Pneumothorax, COVIDx CXR-2, COVID Rural, and EdemaSeverity. In both zero-shot and fine-tuning settings, our model has demonstrated strong performance compared with the former methods on disease classification and grounding.

Multi-person motion prediction is a challenging problem due to the dependency of motion on both individual past movements and interactions with other people. Transformer-based methods have shown promising results on this task, but they miss the explicit relation representation between joints, such as skeleton structure and pairwise distance, which is crucial for accurate interaction modeling. In this paper, we propose the Joint-Relation Transformer, which utilizes relation information to enhance interaction modeling and improve future motion prediction. Our relation information contains the relative distance and the intra-/inter-person physical constraints. To fuse relation and joint information, we design a novel joint-relation fusion layer with relation-aware attention to update both features. Additionally, we supervise the relation information by forecasting future distance. Experiments show that our method achieves a 13.4% improvement of 900ms VIM on 3DPW-SoMoF/RC and 17.8%/12.0% improvement of 3s MPJPE on CMU-Mpcap/MuPoTS-3D dataset. Code is available at https://github.com/MediaBrain-SJTU/JRTransformer.

The goal of this paper is open-vocabulary object detection (OVOD) — building a model that can detect objects beyond the set of categories seen at training, thus enabling the user to specify categories of interest at inference without the need for model retraining. We adopt a standard two- stage object detector architecture, and explore three ways for specifying novel categories: via language descriptions, via image exemplars, or via a combination of the two. We make three contributions: first, we prompt a large language model (LLM) to generate informative language descriptions for object classes, and construct powerful text-based classifiers; second, we employ a visual aggregator on image exemplars that can ingest any number of images as input, forming vision-based classifiers; and third, we provide a simple method to fuse information from language descriptions and image exemplars, yield- ing a multi-modal classifier. When evaluating on the challenging LVIS open-vocabulary bench- mark we demonstrate that: (i) our text-based classifiers outperform all previous OVOD works; (ii) our vision-based classifiers perform as well as text-based classifiers in prior work; (iii) using multi-modal classifiers perform better than either modality alone; and finally, (iv) our text-based and multi-modal classifiers yield better performance than a fully-supervised detector.

Unsupervised multi-object segmentation has shown impressive results on images by utilizing powerful semantics learned from self-supervised pretraining. An additional modality such as depth or motion is often used to facilitate the segmentation in video sequences. However, the performance improvements observed in synthetic sequences, which rely on the robustness of an additional cue, do not translate to more challenging real-world scenarios. In this paper, we propose the first fully unsupervised method for segmenting multiple objects in real-world sequences. Our object-centric learning framework spatially binds objects to slots on each frame and then relates these slots across frames. From these temporally-aware slots, the training objective is to reconstruct the middle frame in a high-level semantic feature space. We propose a masking strategy by dropping a significant portion of tokens in the feature space for efficiency and regularization. Additionally, we address over-clustering by merging slots based on similarity. Our method can successfully segment multiple instances of complex and high-variety classes in YouTube videos.

This paper targets on self-supervised tumor segmentation. We make the following contributions: (i) we take inspiration from the observation that tumors are often characterised independently of their contexts, we propose a novel proxy task "layer-decomposition", that closely matches the goal of the downstream task, and design a scalable pipeline for generating synthetic tumor data for pre-training; (ii) we propose a two-stage Sim2Real training regime for unsupervised tumor segmentation, where we first pre-train a model with simulated tumors, and then adopt a self-training strategy for downstream data adaptation; (iii) when evaluating on different tumor segmentation benchmarks, e.g. BraTS2018 for brain tumor segmentation and LiTS2017 for liver tumor segmentation, our approach achieves state-of-the-art segmentation performance under the unsupervised setting. While transferring the model for tumor segmentation under a low-annotation regime, the proposed approach also outperforms all existing self-supervised approaches; (iv) we conduct extensive ablation studies to analyse the critical components in data simulation, and validate the necessity of different proxy tasks. We demonstrate that, with sufficient texture randomization in simulation, model trained on synthetic data can effortlessly generalise to datasets with real tumors.

