Sirui Zhao

Shukang Yin, Chaoyou Fu, Sirui Zhao et al. · tencent-ai

Recently, Multimodal Large Language Model (MLLM) represented by GPT-4V has been a new rising research hotspot, which uses powerful Large Language Models (LLMs) as a brain to perform multimodal tasks. The surprising emergent capabilities of MLLM, such as writing stories based on images and OCR-free math reasoning, are rare in traditional multimodal methods, suggesting a potential path to artificial general intelligence. To this end, both academia and industry have endeavored to develop MLLMs that can compete with or even better than GPT-4V, pushing the limit of research at a surprising speed. In this paper, we aim to trace and summarize the recent progress of MLLMs. First of all, we present the basic formulation of MLLM and delineate its related concepts, including architecture, training strategy and data, as well as evaluation. Then, we introduce research topics about how MLLMs can be extended to support more granularity, modalities, languages, and scenarios. We continue with multimodal hallucination and extended techniques, including Multimodal ICL (M-ICL), Multimodal CoT (M-CoT), and LLM-Aided Visual Reasoning (LAVR). To conclude the paper, we discuss existing challenges and point out promising research directions. In light of the fact that the era of MLLM has only just begun, we will keep updating this survey and hope it can inspire more research. An associated GitHub link collecting the latest papers is available at https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models.

Shukang Yin, Chaoyou Fu, Sirui Zhao et al.

Hallucination is a big shadow hanging over the rapidly evolving Multimodal Large Language Models (MLLMs), referring to the phenomenon that the generated text is inconsistent with the image content. In order to mitigate hallucinations, existing studies mainly resort to an instruction-tuning manner that requires retraining the models with specific data. In this paper, we pave a different way, introducing a training-free method named Woodpecker. Like a woodpecker heals trees, it picks out and corrects hallucinations from the generated text. Concretely, Woodpecker consists of five stages: key concept extraction, question formulation, visual knowledge validation, visual claim generation, and hallucination correction. Implemented in a post-remedy manner, Woodpecker can easily serve different MLLMs, while being interpretable by accessing intermediate outputs of the five stages. We evaluate Woodpecker both quantitatively and qualitatively and show the huge potential of this new paradigm. On the POPE benchmark, our method obtains a 30.66%/24.33% improvement in accuracy over the baseline MiniGPT-4/mPLUG-Owl. The source code is released at https://github.com/BradyFU/Woodpecker.

Mingjia Yin, Hao Wang, Xiang Xu et al.

The sequential recommendation system has been widely studied for its promising effectiveness in capturing dynamic preferences buried in users' sequential behaviors. Despite the considerable achievements, existing methods usually focus on intra-sequence modeling while overlooking exploiting global collaborative information by inter-sequence modeling, resulting in inferior recommendation performance. Therefore, previous works attempt to tackle this problem with a global collaborative item graph constructed by pre-defined rules. However, these methods neglect two crucial properties when capturing global collaborative information, i.e., adaptiveness and personalization, yielding sub-optimal user representations. To this end, we propose a graph-driven framework, named Adaptive and Personalized Graph Learning for Sequential Recommendation (APGL4SR), that incorporates adaptive and personalized global collaborative information into sequential recommendation systems. Specifically, we first learn an adaptive global graph among all items and capture global collaborative information with it in a self-supervised fashion, whose computational burden can be further alleviated by the proposed SVD-based accelerator. Furthermore, based on the graph, we propose to extract and utilize personalized item correlations in the form of relative positional encoding, which is a highly compatible manner of personalizing the utilization of global collaborative information. Finally, the entire framework is optimized in a multi-task learning paradigm, thus each part of APGL4SR can be mutually reinforced. As a generic framework, APGL4SR can outperform other baselines with significant margins. The code is available at https://github.com/Graph-Team/APGL4SR.

Shukang Yin, Shiwei Wu, Tong Xu et al.

