Yang Liu

Yang Liu, Guanbin Li, Liang Lin

Existing visual question answering methods often suffer from cross-modal spurious correlations and oversimplified event-level reasoning processes that fail to capture event temporality, causality, and dynamics spanning over the video. In this work, to address the task of event-level visual question answering, we propose a framework for cross-modal causal relational reasoning. In particular, a set of causal intervention operations is introduced to discover the underlying causal structures across visual and linguistic modalities. Our framework, named Cross-Modal Causal RelatIonal Reasoning (CMCIR), involves three modules: i) Causality-aware Visual-Linguistic Reasoning (CVLR) module for collaboratively disentangling the visual and linguistic spurious correlations via front-door and back-door causal interventions; ii) Spatial-Temporal Transformer (STT) module for capturing the fine-grained interactions between visual and linguistic semantics; iii) Visual-Linguistic Feature Fusion (VLFF) module for learning the global semantic-aware visual-linguistic representations adaptively. Extensive experiments on four event-level datasets demonstrate the superiority of our CMCIR in discovering visual-linguistic causal structures and achieving robust event-level visual question answering. The datasets, code, and models are available at https://github.com/HCPLab-SYSU/CMCIR.

Zhihong Chen, Yuhao Du, Jinpeng Hu et al.

Medical vision-and-language pre-training provides a feasible solution to extract effective vision-and-language representations from medical images and texts. However, few studies have been dedicated to this field to facilitate medical vision-and-language understanding. In this paper, we propose a self-supervised learning paradigm with multi-modal masked autoencoders (M$^3$AE), which learn cross-modal domain knowledge by reconstructing missing pixels and tokens from randomly masked images and texts. There are three key designs to make this simple approach work. First, considering the different information densities of vision and language, we adopt different masking ratios for the input image and text, where a considerably larger masking ratio is used for images. Second, we use visual and textual features from different layers to perform the reconstruction to deal with different levels of abstraction in visual and language. Third, we develop different designs for vision and language decoders (i.e., a Transformer for vision and a multi-layer perceptron for language). To perform a comprehensive evaluation and facilitate further research, we construct a medical vision-and-language benchmark including three tasks. Experimental results demonstrate the effectiveness of our approach, where state-of-the-art results are achieved on all downstream tasks. Besides, we conduct further analysis to better verify the effectiveness of different components of our approach and various settings of pre-training. The source code is available at~\url{https://github.com/zhjohnchan/M3AE}.

Ting Lei, Fabian Caba, Qingchao Chen et al.

Human Object Interaction (HOI) detection aims to localize and infer the relationships between a human and an object. Arguably, training supervised models for this task from scratch presents challenges due to the performance drop over rare classes and the high computational cost and time required to handle long-tailed distributions of HOIs in complex HOI scenes in realistic settings. This observation motivates us to design an HOI detector that can be trained even with long-tailed labeled data and can leverage existing knowledge from pre-trained models. Inspired by the powerful generalization ability of the large Vision-Language Models (VLM) on classification and retrieval tasks, we propose an efficient Adaptive HOI Detector with Concept-guided Memory (ADA-CM). ADA-CM has two operating modes. The first mode makes it tunable without learning new parameters in a training-free paradigm. Its second mode incorporates an instance-aware adapter mechanism that can further efficiently boost performance if updating a lightweight set of parameters can be afforded. Our proposed method achieves competitive results with state-of-the-art on the HICO-DET and V-COCO datasets with much less training time. Code can be found at https://github.com/ltttpku/ADA-CM.

Jiangbo Pei, Zhuqing Jiang, Aidong Men et al.

Source-free unsupervised domain adaptation (SFUDA) aims to learn a target domain model using unlabeled target data and the knowledge of a well-trained source domain model. Most previous SFUDA works focus on inferring semantics of target data based on the source knowledge. Without measuring the transferability of the source knowledge, these methods insufficiently exploit the source knowledge, and fail to identify the reliability of the inferred target semantics. However, existing transferability measurements require either source data or target labels, which are infeasible in SFUDA. To this end, firstly, we propose a novel Uncertainty-induced Transferability Representation (UTR), which leverages uncertainty as the tool to analyse the channel-wise transferability of the source encoder in the absence of the source data and target labels. The domain-level UTR unravels how transferable the encoder channels are to the target domain and the instance-level UTR characterizes the reliability of the inferred target semantics. Secondly, based on the UTR, we propose a novel Calibrated Adaption Framework (CAF) for SFUDA, including i)the source knowledge calibration module that guides the target model to learn the transferable source knowledge and discard the non-transferable one, and ii)the target semantics calibration module that calibrates the unreliable semantics. With the help of the calibrated source knowledge and the target semantics, the model adapts to the target domain safely and ultimately better. We verified the effectiveness of our method using experimental results and demonstrated that the proposed method achieves state-of-the-art performances on the three SFUDA benchmarks. Code is available at https://github.com/SPIresearch/UTR.

Weixing Chen, Yang Liu, Ce Wang et al.

Radiology Report Generation (RRG) is essential for computer-aided diagnosis and medication guidance, which can relieve the heavy burden of radiologists by automatically generating the corresponding radiology reports according to the given radiology image. However, generating accurate lesion descriptions remains challenging due to spurious correlations from visual-linguistic biases and inherent limitations of radiological imaging, such as low resolution and noise interference. To address these issues, we propose a two-stage framework named CrossModal Causal Representation Learning (CMCRL), consisting of the Radiological Cross-modal Alignment and Reconstruction Enhanced (RadCARE) pre-training and the Visual-Linguistic Causal Intervention (VLCI) fine-tuning. In the pre-training stage, RadCARE introduces a degradation-aware masked image restoration strategy tailored for radiological images, which reconstructs high-resolution patches from low-resolution inputs to mitigate noise and detail loss. Combined with a multiway architecture and four adaptive training strategies (e.g., text postfix generation with degraded images and text prefixes), RadCARE establishes robust cross-modal correlations even with incomplete data. In the VLCI phase, we deploy causal front-door intervention through two modules: the Visual Deconfounding Module (VDM) disentangles local-global features without fine-grained annotations, while the Linguistic Deconfounding Module (LDM) eliminates context bias without external terminology databases. Experiments on IU-Xray and MIMIC-CXR show that our CMCRL pipeline significantly outperforms state-of-the-art methods, with ablation studies confirming the necessity of both stages. Code and models are available at https://github.com/WissingChen/CMCRL.

Dezhao Luo, Jiabo Huang, Shaogang Gong et al.