The objective of this paper is a temporal alignment network that ingests long term video sequences, and associated text sentences, in order to: (1) determine if a sentence is alignable with the video; and (2) if it is alignable, then determine its alignment. The challenge is to train such networks from large-scale datasets, such as HowTo100M, where the associated text sentences have significant noise, and are only weakly aligned when relevant. Apart from proposing the alignment network, we also make four contributions: (i) we describe a novel co-training method that enables to denoise and train on raw instructional videos without using manual annotation, de-spite the considerable noise; (ii) to benchmark the align-ment performance, we manually curate a 10-hour subset of HowTo100M, totalling 80 videos, with sparse temporal de-scriptions. Our proposed model, trained on HowTo100M, outperforms strong baselines (CLIP, MIL-NCE) on this alignment dataset by a significant margin; (iii) we ap-ply the trained model in the zero-shot settings to mul-tiple downstream video understanding tasks and achieve state-of-the-art results, including text-video retrieval on YouCook2, and weakly supervised video action segmentation on Breakfast-Action. (iv) we use the automatically-aligned HowTo100M annotations for end-to-end finetuning of the backbone model, and obtain improved performance on downstream action recognition tasks.

The objective of this paper is few-shot object detection (FSOD) - the task of expanding an object detector for a new category given only a few instances for training. We introduce a simple pseudo-labelling method to source high-quality pseudo-annotations from the training set, for each new category, vastly increasing the number of training instances and reducing class imbalance; our method finds previously unlabelled instances. Naively training with model predictions yields sub-optimal performance; we present two novel methods to improve the precision of the pseudo-labelling process: first, we introduce a verification technique to remove candidate detections with incorrect class labels; second, we train a specialised model to correct poor quality bounding boxes. After these two novel steps, we obtain a large set of high-quality pseudo-annotations that allow our final detector to be trained end-to-end. Additionally, we demonstrate our method maintains base class performance, and the utility of simple augmentations in FSOD. While benchmarking on PASCAL VOC and MS-COCO, our method achieves state-of-the-art or second-best performance compared to existing approaches across all number of shots.

In this paper, we present a framework for reading analog clocks in natural images or videos. Specifically, we make the following contributions: First, we create a scalable pipeline for generating synthetic clocks, significantly reducing the requirements for the labour-intensive annotations; Second, we introduce a clock recognition architecture based on spatial transformer networks (STN), which is trained end-to-end for clock alignment and recognition. We show that the model trained on the proposed synthetic dataset generalises towards real clocks with good accuracy, advocating a Sim2Real training regime; Third, to further reduce the gap between simulation and real data, we leverage the special property of “time”, i.e. uniformity, to generate reliable pseudo-labels on real unlabelled clock videos, and show that training on these videos offers further improvements while still requiring zero manual annotations. Lastly, we introduce three benchmark datasets based on COCO, Open Images, and The Clock movie, with full annotations for time, accurate to the minute.

We present a simple yet effective self-supervised framework for audio-visual representation learning, to localize the sound source in videos. To understand what enables to learn useful representations, we systematically investigate the effects of data augmentations, and reveal that (1) composition of data augmentations plays a critical role, i.e. explicitly encouraging the audio-visual representations to be invariant to various transformations (transformation invariance); (2) enforcing geometric consistency substantially improves the quality of learned representations, i.e. the detected sound source should follow the same transformation applied on input video frames (transformation equivariance). Extensive experiments demonstrate that our model significantly outperforms previous methods on two sound localization benchmarks, namely, Flickr-SoundNet and VGG-Sound. Additionally, we also evaluate audio retrieval and cross-modal retrieval tasks. In both cases, our self-supervised models demonstrate superior retrieval performances, even competitive with the supervised approach in audio retrieval. This reveals the proposed framework learns strong multi-modal representations that are beneficial to sound localisation and generalization to further applications. The project page is https://jinxiang-liu.github.io/SSL-TIE.