Automatic Micro-Expression (ME) spotting in long videos is a crucial step in ME analysis but also a challenging task due to the short duration and low intensity of MEs. When solving this problem, previous works generally lack in considering the structures of human faces and the correspondence between expressions and relevant facial muscles. To address this issue for better performance of ME spotting, this paper seeks to extract finer spatial features by modeling the relationships between facial Regions of Interest (ROIs). Specifically, we propose a graph convolutional-based network, called Action-Unit-aWare Graph Convolutional Network (AUW-GCN). Furthermore, to inject prior information and to cope with the problem of small datasets, AU-related statistics are encoded into the network. Comprehensive experiments show that our results outperform baseline methods consistently and achieve new SOTA performance in two benchmark datasets,CAS(ME)^2 and SAMM-LV. Our code is available at https://github.com/xjtupanda/AUW-GCN.

Chao Zhang, Shiwei Wu, Sirui Zhao et al.

Affordance-centric Question-driven Task Completion (AQTC) for Egocentric Assistant introduces a groundbreaking scenario. In this scenario, through learning instructional videos, AI assistants provide users with step-by-step guidance on operating devices. In this paper, we present a solution for enhancing video alignment to improve multi-step inference. Specifically, we first utilize VideoCLIP to generate video-script alignment features. Afterwards, we ground the question-relevant content in instructional videos. Then, we reweight the multimodal context to emphasize prominent features. Finally, we adopt GRU to conduct multi-step inference. Through comprehensive experiments, we demonstrate the effectiveness and superiority of our method, which secured the 2nd place in CVPR'2023 AQTC challenge. Our code is available at https://github.com/zcfinal/LOVEU-CVPR23-AQTC.

Xianglong Shi, Silin Cheng, Sirui Zhao et al.

Existing Weakly-Supervised Referring Expression Comprehension (WREC) methods, while effective, are fundamentally limited by a one-to-one mapping assumption, hindering their ability to handle expressions corresponding to zero or multiple targets in realistic scenarios. To bridge this gap, we introduce the Weakly-Supervised Generalized Referring Expression Comprehension task (WGREC), a more practical paradigm that handles expressions with variable numbers of referents. However, extending WREC to WGREC presents two fundamental challenges: supervisory signal ambiguity, where weak image-level supervision is insufficient for training a model to infer the correct number and identity of referents, and semantic representation collapse, where standard Euclidean similarity forces hierarchically-related concepts into non-discriminative clusters, blurring categorical boundaries. To tackle these challenges, we propose a novel WGREC framework named Linguistic Instance-Split Hyperbolic-Euclidean (LIHE), which operates in two stages. The first stage, Referential Decoupling, predicts the number of target objects and decomposes the complex expression into simpler sub-expressions. The second stage, Referent Grounding, then localizes these sub-expressions using HEMix, our innovative hybrid similarity module that synergistically combines the precise alignment capabilities of Euclidean proximity with the hierarchical modeling strengths of hyperbolic geometry. This hybrid approach effectively prevents semantic collapse while preserving fine-grained distinctions between related concepts. Extensive experiments demonstrate LIHE establishes the first effective weakly supervised WGREC baseline on gRefCOCO and Ref-ZOM, while HEMix achieves consistent improvements on standard REC benchmarks, improving IoU@0.5 by up to 2.5\%. The code is available at https://anonymous.4open.science/r/LIHE.

Sirui Zhao, Huaying Tang, Xinglong Mao et al.

One of the most important subconscious reactions, micro-expression (ME), is a spontaneous, subtle, and transient facial expression that reveals human beings' genuine emotion. Therefore, automatically recognizing ME (MER) is becoming increasingly crucial in the field of affective computing, providing essential technical support for lie detection, clinical psychological diagnosis, and public safety. However, the ME data scarcity has severely hindered the development of advanced data-driven MER models. Despite the recent efforts by several spontaneous ME databases to alleviate this problem, there is still a lack of sufficient data. Hence, in this paper, we overcome the ME data scarcity problem by collecting and annotating a dynamic spontaneous ME database with the largest current ME data scale called DFME (Dynamic Facial Micro-expressions). Specifically, the DFME database contains 7,526 well-labeled ME videos spanning multiple high frame rates, elicited by 671 participants and annotated by more than 20 professional annotators over three years. Furthermore, we comprehensively verify the created DFME, including using influential spatiotemporal video feature learning models and MER models as baselines, and conduct emotion classification and ME action unit classification experiments. The experimental results demonstrate that the DFME database can facilitate research in automatic MER, and provide a new benchmark for this field. DFME will be published via https://mea-lab-421.github.io.