The correlation between the vision and text is essential for video moment retrieval (VMR), however, existing methods heavily rely on separate pre-training feature extractors for visual and textual understanding. Without sufficient temporal boundary annotations, it is non-trivial to learn universal video-text alignments. In this work, we explore multi-modal correlations derived from large-scale image-text data to facilitate generalisable VMR. To address the limitations of image-text pre-training models on capturing the video changes, we propose a generic method, referred to as Visual-Dynamic Injection (VDI), to empower the model's understanding of video moments. Whilst existing VMR methods are focusing on building temporal-aware video features, being aware of the text descriptions about the temporal changes is also critical but originally overlooked in pre-training by matching static images with sentences. Therefore, we extract visual context and spatial dynamic information from video frames and explicitly enforce their alignments with the phrases describing video changes (e.g. verb). By doing so, the potentially relevant visual and motion patterns in videos are encoded in the corresponding text embeddings (injected) so to enable more accurate video-text alignments. We conduct extensive experiments on two VMR benchmark datasets (Charades-STA and ActivityNet-Captions) and achieve state-of-the-art performances. Especially, VDI yields notable advantages when being tested on the out-of-distribution splits where the testing samples involve novel scenes and vocabulary.

Chengcheng Ma, Yang Liu, Jiankang Deng et al.

Pretrained vision-language models (VLMs) such as CLIP have shown impressive generalization capability in downstream vision tasks with appropriate text prompts. Instead of designing prompts manually, Context Optimization (CoOp) has been recently proposed to learn continuous prompts using taskspecific training data. Despite the performance improvements on downstream tasks, several studies have reported that CoOp suffers from the overfitting issue in two aspects: (i) the test accuracy on base classes first improves and then worsens during training;(ii) the test accuracy on novel classes keeps decreasing. However, none of the existing studies can understand and mitigate such overfitting problems. In this study, we first explore the cause of overfitting by analyzing the gradient flow. Comparative experiments reveal that CoOp favors generalizable and spurious features in the early and later training stages, respectively, leading to the non-overfitting and overfitting phenomena. Given those observations, we propose Subspace Prompt Tuning (SubPT) to project the gradients in back-propagation onto the low-rank subspace spanned by the early-stage gradient flow eigenvectors during the entire training process and successfully eliminate the overfitting problem. In addition, we equip CoOp with a Novel Feature Learner (NFL) to enhance the generalization ability of the learned prompts onto novel categories beyond the training set, needless of image training data. Extensive experiments on 11 classification datasets demonstrate that SubPT+NFL consistently boost the performance of CoOp and outperform the state-of-the-art CoCoOp approach. Experiments on more challenging vision downstream tasks, including open-vocabulary object detection and zero-shot semantic segmentation, also verify the effectiveness of the proposed method. Codes can be found at https://tinyurl.com/mpe64f89.

Yinsong Xu, Zhuqing Jiang, Aidong Men et al.

Existing domain adaptation methods assume that domain discrepancies are caused by a few discrete attributes and variations, e.g., art, real, painting, quickdraw, etc. We argue that this is not realistic as it is implausible to define the real-world datasets using a few discrete attributes. Therefore, we propose to investigate a new problem namely the Continuous Domain Adaptation (CDA) through the lens where infinite domains are formed by continuously varying attributes. Leveraging knowledge of two labeled source domains and several observed unlabeled target domains data, the objective of CDA is to learn a generalized model for whole data distribution with the continuous attribute. Besides the contributions of formulating a new problem, we also propose a novel approach as a strong CDA baseline. To be specific, firstly we propose a novel alternating training strategy to reduce discrepancies among multiple domains meanwhile generalize to unseen target domains. Secondly, we propose a continuity constraint when estimating the cross-domain divergence measurement. Finally, to decouple the discrepancy from the mini-batch size, we design a domain-specific queue to maintain the global view of the source domain that further boosts the adaptation performances. Our method is proven to achieve the state-of-the-art in CDA problem using extensive experiments. The code is available at https://github.com/SPIresearch/CDA.

Yang Liu, Yushen Wei, Hong Yan et al.

Visual representation learning is ubiquitous in various real-world applications, including visual comprehension, video understanding, multi-modal analysis, human-computer interaction, and urban computing. Due to the emergence of huge amounts of multi-modal heterogeneous spatial/temporal/spatial-temporal data in big data era, the lack of interpretability, robustness, and out-of-distribution generalization are becoming the challenges of the existing visual models. The majority of the existing methods tend to fit the original data/variable distributions and ignore the essential causal relations behind the multi-modal knowledge, which lacks unified guidance and analysis about why modern visual representation learning methods easily collapse into data bias and have limited generalization and cognitive abilities. Inspired by the strong inference ability of human-level agents, recent years have therefore witnessed great effort in developing causal reasoning paradigms to realize robust representation and model learning with good cognitive ability. In this paper, we conduct a comprehensive review of existing causal reasoning methods for visual representation learning, covering fundamental theories, models, and datasets. The limitations of current methods and datasets are also discussed. Moreover, we propose some prospective challenges, opportunities, and future research directions for benchmarking causal reasoning algorithms in visual representation learning. This paper aims to provide a comprehensive overview of this emerging field, attract attention, encourage discussions, bring to the forefront the urgency of developing novel causal reasoning methods, publicly available benchmarks, and consensus-building standards for reliable visual representation learning and related real-world applications more efficiently.

Minghang Zheng, Jiahua Zhang, Qingchao Chen et al.

Visual grounding aims to localize the object referred to in an image based on a natural language query. Although progress has been made recently, accurately localizing target objects within multiple-instance distractions (multiple objects of the same category as the target) remains a significant challenge. Existing methods demonstrate a significant performance drop when there are multiple distractions in an image, indicating an insufficient understanding of the fine-grained semantics and spatial relationships between objects. In this paper, we propose a novel approach, the Relation and Semantic-sensitive Visual Grounding (ResVG) model, to address this issue. Firstly, we enhance the model's understanding of fine-grained semantics by injecting semantic prior information derived from text queries into the model. This is achieved by leveraging text-to-image generation models to produce images representing the semantic attributes of target objects described in queries. Secondly, we tackle the lack of training samples with multiple distractions by introducing a relation-sensitive data augmentation method. This method generates additional training data by synthesizing images containing multiple objects of the same category and pseudo queries based on their spatial relationships. The proposed ReSVG model significantly improves the model's ability to comprehend both object semantics and spatial relations, leading to enhanced performance in visual grounding tasks, particularly in scenarios with multiple-instance distractions. We conduct extensive experiments to validate the effectiveness of our methods on five datasets. Code is available at https://github.com/minghangz/ResVG.

Ting Lei, Shaofeng Yin, Yuxin Peng et al.