We explore a feed-forward approach for decomposing a video into layers, where each layer contains an object of interest along with its associated shadows, reflections, and other visual effects. This problem is challenging since associated effects vary widely with the 3D geometry and lighting conditions in the scene, and ground-truth labels for visual effects are difficult (and in some cases impractical) to collect. We take a self-supervised approach and train a neural network to produce a foreground image and alpha matte from a rough object segmentation mask under a reconstruction and sparsity loss. Under reconstruction loss, the layer decomposition problem is underdetermined: many combinations of layers may reconstruct the input video.Inspired by the game theory concept of focal points---or \emph{Schelling points}---we pose the problem as a coordination game, where each player (network) predicts the effects for a single object without knowledge of the other players' choices. The players learn to converge on the ``natural'' layer decomposition in order to maximize the likelihood of their choices aligning with the other players'. We train the network to play this game with itself, and show how to design the rules of this game so that the focal point lies at the correct layer decomposition. We demonstrate feed-forward results on a challenging synthetic dataset, then show that pretraining on this dataset significantly reduces optimization time for real videos.

Semantic segmentation has a broad range of applications, but its real-world impact has been significantly limited by the prohibitive annotation costs necessary to enable deployment. Segmentation methods that forgo supervision can side-step these costs, but exhibit the inconvenient requirement to provide labelled examples from the target distribution to assign concept names to predictions. An alternative line of work in language-image pre-training has recently demonstrated the potential to produce models that can both assign names across large vocabularies of concepts and enable zero-shot transfer for classification, but do not demonstrate commensurate segmentation abilities.We leverage the retrieval abilities of one such language-image pre-trained model, CLIP, to dynamically curate training sets from unlabelled images for arbitrary collections of concept names, and leverage the robust correspondences offered by modern image representations to co-segment entities among the resulting collections. The synthetic segment collections are then employed to construct a segmentation model (without requiring pixel labels) whose knowledge of concepts is inherited from the scalable pre-training process of CLIP. We demonstrate that our approach, termed Retrieve and Co-segment (ReCo) performs favourably to conventional unsupervised segmentation approaches while inheriting the convenience of nameable predictions and zero-shot transfer. We also demonstrate ReCo’s ability to generate specialist segmenters for extremely rare objects.

The objective of this paper is a model that is able to discover, track and segment multiple moving objects in a video. We make four contributions: First, we introduce an object-centric segmentation model with a depth-ordered layer representation. This is implemented using a variant of the transformer architecture that ingests optical flow, where each query vector specifies an object and its layer for the entire video. The model can effectively discover multiple moving objects and handle mutual occlusions; Second, we introduce a scalable pipeline for generating multi-object synthetic training data via layer compositions, that is used to train the proposed model, significantly reducing the requirements for labour-intensive annotations, and supporting Sim2Real generalisation; Third, we conduct thorough ablation studies, showing that the model is able to learn object permanence and temporal shape consistency, and is able to predict amodal segmentation masks; Fourth, we evaluate our model, trained only on synthetic data, on standard video segmentation benchmarks, DAVIS, MoCA, SegTrack, FBMS-59, and achieve state-of-the-art performance among existing methods that do not rely on any manual annotations. With test-time adaptation, we observe further performance boosts.

The objective of this work is to localize sound sources that are visible in a video without using manual annotations. Our key technical contribution is to show that, by training the network to explicitly discriminate challenging image fragments, even for images that do contain the object emitting the sound, we can significantly boost the localization performance. We do so elegantly by introducing a mechanism to mine hard samples and add them to a contrastive learning formulation automatically. We show that our algorithm achieves state-of-the-art performance on the popular Flickr SoundNet dataset. Furthermore, we introduce the VGG-Sound Source (VGG-SS) benchmark, a new set of annotations for the recently-introduced VGG-Sound dataset, where the sound sources visible in each video clip are explicitly marked with bounding box annotations. This dataset is 20 times larger than analogous existing ones, contains 5K videos spanning over 200 categories, and, differently from Flickr SoundNet, is video-based. On VGG-SS, we also show that our algorithm achieves state-of-the-art performance against several baselines. Code and datasets can be found at http://www.robots.ox.ac.uk/˜vgg/research/lvs/.

