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.

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.

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.

We propose ReKV, a novel training-free approach that enables efficient streaming video question-answering (StreamingVQA), by seamlessly integrating with existing Video Large Language Models (Video-LLMs). Traditional VideoQA systems struggle with long videos, as they must process entire videos before responding to queries, and repeat this process for each new question. In contrast, our approach analyzes long videos in a streaming manner, allowing for prompt responses as soon as user queries are received. Building on a common Video-LLM, we first incorporate a sliding-window attention mechanism, ensuring that input frames attend to a limited number of preceding frames, thereby reducing computational overhead. To prevent information loss, we store processed video key-value caches (KV-Caches) in RAM and disk, reloading them into GPU memory as needed. Additionally, we introduce a retrieval method that leverages an external retriever or the parameters within Video-LLMs to retrieve only query-relevant KV-Caches, ensuring both efficiency and accuracy in question answering. ReKV enables the separation of video analyzing and question-answering across different processes and GPUs, significantly enhancing the efficiency of StreamingVQA. Through comprehensive experimentation, we validate the efficacy and practicality of our approach, which significantly boosts efficiency and enhances applicability over existing VideoQA models.

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.

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.

iamese and Transformer trackers have demon strated exceptional performance in visual object tracking. These methods utilize initial and potentially online templates to locate the target in subsequent frames. Despite their success, these trackers are vulnerable to changes in the target's appearance due to slow template updates and interference from similar objects, resulting from the absence of scene information. To address these issues, we introduce a reference region within our tracker. The reference region is updated rapidly, providing short-term scene information. By associating the initial template, reference region, and current search region, we enhance the tracker's ability to adapt to changes in target appearance and discriminate between the target and other objects. Additionally, we propose a novel Reference-Enhance (RE) module, which aggregates contextually relevant information from the reference region to enhance the template feature. Extensive experiments show our method achieves state-of-the-art performance on six popular visual object tracking benchmarks while running at over 40 FPS.

With the prevalence of multimodal learning, camera-LiDAR fusion has gained popularity in 3D object detection. Many fusion approaches have been proposed, falling into two main categories: sparse-only or dense-only, differentiated by their feature representation within the fusion module. We analyze these approaches within a shared taxonomy, identifying two key challenges: (1) Sparse-only methodologies maintain 3D geometric prior but fail to capture the semantic richness from camera data, and (2) Dense-only strategies preserve semantic continuity at the expense of precise geometric information derived from LiDAR. Upon analysis, we deduce that due to their respective architectural designs, some degree of information loss is inevitable. To counteract this loss, we introduce Sparse Dense Fusion (SD-Fusion), an innovative framework combining both sparse and dense fusion modules via the Transformer architecture. The simple yet effective fusion strategy enhances semantic texture and simultaneously leverages spatial structure data. Employing our SD-Fusion strategy, we assemble two popular methods with moderate performance, achieving a 4.3% increase in mAP and a 2.5% rise in NDS, thus ranking first in the nuScenes benchmark. Comprehensive ablation studies validate the effectiveness of our approach and empirically support our findings.

Image-based virtual try-on focuses on changing the model's garment item to the target ones and preserving other visual features. To preserve the texture detail of the given in-shop garment, former methods use geometry-based methods (e.g., Thin-plate-spline interpolation) to realize garment warping. However, due to limited degree of freedom, geometry-based methods perform poorly when garment self-occlusion occurs, which is common in daily life. To address this challenge, we propose a novel occlusion-focused virtual try-on system. Compared to previous ones, our system contains three critical submodules, namely, Garment Part Modeling (GPM), a group of Garment Part Generators (GPGs), and Overlap Relation Estimator (ORE). GPM takes the pose landmarks as input, and progressively models the mask of body parts and garments. Based on these masks, GPGs are introduced to generate each garment part. Finally, ORE is proposed to model the overlap relationships between each garment part, and we bind the generated garments under the guidance of overlap relationships predicted by ORE. To make the most of extracted overlap relationships, we proposed an IoU-based hard example mining method for loss terms to handle the sparsity of the self-occlusion samples in the dataset. Furthermore, we introduce part affinity field as pose representation instead of landmark used widely by previous methods and achieve accuracy improvement on try-on layout estimation stage. We evaluate our model on the VITON dataset and found it can outperform previous approaches, especially on samples with garment self-occlusion.

In this work, we address the task of video background music generation. Some previous works achieve effective music generation but are unable to generate melodious music specifically for a given video, and none of them considers the video-music rhythmic consistency. To generate the background music that matches the given video, we first establish the rhythmic relationships between video and background music. In particular, we connect timing, motion speed, and motion saliency from video with beat, simu-note density, and simu-note strength from music, respectively. We then propose CMT, a Controllable Music Transformer that enables the local control of the aforementioned rhythmic features, as well as the global control of the music genre and the used instrument specified by users. Objective and subjective evaluations show that the generated background music has achieved satisfactory compatibility with the input videos, and at the same time, impressive music quality.