Zero-shot Human-Object Interaction (HOI) detection has emerged as a frontier topic due to its capability to detect HOIs beyond a predefined set of categories. This task entails not only identifying the interactiveness of human-object pairs and localizing them but also recognizing both seen and unseen interaction categories. In this paper, we introduce a novel framework for zero-shot HOI detection using Conditional Multi-Modal Prompts, namely CMMP. This approach enhances the generalization of large foundation models, such as CLIP, when fine-tuned for HOI detection. Unlike traditional prompt-learning methods, we propose learning decoupled vision and language prompts for interactiveness-aware visual feature extraction and generalizable interaction classification, respectively. Specifically, we integrate prior knowledge of different granularity into conditional vision prompts, including an input-conditioned instance prior and a global spatial pattern prior. The former encourages the image encoder to treat instances belonging to seen or potentially unseen HOI concepts equally while the latter provides representative plausible spatial configuration of the human and object under interaction. Besides, we employ language-aware prompt learning with a consistency constraint to preserve the knowledge of the large foundation model to enable better generalization in the text branch. Extensive experiments demonstrate the efficacy of our detector with conditional multi-modal prompts, outperforming previous state-of-the-art on unseen classes of various zero-shot settings. The code and models are available at \url{https://github.com/ltttpku/CMMP}.

Haoyu Liu, Yang Liu, Hongkai He et al.

People always desire an embodied agent that can perform a task by understanding language instruction. Moreover, they also want to monitor and expect agents to understand commands the way they expected. But, how to build such an embodied agent is still unclear. Recently, people can explore this problem with the Vision-and-Language Interaction benchmark ALFRED, which requires an agent to perform complicated daily household tasks following natural language instructions in unseen scenes. In this paper, we propose LEBP -- Language Expectation and Binding Policy Module to tackle the ALFRED. The LEBP contains a two-stream process: 1) It first conducts a language expectation module to generate an expectation describing how to perform tasks by understanding the language instruction. The expectation consists of a sequence of sub-steps for the task (e.g., Pick an apple). The expectation allows people to access and check the understanding results of instructions before the agent takes actual actions, in case the task might go wrong. 2) Then, it uses the binding policy module to bind sub-steps in expectation to actual actions to specific scenarios. Actual actions include navigation and object manipulation. Experimental results suggest our approach achieves comparable performance to currently published SOTA methods and can avoid large decay from seen scenarios to unseen scenarios.

Jun Rao, Liang Ding, Shuhan Qi et al.

Although the vision-and-language pretraining (VLP) equipped cross-modal image-text retrieval (ITR) has achieved remarkable progress in the past two years, it suffers from a major drawback: the ever-increasing size of VLP models restricts its deployment to real-world search scenarios (where the high latency is unacceptable). To alleviate this problem, we present a novel plug-in dynamic contrastive distillation (DCD) framework to compress the large VLP models for the ITR task. Technically, we face the following two challenges: 1) the typical uni-modal metric learning approach is difficult to directly apply to the cross-modal tasks, due to the limited GPU memory to optimize too many negative samples during handling cross-modal fusion features. 2) it is inefficient to static optimize the student network from different hard samples, which have different effects on distillation learning and student network optimization. We try to overcome these challenges from two points. First, to achieve multi-modal contrastive learning, and balance the training costs and effects, we propose to use a teacher network to estimate the difficult samples for students, making the students absorb the powerful knowledge from pre-trained teachers, and master the knowledge from hard samples. Second, to dynamic learn from hard sample pairs, we propose dynamic distillation to dynamically learn samples of different difficulties, from the perspective of better balancing the difficulty of knowledge and students' self-learning ability. We successfully apply our proposed DCD strategy to two state-of-the-art vision-language pretrained models, i.e. ViLT and METER. Extensive experiments on MS-COCO and Flickr30K benchmarks show the effectiveness and efficiency of our DCD framework. Encouragingly, we can speed up the inference at least 129$\times$ compared to the existing ITR models.

Yang Liu, Ying Tan, Jingzhou Luo et al.

Existing visual question reasoning methods usually fail to explicitly discover the inherent causal mechanism and ignore jointly modeling cross-modal event temporality and causality. In this paper, we propose a visual question reasoning framework named Cross-Modal Question Reasoning (CMQR), to discover temporal causal structure and mitigate visual spurious correlation by causal intervention. To explicitly discover visual causal structure, the Visual Causality Discovery (VCD) architecture is proposed to find question-critical scene temporally and disentangle the visual spurious correlations by attention-based front-door causal intervention module named Local-Global Causal Attention Module (LGCAM). To align the fine-grained interactions between linguistic semantics and spatial-temporal representations, we build an Interactive Visual-Linguistic Transformer (IVLT) that builds the multi-modal co-occurrence interactions between visual and linguistic content. Extensive experiments on four datasets demonstrate the superiority of CMQR for discovering visual causal structures and achieving robust question reasoning.

Zineng Tang, Ziyi Yang, Mahmoud Khademi et al.

We present CoDi-2, a versatile and interactive Multimodal Large Language Model (MLLM) that can follow complex multimodal interleaved instructions, conduct in-context learning (ICL), reason, chat, edit, etc., in an any-to-any input-output modality paradigm. By aligning modalities with language for both encoding and generation, CoDi-2 empowers Large Language Models (LLMs) to not only understand complex modality-interleaved instructions and in-context examples, but also autoregressively generate grounded and coherent multimodal outputs in the continuous feature space. To train CoDi-2, we build a large-scale generation dataset encompassing in-context multimodal instructions across text, vision, and audio. CoDi-2 demonstrates a wide range of zero-shot capabilities for multimodal generation, such as in-context learning, reasoning, and compositionality of any-to-any modality generation through multi-round interactive conversation. CoDi-2 surpasses previous domain-specific models on tasks such as subject-driven image generation, vision transformation, and audio editing. CoDi-2 signifies a substantial breakthrough in developing a comprehensive multimodal foundation model adept at interpreting in-context language-vision-audio interleaved instructions and producing multimodal outputs.

Chu Chen, Yang Liu, Se Weon Park et al.

Chemical Exchange Saturation Transfer (CEST) MRI demonstrates its capability in significantly enhancing the detection of proteins and metabolites with low concentrations through exchangeable protons. The clinical application of CEST, however, is constrained by its low contrast and low signal-to-noise ratio (SNR) in the acquired data. Denoising, as one of the post-processing stages for CEST data, can effectively improve the accuracy of CEST quantification. In this work, by modeling spatial variant z-spectrums into low-dimensional subspace, we introduce Implicit Regression in Subspace (IRIS), which is an unsupervised denoising algorithm utilizing the excellent property of implicit neural representation for continuous mapping. Experiments conducted on both synthetic and in-vivo data demonstrate that our proposed method surpasses other CEST denoising methods regarding both qualitative and quantitative performance.

