Multimodal Large Language Models (MLLMs) have revolutionized video understanding, yet are still limited by context length when processing long videos. Recent methods compress videos by leveraging visual redundancy uniformly, yielding promising results. Nevertheless, our quantitative analysis shows that redundancy varies significantly across time and model layers, necessitating a more flexible compression strategy. We propose AdaReTaKe, a training-free method that flexibly reduces visual redundancy by allocating compression ratios among time and layers with theoretical guarantees. Integrated into state-of-the-art MLLMs, AdaReTaKe improves processing capacity from 256 to 2048 frames while preserving critical information. Experiments on VideoMME, MLVU, LongVideoBench, and LVBench datasets demonstrate that AdaReTaKe outperforms existing methods by 2.3% and 2.8% for 7B and 72B models, respectively, with even greater improvements of 5.9% and 6.0% on the longest LVBench. Our code is available at https://github.com/SCZwangxiao/video-FlexReduc.git.

Inferring the affordance of an object and grasping it in a task-oriented manner is crucial for robots to successfully complete manipulation tasks. Affordance indicates where and how to grasp an object by taking its functionality into account, serving as the foundation for effective task-oriented grasping. However, current task-oriented methods often depend on extensive training data that is confined to specific tasks and objects, making it difficult to generalize to novel objects and complex scenes. In this paper, we introduce AffordGrasp, a novel open-vocabulary grasping framework that leverages the reasoning capabilities of vision-language models (VLMs) for in-context affordance reasoning. Unlike existing methods that rely on explicit task and object specifications, our approach infers tasks directly from implicit user instructions, enabling more intuitive and seamless human-robot interaction in everyday scenarios. Building on the reasoning outcomes, our framework identifies task-relevant objects and grounds their part-level affordances using a visual grounding module. This allows us to generate task-oriented grasp poses precisely within the affordance regions of the object, ensuring both functional and context-aware robotic manipulation. Extensive experiments demonstrate that AffordGrasp achieves state-of-the-art performance in both simulation and real-world scenarios, highlighting the effectiveness of our method. We believe our approach advances robotic manipulation techniques and contributes to the broader field of embodied AI. Project website: https://eqcy.github.io/affordgrasp/.

Recent Multi-modal Large Language Models (MLLMs) have made great progress in video understanding. However, their performance on videos involving human actions is still limited by the lack of high-quality data. To address this, we introduce a two-stage data annotation pipeline. First, we design strategies to accumulate videos featuring clear human actions from the Internet. Second, videos are annotated in a standardized caption format that uses human attributes to distinguish individuals and chronologically details their actions and interactions. Through this pipeline, we curate two datasets, namely HAICTrain and HAICBench. \textbf{HAICTrain} comprises 126K video-caption pairs generated by Gemini-Pro and verified for training purposes. Meanwhile, \textbf{HAICBench} includes 500 manually annotated video-caption pairs and 1,400 QA pairs, for a comprehensive evaluation of human action understanding. Experimental results demonstrate that training with HAICTrain not only significantly enhances human understanding abilities across 4 benchmarks, but can also improve text-to-video generation results. Both the HAICTrain and HAICBench are released at https://huggingface.co/datasets/KuaishouHAIC/HAIC.

Video Large Language Models (VideoLLMs) have achieved remarkable progress in video understanding. However, existing VideoLLMs often inherit the limitations of their backbone LLMs in handling long sequences, leading to challenges for long video understanding. Common solutions either simply uniformly sample videos' frames or compress visual tokens, which focus primarily on low-level temporal visual redundancy, overlooking high-level knowledge redundancy. This limits the achievable compression rate with minimal loss. To this end. we introduce a training-free method, $\textbf{ReTaKe}$, containing two novel modules DPSelect and PivotKV, to jointly model and reduce both temporal visual redundancy and knowledge redundancy for long video understanding. Specifically, DPSelect identifies keyframes with local maximum peak distance based on their visual features, which are closely aligned with human video perception. PivotKV employs the obtained keyframes as pivots and conducts KV-Cache compression for the non-pivot tokens with low attention scores, which are derived from the learned prior knowledge of LLMs. Experiments on benchmarks VideoMME, MLVU, and LVBench, show that ReTaKe can support 4x longer video sequences with minimal performance loss (<1%) and outperform all similar-size VideoLLMs with 3%-5%, even surpassing or on par with much larger ones. Our code is available at https://github.com/SCZwangxiao/video-ReTaKe

Multimodal Large Language Models (MLLMs) have recently received substantial interest, which shows their emerging potential as general-purpose models for various vision-language tasks. MLLMs involve significant external knowledge within their parameters; however, it is challenging to continually update these models with the latest knowledge, which involves huge computational costs and poor interpretability. Retrieval augmentation techniques have proven to be effective plugins for both LLMs and MLLMs. In this study, we propose multimodal adaptive Retrieval-Augmented Bootstrapping Language-Image Pre-training (RA-BLIP), a novel retrieval-augmented framework for various MLLMs. Considering the redundant information within vision modality, we first leverage the question to instruct the extraction of visual information through interactions with one set of learnable queries, minimizing irrelevant interference during retrieval and generation. Besides, we introduce a pre-trained multimodal adaptive fusion module to achieve question text-to-multimodal retrieval and integration of multimodal knowledge by projecting visual and language modalities into a unified semantic space. Furthermore, we present an Adaptive Selection Knowledge Generation (ASKG) strategy to train the generator to autonomously discern the relevance of retrieved knowledge, which realizes excellent denoising performance. Extensive experiments on open multimodal question-answering datasets demonstrate that RA-BLIP achieves significant performance and surpasses the state-of-the-art retrieval-augmented models.