Animals have evolved highly functional visual systems to understand motion, assisting perception even under complex environments. In this paper, we work towards developing a computer vision system able to segment objects by exploiting motion cues, i.e. motion segmentation. To achieve this, we introduce a simple variant of the Transformer to segment optical flow frames into primary objects and the background, which can be trained in a self-supervised manner, i.e. without using any manual annotations. Despite using only optical flow, and no appearance information, as input, our approach achieves superior results compared to previous state-of-the-art self-supervised methods on public benchmarks (DAVIS2016, SegTrackv2, FBMS59), while being an order of magnitude faster. On a challenging camouflage dataset (MoCA), we significantly outperform other self-supervised approaches, and are competitive with the top supervised approach, highlighting the importance of motion cues and the potential bias towards appearance in existing video segmentation models.

Recent interest in self-supervised dense tracking has yielded rapid progress, but performance still remains far from supervised methods. We propose a dense tracking model trained on videos without any annotations that surpasses previous self-supervised methods on existing benchmarks by a significant margin (+15%), and achieves performance comparable to supervised methods. In this paper, we first reassess the traditional choices used for self-supervised training and reconstruction loss by conducting thorough experiments that finally elucidate the optimal choices. Second, we further improve on existing methods by augmenting our architecture with a crucial memory component. Third, we benchmark on large-scale semi-supervised video object segmentation (aka. dense tracking), and propose a new metric: generalizability. Our first two contributions yield a self-supervised network that for the first time is competitive with supervised methods on standard evaluation metrics of dense tracking. When measuring generalizability, we show self-supervised approaches are actually superior to the majority of supervised methods. We believe this new generalizability metric can better capture the real-world use-cases for dense tracking, and will spur new interest in this research direction.

Our goal is to collect a large-scale audio-visual dataset with low label noise from videos `in the wild' using computer vision techniques. The resulting dataset can be used for training and evaluating audio recognition models. We make three contributions. First, we propose a scalable pipeline based on computer vision techniques to create an audio dataset from open-source media. Our pipeline involves obtaining videos from YouTube; using image classification algorithms to localize audio-visual correspondence; and filtering out ambient noise using audio verification. Second, we use this pipeline to curate the VGGSound dataset consisting of more than 200k videos for 300 audio classes. Third, we investigate various Convolutional Neural Network (CNN) architectures and aggregation approaches to establish audio recognition baselines for our new dataset. Compared to existing audio datasets, VGGSound ensures audio-visual correspondence and is collected under unconstrained conditions. Code and the dataset are available at http://www.robots.ox.ac.uk/~vgg/data/vggsound/.

The objective of this paper is visual-only self-supervised video representation learning. We make the following contributions: (i) we investigate the benefit of adding semantic-class positives to instance-based Info Noise Contrastive Estimation (InfoNCE) training, showing that this form of supervised contrastive learning leads to a clear improvement in performance; (ii) we propose a novel self-supervised co-training scheme to improve the popular infoNCE loss, exploiting the complementary information from different views, RGB streams and optical flow, of the same data source by using one view to obtain positive class samples for the other; (iii) we thoroughly evaluate the quality of the learnt representation on two different downstream tasks: action recognition and video retrieval. In both cases, the proposed approach demonstrates state-of-the-art or comparable performance with other self-supervised approaches, whilst being significantly more efficient to train, i.e. requiring far less training data to achieve similar performance.

The objective of this work is speaker recognition under noisy and unconstrained conditions. We make two key contributions. First, we introduce a very large-scale audio-visual dataset collected from open source media using a fully automated pipeline. Most existing datasets for speaker identification contain samples obtained under quite constrained conditions, and usually require manual annotations, hence are limited in size. We propose a pipeline based on computer vision techniques to create the dataset from open-source media. Our pipeline involves obtaining videos from YouTube; performing active speaker verification using a two-stream synchronization Convolutional Neural Network (CNN), and confirming the identity of the speaker using CNN based facial recognition. We use this pipeline to curate VoxCeleb which contains contains over a million ‘real-world’ utterances from over 6000 speakers. This is several times larger than any publicly available speaker recognition dataset. Second, we develop and compare different CNN architectures with various aggregation methods and training loss functions that can effectively recognise identities from voice under various conditions. The models trained on our dataset surpass the performance of previous works by a significant margin.