Yang Liu, Pengxiang Ding, Siteng Huang et al.

Fueled by the Large Language Models (LLMs) wave, Large Visual-Language Models (LVLMs) have emerged as a pivotal advancement, bridging the gap between image and text. However, video making it challenging for LVLMs to perform adequately due to the complexity of the relationship between language and spatial-temporal data structure. Recent Large Video-Language Models (LVidLMs) align feature of static visual data like image into latent space of language feature, by general multi-modal tasks to leverage abilities of LLMs sufficiently. In this paper, we explore fine-grained alignment approach via object trajectory for different modalities across both spatial and temporal dimensions simultaneously. Thus, we propose a novel LVidLM by trajectory-guided Pixel-Temporal Alignment, dubbed PiTe, that exhibits promising applicable model property. To achieve fine-grained video-language alignment, we curate a multi-modal pre-training dataset PiTe-143k, the dataset provision of moving trajectories in pixel level for all individual objects, that appear and mention in the video and caption both, by our automatic annotation pipeline. Meanwhile, PiTe demonstrates astounding capabilities on myriad video-related multi-modal tasks through beat the state-of-the-art methods by a large margin.

Yinsong Xu, Aidong Men, Yang Liu et al.

In deep learning, initializing models with pre-trained weights has become the de facto practice for various downstream tasks. Many unsupervised domain adaptation (UDA) methods typically adopt a backbone pre-trained on ImageNet, and focus on reducing the source-target domain discrepancy. However, the impact of pre-training on adaptation received little attention. In this study, we delve into UDA from the novel perspective of pre-training. We first demonstrate the impact of pre-training by analyzing the dynamic distribution discrepancies between pre-training data domain and the source/ target domain during adaptation. Then, we reveal that the target error also stems from the pre-training in the following two factors: 1) empirically, target error arises from the gradually degenerative pre-trained knowledge during adaptation; 2) theoretically, the error bound depends on difference between the gradient of loss function, \ie, on the target domain and pre-training data domain. To address these two issues, we redefine UDA as a three-domain problem, \ie, source domain, target domain, and pre-training data domain; then we propose a novel framework, named TriDA. We maintain the pre-trained knowledge and improve the error bound by incorporating pre-training data into adaptation for both vanilla UDA and source-free UDA scenarios. For efficiency, we introduce a selection strategy for pre-training data, and offer a solution with synthesized images when pre-training data is unavailable during adaptation. Notably, TriDA is effective even with a small amount of pre-training or synthesized images, and seamlessly complements the two scenario UDA methods, demonstrating state-of-the-art performance across multiple benchmarks. We hope our work provides new insights for better understanding and application of domain adaptation.

Wentao Mo, Yang Liu

In 3D Visual Question Answering (3D VQA), the scarcity of fully annotated data and limited visual content diversity hampers the generalization to novel scenes and 3D concepts (e.g., only around 800 scenes are utilized in ScanQA and SQA dataset). Current approaches resort supplement 3D reasoning with 2D information. However, these methods face challenges: either they use top-down 2D views that introduce overly complex and sometimes question-irrelevant visual clues, or they rely on globally aggregated scene/image-level representations from 2D VLMs, losing the fine-grained vision-language correlations. To overcome these limitations, our approach utilizes question-conditional 2D view selection procedure, pinpointing semantically relevant 2D inputs for crucial visual clues. We then integrate this 2D knowledge into the 3D-VQA system via a two-branch Transformer structure. This structure, featuring a Twin-Transformer design, compactly combines 2D and 3D modalities and captures fine-grained correlations between modalities, allowing them mutually augmenting each other. Integrating proposed mechanisms above, we present BridgeQA, that offers a fresh perspective on multi-modal transformer-based architectures for 3D-VQA. Experiments validate that BridgeQA achieves state-of-the-art on 3D-VQA datasets and significantly outperforms existing solutions. Code is available at $\href{https://github.com/matthewdm0816/BridgeQA}{\text{this URL}}$.

Xiaojun Jia, Sensen Gao, Qing Guo et al.

Vision-language pre-training (VLP) models excel at interpreting both images and text but remain vulnerable to multimodal adversarial examples (AEs). Advancing the generation of transferable AEs, which succeed across unseen models, is key to developing more robust and practical VLP models. Previous approaches augment image-text pairs to enhance diversity within the adversarial example generation process, aiming to improve transferability by expanding the contrast space of image-text features. However, these methods focus solely on diversity around the current AEs, yielding limited gains in transferability. To address this issue, we propose to increase the diversity of AEs by leveraging the intersection regions along the adversarial trajectory during optimization. Specifically, we propose sampling from adversarial evolution triangles composed of clean, historical, and current adversarial examples to enhance adversarial diversity. We provide a theoretical analysis to demonstrate the effectiveness of the proposed adversarial evolution triangle. Moreover, we find that redundant inactive dimensions can dominate similarity calculations, distorting feature matching and making AEs model-dependent with reduced transferability. Hence, we propose to generate AEs in the semantic image-text feature contrast space, which can project the original feature space into a semantic corpus subspace. The proposed semantic-aligned subspace can reduce the image feature redundancy, thereby improving adversarial transferability. Extensive experiments across different datasets and models demonstrate that the proposed method can effectively improve adversarial transferability and outperform state-of-the-art adversarial attack methods. The code is released at https://github.com/jiaxiaojunQAQ/SA-AET.

Jiayi Gao, Changcheng Hua, Qingchao Chen et al.

Identity-preserving text-to-video (IPT2V) generation creates videos faithful to both a reference subject image and a text prompt. While fine-tuning large pretrained video diffusion models on ID-matched data achieves state-of-the-art results on IPT2V, data scarcity and high tuning costs hinder broader improvement. We thus introduce a Training-Free Prompt, Image, and Guidance Enhancement (TPIGE) framework that bridges the semantic gap between the video description and the reference image and design sampling guidance that enhances identity preservation and video quality, achieving performance gains at minimal cost.Specifically, we first propose Face Aware Prompt Enhancement, using GPT-4o to enhance the text prompt with facial details derived from the reference image. We then propose Prompt Aware Reference Image Enhancement, leveraging an identity-preserving image generator to refine the reference image, rectifying conflicts with the text prompt. The above mutual refinement significantly improves input quality before video generation. Finally, we propose ID-Aware Spatiotemporal Guidance Enhancement, utilizing unified gradients to optimize identity preservation and video quality jointly during generation.Our method outperforms prior work and is validated by automatic and human evaluations on a 1000 video test set, winning first place in the ACM Multimedia 2025 Identity-Preserving Video Generation Challenge, demonstrating state-of-the-art performance and strong generality. The code is available at https://github.com/Andyplus1/IPT2V.git.