Knowledge distillation is a mainstream algorithm in model compression by transferring knowledge from the larger model (teacher) to the smaller model (student) to improve the performance of student. Despite many efforts, existing methods mainly investigate the consistency between instance-level feature representation or prediction, which neglects the category-level information and the difficulty of each sample, leading to undesirable performance. To address these issues, we propose a novel preview-based category contrastive learning method for knowledge distillation (PCKD). It first distills the structural knowledge of both instance-level feature correspondence and the relation between instance features and category centers in a contrastive learning fashion, which can explicitly optimize the category representation and explore the distinct correlation between representations of instances and categories, contributing to discriminative category centers and better classification results. Besides, we introduce a novel preview strategy to dynamically determine how much the student should learn from each sample according to their difficulty. Different from existing methods that treat all samples equally and curriculum learning that simply filters out hard samples, our method assigns a small weight for hard instances as a preview to better guide the student training. Extensive experiments on several challenging datasets, including CIFAR-100 and ImageNet, demonstrate the superiority over state-of-the-art methods.

Recently, video-language understanding has achieved great success through large-scale pre-training. However, data scarcity remains a prevailing challenge. This study quantitatively reveals an "impossible trinity" among data quantity, diversity, and quality in pre-training datasets. Recent efforts seek to refine large-scale, diverse ASR datasets compromised by low quality through synthetic annotations. These methods successfully leverage useful information in multimodal video content (frames, tags, ASR transcripts, etc.) to refine the original annotations. Nevertheless, they struggle to mitigate noise within synthetic annotations and lack scalability as the dataset size expands. To address these issues, we introduce the Video DataFlywheel framework, which iteratively refines video annotations with improved noise control methods. For iterative refinement, we first leverage a video-language model to generate synthetic annotations, resulting in a refined dataset. Then, we pre-train on it and fine-tune on human refinement examples for a stronger model. These processes are repeated for continuous improvement. For noise control, we present AdaTaiLr, a novel noise control method that requires weaker assumptions on noise distribution, thereby proving more effective in large datasets with theoretical guarantees. The combination of iterative refinement and AdaTaiLr can achieve better scalability in video-language understanding. Extensive experiments show that our framework outperforms existing data refinement baselines, delivering a 3% performance boost and improving dataset quality with minimal diversity loss. Furthermore, our refined dataset facilitates significant improvements in various video-language understanding tasks, including video question answering and text-video retrieval.

Current parameter-efficient fine-tuning (PEFT) methods build adapters without considering the context of downstream task to learn, or the context of important knowledge to maintain. As a result, there is often a performance gap compared to full-parameter finetuning, and meanwhile the finetuned model suffers from catastrophic forgetting of the pre-trained world knowledge. In this paper, we propose CorDA, a Context-oriented Decomposition Adaptation method that builds learnable adapters from weight decomposition oriented by the context of downstream task or world knowledge. Concretely, we collect a few data samples, and perform singular value decomposition for each linear layer of a pre-trained LLM multiplied by the covariance matrix of the input activation using these samples. By doing so, the context of the representative samples is captured through deciding the factorizing orientation. Our method enables two options, the knowledge-preserved adaptation and the instruction-previewed adaptation. For the former, we use question-answering samples to obtain the covariance matrices, and use the decomposed components with the smallest $r$ singular values to initialize a learnable adapter, with the others frozen such that the world knowledge is better preserved. For the latter, we use the instruction data from the finetuning task, such as math or coding, to orientate the decomposition and train the largest $r$ components that capture the main characteristics of the task to learn. We conduct extensive experiments on Math, Code, and Instruction Following tasks. Our knowledge-preserved adaptation not only achieves better performance than LoRA on finetuning tasks, but also mitigates the forgetting of world knowledge. Our instruction-previewed adaptation is able to further enhance the finetuning performance, surpassing full-parameter finetuning and the state-of-the-art PEFT methods.

Large Language Models (LLMs) face significant deployment challenges due to their substantial memory requirements and the computational demands of auto-regressive text generation process. This paper addresses these challenges by focusing on the quantization of LLMs, a technique that reduces memory consumption by converting model parameters and activations into low-bit integers. We critically analyze the existing quantization approaches, identifying their limitations in balancing the accuracy and efficiency of the quantized LLMs. To advance beyond these limitations, we propose WKVQuant, a PTQ framework especially designed for quantizing weights and the key/value (KV) cache of LLMs. Specifically, we incorporates past-only quantization to improve the computation of attention. Additionally, we introduce two-dimensional quantization strategy to handle the distribution of KV cache, along with a cross-block reconstruction regularization for parameter optimization. Experiments show that WKVQuant achieves almost comparable memory savings to weight-activation quantization, while also approaching the performance of weight-only quantization.

This paper aims to tackle a novel task - Temporal Sentence Grounding in Streaming Videos (TSGSV). The goal of TSGSV is to evaluate the relevance between a video stream and a given sentence query. Unlike regular videos, streaming videos are acquired continuously from a particular source, and are always desired to be processed on-the-fly in many applications such as surveillance and live-stream analysis. Thus, TSGSV is challenging since it requires the model to infer without future frames and process long historical frames effectively, which is untouched in the early methods. To specifically address the above challenges, we propose two novel methods: (1) a TwinNet structure that enables the model to learn about upcoming events; and (2) a language-guided feature compressor that eliminates redundant visual frames and reinforces the frames that are relevant to the query. We conduct extensive experiments using ActivityNet Captions, TACoS, and MAD datasets. The results demonstrate the superiority of our proposed methods. A systematic ablation study also confirms their effectiveness.