Convolutional Neural Networks (CNNs), which are currently state-of-the-art for most image analysis tasks, are ill suited to leveraging the key benefits of ultrasound imaging - specifically, ultrasound's portability and real-time capabilities. CNNs have large memory footprints, which obstructs their implementation on mobile devices, and require numerous floating point operations, which results in slow CPU inference times. In this paper, we propose three approaches to training efficient CNNs that can operate in real-time on a CPU (catering to the clinical setting), with a low memory footprint, for minimal compromise in accuracy. We first demonstrate the power of 'thin' CNNs, with very few feature channels, for fast medical image segmentation. We then leverage separable convolutions to further speed up inference, reduce parameter count and facilitate mobile deployment. Lastly, we propose a novel knowledge distillation technique to boost the accuracy of light-weight models, while maintaining inference speed-up. For a negligible sacrifice in test set Dice performance on the challenging ultrasound analysis task of nerve segmentation, our final proposed model processes images at 30fps on a CPU, which is 9× faster than the standard U-Net, while requiring 420× less space in memory.

We present a method for retiming people in an ordinary, natural video --- manipulating and editing the time in which different motions of individuals in the video occur. We can temporally align different motions, change the speed of certain actions (speeding up/slowing down, or entirely "freezing" people), or "erase" selected people from the video altogether. We achieve these effects computationally via a dedicated learning-based layered video representation, where each frame in the video is decomposed into separate RGBA layers, representing the appearance of different people in the video. A key property of our model is that it not only disentangles the direct motions of each person in the input video, but also correlates each person automatically with the scene changes they generate---e.g., shadows, reflections, and motion of loose clothing. The layers can be individually retimed and recombined into a new video, allowing us to achieve realistic, high-quality renderings of retiming effects for real-world videos depicting complex actions and involving multiple individuals, including dancing, trampoline jumping, or group running.

The objective of this paper is speaker recognition `in the wild' - where utterances may be of variable length and also contain irrelevant signals. Crucial elements in the design of deep networks for this task are the type of trunk (frame level) network, and the method of temporal aggregation. We propose a powerful speaker recognition deep network, using a `thin-ResNet' trunk architecture, and a dictionary-based NetVLAD or GhostVLAD layer to aggregate features across time, that can be trained end-to-end. We show that our network achieves state of the art performance by a significant margin on the VoxCeleb1 test set for speaker recognition, whilst requiring fewer parameters than previous methods. We also investigate the effect of utterance length on performance, and conclude that for `in the wild' data, a longer length is beneficial.

In this paper, we introduce a new large-scale face dataset named VGGFace2. The dataset contains 3.31 million images of 9131 subjects, with an average of 362.6 images for each subject. Images are downloaded from Google Image Search and have large variations in pose, age, illumination, ethnicity and profession (e.g. actors, athletes, politicians). The dataset was collected with three goals in mind: (i) to have both a large number of identities and also a large number of images for each identity; (ii) to cover a large range of pose, age and ethnicity; and (iii) to minimise the label noise. We describe how the dataset was collected, in particular the automated and manual filtering stages to ensure a high accuracy for the images of each identity. To assess face recognition performance using the new dataset, we train ResNet-50 (with and without Squeeze-and-Excitation blocks) Convolutional Neural Networks on VGGFace2, on MS-Celeb-1M, and on their union, and show that training on VGGFace2 leads to improved recognition performance over pose and age. Finally, using the models trained on these datasets, we demonstrate state-of-the-art performance on the IJB-A and IJB-B face recognition benchmarks, exceeding the previous state-of-the-art by a large margin. The dataset and models are publicly available.