Yang Liu, Muzhi Zhu, Hengtao Li et al.

Powered by large-scale pre-training, vision foundation models exhibit significant potential in open-world image understanding. However, unlike large language models that excel at directly tackling various language tasks, vision foundation models require a task-specific model structure followed by fine-tuning on specific tasks. In this work, we present Matcher, a novel perception paradigm that utilizes off-the-shelf vision foundation models to address various perception tasks. Matcher can segment anything by using an in-context example without training. Additionally, we design three effective components within the Matcher framework to collaborate with these foundation models and unleash their full potential in diverse perception tasks. Matcher demonstrates impressive generalization performance across various segmentation tasks, all without training. For example, it achieves 52.7% mIoU on COCO-20$^i$ with one example, surpassing the state-of-the-art specialist model by 1.6%. In addition, Matcher achieves 33.0% mIoU on the proposed LVIS-92$^i$ for one-shot semantic segmentation, outperforming the state-of-the-art generalist model by 14.4%. Our visualization results further showcase the open-world generality and flexibility of Matcher when applied to images in the wild. Our code can be found at https://github.com/aim-uofa/Matcher.

Yushen Wei, Yang Liu, Hong Yan et al.

Existing methods for video question answering (VideoQA) often suffer from spurious correlations between different modalities, leading to a failure in identifying the dominant visual evidence and the intended question. Moreover, these methods function as black boxes, making it difficult to interpret the visual scene during the QA process. In this paper, to discover critical video segments and frames that serve as the visual causal scene for generating reliable answers, we present a causal analysis of VideoQA and propose a framework for cross-modal causal relational reasoning, named Visual Causal Scene Refinement (VCSR). Particularly, a set of causal front-door intervention operations is introduced to explicitly find the visual causal scenes at both segment and frame levels. Our VCSR involves two essential modules: i) the Question-Guided Refiner (QGR) module, which refines consecutive video frames guided by the question semantics to obtain more representative segment features for causal front-door intervention; ii) the Causal Scene Separator (CSS) module, which discovers a collection of visual causal and non-causal scenes based on the visual-linguistic causal relevance and estimates the causal effect of the scene-separating intervention in a contrastive learning manner. Extensive experiments on the NExT-QA, Causal-VidQA, and MSRVTT-QA datasets demonstrate the superiority of our VCSR in discovering visual causal scene and achieving robust video question answering. The code is available at https://github.com/YangLiu9208/VCSR.

Simion-Vlad Bogolin, Ioana Croitoru, Hailin Jin et al.

Profiting from large-scale training datasets, advances in neural architecture design and efficient inference, joint embeddings have become the dominant approach for tackling cross-modal retrieval. In this work we first show that, despite their effectiveness, state-of-the-art joint embeddings suffer significantly from the longstanding "hubness problem" in which a small number of gallery embeddings form the nearest neighbours of many queries. Drawing inspiration from the NLP literature, we formulate a simple but effective framework called Querybank Normalisation (QB-Norm) that re-normalises query similarities to account for hubs in the embedding space. QB-Norm improves retrieval performance without requiring retraining. Differently from prior work, we show that QB-Norm works effectively without concurrent access to any test set queries. Within the QB-Norm framework, we also propose a novel similarity normalisation method, the Dynamic Inverted Softmax, that is significantly more robust than existing approaches. We showcase QB-Norm across a range of cross modal retrieval models and benchmarks where it consistently enhances strong baselines beyond the state of the art. Code is available at https://vladbogo.github.io/QB-Norm/.

Jiaheng Wei, Yuanshun Yao, Jean-Francois Ton et al.

LLM hallucination, i.e. generating factually incorrect yet seemingly convincing answers, is currently a major threat to the trustworthiness and reliability of LLMs. The first step towards solving this complicated problem is to measure it. However, existing hallucination metrics require having a benchmark dataset with gold-standard answers, i.e. "best" or "correct" answers written by humans. Such requirements make hallucination measurement costly and prone to human errors. In this work, we propose Factualness Evaluations via Weighting LLMs (FEWL), an innovative hallucination metric that is specifically designed for the scenario when gold-standard answers are absent. FEWL leverages the answers from off-the-shelf LLMs that serve as a proxy of gold-standard answers. The key challenge is how to quantify the expertise of reference LLMs resourcefully. We show FEWL has certain theoretical guarantees and demonstrate empirically it gives more accurate hallucination measures than naively using reference LLMs. We also show how to leverage FEWL to reduce hallucination through both in-context learning and supervised fine-tuning. Extensive experiment results on Truthful-QA, CHALE, and HaluEval datasets demonstrate the effectiveness of FEWL.

Yanxing Huang, Zihan Tang, Zejin Lin et al.

The key limitation of the verification performance lies in the ability of error detection. With this intuition we designed several variants of pessimistic verification, which are simple workflows that could significantly improve the verification of open-ended math questions. In pessimistic verification we construct multiple parallel verifications for the same proof, and the proof is deemed incorrect if any one of them reports an error. This simple technique significantly improves the performance across many math verification benchmarks without incurring substantial computational resources. Its token efficiency even surpassed extended long-CoT in test-time scaling. Our case studies further indicate that the majority of false negatives in stronger models are actually caused by annotation errors in the original dataset, so our method's performance is in fact underestimated. Self-verification for mathematical problems can effectively improve the reliability and performance of language model outputs, and it also plays a critical role in enabling long-horizon mathematical tasks. We believe that research on pessimistic verification will help enhance the mathematical capabilities of language models across a wide range of tasks.

Xinshuai Song, Weixing Chen, Yang Liu et al.

Existing Vision-Language Navigation (VLN) methods primarily focus on single-stage navigation, limiting their effectiveness in multi-stage and long-horizon tasks within complex and dynamic environments. To address these limitations, we propose a novel VLN task, named Long-Horizon Vision-Language Navigation (LH-VLN), which emphasizes long-term planning and decision consistency across consecutive subtasks. Furthermore, to support LH-VLN, we develop an automated data generation platform NavGen, which constructs datasets with complex task structures and improves data utility through a bidirectional, multi-granularity generation approach. To accurately evaluate complex tasks, we construct the Long-Horizon Planning and Reasoning in VLN (LHPR-VLN) benchmark consisting of 3,260 tasks with an average of 150 task steps, serving as the first dataset specifically designed for the long-horizon vision-language navigation task. Furthermore, we propose Independent Success Rate (ISR), Conditional Success Rate (CSR), and CSR weight by Ground Truth (CGT) metrics, to provide fine-grained assessments of task completion. To improve model adaptability in complex tasks, we propose a novel Multi-Granularity Dynamic Memory (MGDM) module that integrates short-term memory blurring with long-term memory retrieval to enable flexible navigation in dynamic environments. Our platform, benchmark and method supply LH-VLN with a robust data generation pipeline, comprehensive model evaluation dataset, reasonable metrics, and a novel VLN model, establishing a foundational framework for advancing LH-VLN.

Yang Liu, Daizong Liu, Wei Hu

This paper tackles the challenging task of 3D visual grounding-locating a specific object in a 3D point cloud scene based on text descriptions. Existing methods fall into two categories: top-down and bottom-up methods. Top-down methods rely on a pre-trained 3D detector to generate and select the best bounding box, resulting in time-consuming processes. Bottom-up methods directly regress object bounding boxes with coarse-grained features, producing worse results. To combine their strengths while addressing their limitations, we propose a joint top-down and bottom-up framework, aiming to enhance the performance while improving the efficiency. Specifically, in the first stage, we propose a bottom-up based proposal generation module, which utilizes lightweight neural layers to efficiently regress and cluster several coarse object proposals instead of using a complex 3D detector. Then, in the second stage, we introduce a top-down based proposal consolidation module, which utilizes graph design to effectively aggregate and propagate the query-related object contexts among the generated proposals for further refinement. By jointly training these two modules, we can avoid the inherent drawbacks of the complex proposals in the top-down framework and the coarse proposals in the bottom-up framework. Experimental results on the ScanRefer benchmark show that our framework is able to achieve the state-of-the-art performance.

Ruoxi Cheng, Haoxuan Ma, Weixin Wang et al.

Alignment is vital for safely deploying large language models (LLMs). Existing techniques are either reward-based (train a reward model on preference pairs and optimize with reinforcement learning) or reward-free (directly fine-tune on ranked outputs). Recent research shows that well-tuned reward-based pipelines remain robust, and single-response demonstrations can outperform pairwise preference data. However, two challenges persist: (1) imbalanced safety datasets that overrepresent common hazards while neglecting long-tail threats; and (2) static reward models that ignore task difficulty, limiting optimization efficiency and attainable gains. We propose DR-IRL (Dynamically adjusting Rewards through Inverse Reinforcement Learning). We first train category-specific reward models using a balanced safety dataset covering seven harmful categories via IRL. Then we enhance Group Relative Policy Optimization (GRPO) by introducing dynamic reward scaling--adjusting rewards by task difficulty--data-level hardness by text encoder cosine similarity, model-level responsiveness by reward gaps. Extensive experiments across various benchmarks and LLMs demonstrate that DR-IRL outperforms all baseline methods in safety alignment while maintaining usefulness.

Xinliang Zhou, Yuzhe Han, Zhisheng Chen et al.

In this paper, we address the challenges in automatic sleep stage classification, particularly the high computational cost, inadequate modeling of bidirectional temporal dependencies, and class imbalance issues faced by Transformer-based models. To address these limitations, we propose BiT-MamSleep, a novel architecture that integrates the Triple-Resolution CNN (TRCNN) for efficient multi-scale feature extraction with the Bidirectional Mamba (BiMamba) mechanism, which models both short- and long-term temporal dependencies through bidirectional processing of EEG data. Additionally, BiT-MamSleep incorporates an Adaptive Feature Recalibration (AFR) module and a temporal enhancement block to dynamically refine feature importance, optimizing classification accuracy without increasing computational complexity. To further improve robustness, we apply optimization techniques such as Focal Loss and SMOTE to mitigate class imbalance. Extensive experiments on four public datasets demonstrate that BiT-MamSleep significantly outperforms state-of-the-art methods, particularly in handling long EEG sequences and addressing class imbalance, leading to more accurate and scalable sleep stage classification.

Yunghwei Lai, Weizhi Ma, Yang Liu

Synthetic data generation using large language models (LLMs) has emerged as a promising solution across various domains, particularly in medical field, to mitigate data collection challenges. However, existing studies mainly utilize LLMs to rewrite and complete existing medical records, where the limitations in data privacy, accuracy, and diversity sill exist, and additionally lack the ability to interact like real patients. To address these issues, we propose a realistic patient generation framework, Patient-Zero, which requires no real medical records. Patient-Zero first introduces a medically-aligned multi-step generation architecture, which builds comprehensive patient records through hierarchical medical knowledge injection without real medical records. Then, to optimize the virtual patient's interaction abilities with humans, Patient-Zero designs a dynamic updating mechanism to improve the consistency and conversational performance. Our framework enables the generation of contextually diverse patient records while maintaining strict medical coherence, supported by adaptive dialogue strategies and real-time clinical plausibility verification. Experimental results demonstrate that our model achieves good performance in accuracy, diversity, and consistency. After training with our generated virtual patients, existing models show significant improvements on the MedQA dataset.

Xinhao Cai, Minghang Zheng, Xin Jin et al.

We propose a novel task of text-controlled human object interaction generation in 3D scenes with movable objects. Existing human-scene interaction datasets suffer from insufficient interaction categories and typically only consider interactions with static objects (do not change object positions), and the collection of such datasets with movable objects is difficult and costly. To address this problem, we construct the InteractMove dataset for Movable Human-Object Interaction in 3D Scenes by aligning existing human object interaction data with scene contexts, featuring three key characteristics: 1) scenes containing multiple movable objects with text-controlled interaction specifications (including same-category distractors requiring spatial and 3D scene context understanding), 2) diverse object types and sizes with varied interaction patterns (one-hand, two-hand, etc.), and 3) physically plausible object manipulation trajectories. With the introduction of various movable objects, this task becomes more challenging, as the model needs to identify objects to be interacted with accurately, learn to interact with objects of different sizes and categories, and avoid collisions between movable objects and the scene. To tackle such challenges, we propose a novel pipeline solution. We first use 3D visual grounding models to identify the interaction object. Then, we propose a hand-object joint affordance learning to predict contact regions for different hand joints and object parts, enabling accurate grasping and manipulation of diverse objects. Finally, we optimize interactions with local-scene modeling and collision avoidance constraints, ensuring physically plausible motions and avoiding collisions between objects and the scene. Comprehensive experiments demonstrate our method's superiority in generating physically plausible, text-compliant interactions compared to existing approaches.

Zijing Zhao, Zhu Xu, Qingchao Chen et al.

As a fundamental task for indoor scene understanding, 3D object detection has been extensively studied, and the accuracy on indoor point cloud data has been substantially improved. However, existing researches have been conducted on limited datasets, where the training and testing sets share the same distribution. In this paper, we consider the task of adapting indoor 3D object detectors from one dataset to another, presenting a comprehensive benchmark with ScanNet, SUN RGB-D and 3D Front datasets, as well as our newly proposed large-scale datasets ProcTHOR-OD and ProcFront generated by a 3D simulator. Since indoor point cloud datasets are collected and constructed in different ways, the object detectors are likely to overfit to specific factors within each dataset, such as point cloud quality, bounding box layout and instance features. We conduct experiments across datasets on different adaptation scenarios including synthetic-to-real adaptation, point cloud quality adaptation, layout adaptation and instance feature adaptation, analyzing the impact of different domain gaps on 3D object detectors. We also introduce several approaches to improve adaptation performances, providing baselines for domain adaptive indoor 3D object detection, hoping that future works may propose detectors with stronger generalization ability across domains. Our project homepage can be found in https://jeremyzhao1998.github.io/DAVoteNet-release/.

Dapeng Jiang, Xiangzhe Kong, Jiaqi Han et al.

Cyclic peptides, characterized by geometric constraints absent in linear peptides, offer enhanced biochemical properties, presenting new opportunities to address unmet medical needs. However, designing target-specific cyclic peptides remains underexplored due to limited training data. To bridge the gap, we propose CP-Composer, a novel generative framework that enables zero-shot cyclic peptide generation via composable geometric constraints. Our approach decomposes complex cyclization patterns into unit constraints, which are incorporated into a diffusion model through geometric conditioning on nodes and edges. During training, the model learns from unit constraints and their random combinations in linear peptides, while at inference, novel constraint combinations required for cyclization are imposed as input. Experiments show that our model, despite trained with linear peptides, is capable of generating diverse target-binding cyclic peptides, reaching success rates from 38% to 84% on different cyclization strategies.

Yuanhang Liu, Yanxing Huang, Yanqiao Wang et al. · tsinghua

Large Reasoning Models (LRMs) have made significant progress in mathematical capabilities in recent times. However, these successes have been primarily confined to competition-level problems. In this work, we propose AI Mathematician (AIM) framework, which harnesses the reasoning strength of LRMs to support frontier mathematical research. We have identified two critical challenges of mathematical research compared to competition, {\it the intrinsic complexity of research problems} and {\it the requirement of procedural rigor}. To address these challenges, AIM incorporates two core strategies: an exploration mechanism to foster longer solution paths, and the pessimistic reasonable verification method to ensure reliability. This early version of AIM already exhibits strong capability in tackling research-level tasks. We conducted extensive experiments across several real-world mathematical topics and obtained promising results. AIM is able to autonomously construct substantial portions of proofs and uncover non-trivial insights within each research area. These findings highlight the potential of LRMs in mathematical discovery and suggest that LRM-based agent systems could significantly accelerate mathematical research in the future.

Shanyu Han, Yang Liu, Xiang Yu

We propose a reinforcement learning (RL) framework under a broad class of risk objectives, characterized by convex scoring functions. This class covers many common risk measures, such as variance, Expected Shortfall, entropic Value-at-Risk, and mean-risk utility. To resolve the time-inconsistency issue, we consider an augmented state space and an auxiliary variable and recast the problem as a two-state optimization problem. We propose a customized Actor-Critic algorithm and establish some theoretical approximation guarantees. A key theoretical contribution is that our results do not require the Markov decision process to be continuous. Additionally, we propose an auxiliary variable sampling method inspired by the alternating minimization algorithm, which is convergent under certain conditions. We validate our approach in simulation experiments with a financial application in statistical arbitrage trading, demonstrating the effectiveness of the algorithm.

Ming-Jia Yang, Yu-Xiao Guo, Yang Liu et al.

Generating realistic, room-level indoor scenes with semantically plausible and detailed appearances from in-the-wild images is crucial for various applications in VR, AR, and robotics. The success of NeRF-based generative methods indicates a promising direction to address this challenge. However, unlike their success at the object level, existing scene-level generative methods require additional information, such as multiple views, depth images, or semantic guidance, rather than relying solely on RGB images. This is because NeRF-based methods necessitate prior knowledge of camera poses, which is challenging to approximate for indoor scenes due to the complexity of defining alignment and the difficulty of globally estimating poses from a single image, given the unseen parts behind the camera. To address this challenge, we redefine global poses within the framework of Local-Pose-Alignment (LPA) -- an anchor-based multi-local-coordinate system that uses a selected number of anchors as the roots of these coordinates. Building on this foundation, we introduce LPA-GAN, a novel NeRF-based generative approach that incorporates specific modifications to estimate the priors of camera poses under LPA. It also co-optimizes the pose predictor and scene generation processes. Our ablation study and comparisons with straightforward extensions of NeRF-based object generative methods demonstrate the effectiveness of our approach. Furthermore, visual comparisons with other techniques reveal that our method achieves superior view-to-view consistency and semantic normality.

Yang Liu, Peiran Wu, Jiayu Huo et al.

Unsupervised video-based surgical instrument segmentation has the potential to accelerate the adoption of robot-assisted procedures by reducing the reliance on manual annotations. However, the generally low quality of optical flow in endoscopic footage poses a great challenge for unsupervised methods that rely heavily on motion cues. To overcome this limitation, we propose a novel approach that pinpoints motion boundaries, regions with abrupt flow changes, while selectively discarding frames with globally low-quality flow and adapting to varying motion patterns. Experiments on the EndoVis2017 VOS and EndoVis2017 Challenge datasets show that our method achieves mean Intersection-over-Union (mIoU) scores of 0.75 and 0.72, respectively, effectively alleviating the constraints imposed by suboptimal optical flow. This enables a more scalable and robust surgical instrument segmentation solution in clinical settings. The code will be publicly released.

Dezhao Luo, Shaogang Gong, Jiabo Huang et al.

Video moment retrieval (VMR) aims to locate the most likely video moment(s) corresponding to a text query in untrimmed videos. Training of existing methods is limited by the lack of diverse and generalisable VMR datasets, hindering their ability to generalise moment-text associations to queries containing novel semantic concepts (unseen both visually and textually in a training source domain). For model generalisation to novel semantics, existing methods rely heavily on assuming to have access to both video and text sentence pairs from a target domain in addition to the source domain pair-wise training data. This is neither practical nor scalable. In this work, we introduce a more generalisable approach by assuming only text sentences describing new semantics are available in model training without having seen any videos from a target domain. To that end, we propose a Fine-grained Video Editing framework, termed FVE, that explores generative video diffusion to facilitate fine-grained video editing from the seen source concepts to the unseen target sentences consisting of new concepts. This enables generative hypotheses of unseen video moments corresponding to the novel concepts in the target domain. This fine-grained generative video diffusion retains the original video structure and subject specifics from the source domain while introducing semantic distinctions of unseen novel vocabularies in the target domain. A critical challenge is how to enable this generative fine-grained diffusion process to be meaningful in optimising VMR, more than just synthesising visually pleasing videos. We solve this problem by introducing a hybrid selection mechanism that integrates three quantitative metrics to selectively incorporate synthetic video moments (novel video hypotheses) as enlarged additions to the original source training data, whilst minimising potential ...

Chi Chen, Peng Li, Maosong Sun et al.

Weakly supervised vision-and-language pre-training (WVLP), which learns cross-modal representations with limited cross-modal supervision, has been shown to effectively reduce the data cost of pre-training while maintaining decent performance on downstream tasks. However, current WVLP methods use only local descriptions of images, i.e., object tags, as cross-modal anchors to construct weakly-aligned image-text pairs for pre-training. This affects the data quality and thus the effectiveness of pre-training. In this paper, we propose to directly take a small number of aligned image-text pairs as anchors, and represent each unaligned image and text by its similarities to these anchors, i.e., relative representations. We build a WVLP framework based on the relative representations, namely RELIT, which collects high-quality weakly-aligned image-text pairs from large-scale image-only and text-only data for pre-training through relative representation-based retrieval and generation. Experiments on four downstream tasks show that RELIT achieves new state-of-the-art results under the weakly supervised setting.

Daizong Liu, Xiaoye Qu, Pan Zhou et al.

This paper addresses temporal sentence grounding. Previous works typically solve this task by learning frame-level video features and align them with the textual information. A major limitation of these works is that they fail to distinguish ambiguous video frames with subtle appearance differences due to frame-level feature extraction. Recently, a few methods adopt Faster R-CNN to extract detailed object features in each frame to differentiate the fine-grained appearance similarities. However, the object-level features extracted by Faster R-CNN suffer from missing motion analysis since the object detection model lacks temporal modeling. To solve this issue, we propose a novel Motion-Appearance Reasoning Network (MARN), which incorporates both motion-aware and appearance-aware object features to better reason object relations for modeling the activity among successive frames. Specifically, we first introduce two individual video encoders to embed the video into corresponding motion-oriented and appearance-aspect object representations. Then, we develop separate motion and appearance branches to learn motion-guided and appearance-guided object relations, respectively. At last, both motion and appearance information from two branches are associated to generate more representative features for final grounding. Extensive experiments on two challenging datasets (Charades-STA and TACoS) show that our proposed MARN significantly outperforms previous state-of-the-art methods by a large margin.

Ayush Shrivastava, Karthik Gopalakrishnan, Yang Liu et al.

Interactive robots navigating photo-realistic environments need to be trained to effectively leverage and handle the dynamic nature of dialogue in addition to the challenges underlying vision-and-language navigation (VLN). In this paper, we present VISITRON, a multi-modal Transformer-based navigator better suited to the interactive regime inherent to Cooperative Vision-and-Dialog Navigation (CVDN). VISITRON is trained to: i) identify and associate object-level concepts and semantics between the environment and dialogue history, ii) identify when to interact vs. navigate via imitation learning of a binary classification head. We perform extensive pre-training and fine-tuning ablations with VISITRON to gain empirical insights and improve performance on CVDN. VISITRON's ability to identify when to interact leads to a natural generalization of the game-play mode introduced by Roman et al. (arXiv:2005.00728) for enabling the use of such models in different environments. VISITRON is competitive with models on the static CVDN leaderboard and attains state-of-the-art performance on the Success weighted by Path Length (SPL) metric.

Ioana Croitoru, Simion-Vlad Bogolin, Marius Leordeanu et al.

In recent years, considerable progress on the task of text-video retrieval has been achieved by leveraging large-scale pretraining on visual and audio datasets to construct powerful video encoders. By contrast, despite the natural symmetry, the design of effective algorithms for exploiting large-scale language pretraining remains under-explored. In this work, we are the first to investigate the design of such algorithms and propose a novel generalized distillation method, TeachText, which leverages complementary cues from multiple text encoders to provide an enhanced supervisory signal to the retrieval model. Moreover, we extend our method to video side modalities and show that we can effectively reduce the number of used modalities at test time without compromising performance. Our approach advances the state of the art on several video retrieval benchmarks by a significant margin and adds no computational overhead at test time. Last but not least, we show an effective application of our method for eliminating noise from retrieval datasets. Code and data can be found at https://www.robots.ox.ac.uk/~vgg/research/teachtext/.

Andreea-Maria Oncescu, João F. Henriques, Yang Liu et al.

We introduce QuerYD, a new large-scale dataset for retrieval and event localisation in video. A unique feature of our dataset is the availability of two audio tracks for each video: the original audio, and a high-quality spoken description of the visual content. The dataset is based on YouDescribe, a volunteer project that assists visually-impaired people by attaching voiced narrations to existing YouTube videos. This ever-growing collection of videos contains highly detailed, temporally aligned audio and text annotations. The content descriptions are more relevant than dialogue, and more detailed than previous description attempts, which can be observed to contain many superficial or uninformative descriptions. To demonstrate the utility of the QuerYD dataset, we show that it can be used to train and benchmark strong models for retrieval and event localisation. Data, code and models are made publicly available, and we hope that QuerYD inspires further research on video understanding with written and spoken natural language.

Yang Liu, Samuel Albanie, Arsha Nagrani et al.

The rapid growth of video on the internet has made searching for video content using natural language queries a significant challenge. Human-generated queries for video datasets `in the wild' vary a lot in terms of degree of specificity, with some queries describing specific details such as the names of famous identities, content from speech, or text available on the screen. Our goal is to condense the multi-modal, extremely high dimensional information from videos into a single, compact video representation for the task of video retrieval using free-form text queries, where the degree of specificity is open-ended. For this we exploit existing knowledge in the form of pre-trained semantic embeddings which include 'general' features such as motion, appearance, and scene features from visual content. We also explore the use of more 'specific' cues from ASR and OCR which are intermittently available for videos and find that these signals remain challenging to use effectively for retrieval. We propose a collaborative experts model to aggregate information from these different pre-trained experts and assess our approach empirically on five retrieval benchmarks: MSR-VTT, LSMDC, MSVD, DiDeMo, and ActivityNet. Code and data can be found at www.robots.ox.ac.uk/~vgg/research/collaborative-experts/. This paper contains a correction to results reported in the previous version.