Ji-Xuan He, Jia-Cheng Zhao, Feng-Qi Cui et al.

Low-light image super-resolution (LLISR) is essential for restoring fine visual details and perceptual quality under insufficient illumination conditions with ubiquitous low-resolution devices. Although pioneer methods achieve high performance on single tasks, they solve both tasks in a serial manner, which inevitably leads to artifact amplification, texture suppression, and structural degradation. To address this, we propose Decoupling then Perceive (DTP), a novel frequency-aware framework that explicitly separates luminance and texture into semantically independent components, enabling specialized modeling and coherent reconstruction. Specifically, to adaptively separate the input into low-frequency luminance and high-frequency texture subspaces, we propose a Frequency-aware Structural Decoupling (FSD) mechanism, which lays a solid foundation for targeted representation learning and reconstruction. Based on the decoupled representation, a Semantics-specific Dual-path Representation (SDR) learning strategy that performs targeted enhancement and reconstruction for each frequency component is further designed, facilitating robust luminance adjustment and fine-grained texture recovery. To promote structural consistency and perceptual alignment in the reconstructed output, building upon this dual-path modeling, we further introduce a Cross-frequency Semantic Recomposition (CSR) module that selectively integrates the decoupled representations. Extensive experiments on the most widely used LLISR benchmarks demonstrate the superiority of our DTP framework, improving $+$1.6\% PSNR, $+$9.6\% SSIM, and $-$48\% LPIPS compared to the most state-of-the-art (SOTA) algorithm. Codes are released at https://github.com/JXVision/DTP.

Yubo Huang, Hailong Guo, Fangtai Wu et al.

Existing diffusion-based video generation methods are fundamentally constrained by sequential computation and long-horizon inconsistency, limiting their practical adoption in real-time, streaming audio-driven avatar synthesis. We present Live Avatar, an algorithm-system co-designed framework that enables efficient, high-fidelity, and infinite-length avatar generation using a 14-billion-parameter diffusion model. Our approach introduces Timestep-forcing Pipeline Parallelism (TPP), a distributed inference paradigm that pipelines denoising steps across multiple GPUs, effectively breaking the autoregressive bottleneck and ensuring stable, low-latency real-time streaming. To further enhance temporal consistency and mitigate identity drift and color artifacts, we propose the Rolling Sink Frame Mechanism (RSFM), which maintains sequence fidelity by dynamically recalibrating appearance using a cached reference image. Additionally, we leverage Self-Forcing Distribution Matching Distillation to facilitate causal, streamable adaptation of large-scale models without sacrificing visual quality. Live Avatar demonstrates state-of-the-art performance, reaching 20 FPS end-to-end generation on 5 H800 GPUs, and, to the best of our knowledge, is the first to achieve practical, real-time, high-fidelity avatar generation at this scale. Our work establishes a new paradigm for deploying advanced diffusion models in industrial long-form video synthesis applications.

Chaoyou Fu, Renrui Zhang, Zihan Wang et al.

The surge of interest towards Multi-modal Large Language Models (MLLMs), e.g., GPT-4V(ision) from OpenAI, has marked a significant trend in both academia and industry. They endow Large Language Models (LLMs) with powerful capabilities in visual understanding, enabling them to tackle diverse multi-modal tasks. Very recently, Google released Gemini, its newest and most capable MLLM built from the ground up for multi-modality. In light of the superior reasoning capabilities, can Gemini challenge GPT-4V's leading position in multi-modal learning? In this paper, we present a preliminary exploration of Gemini Pro's visual understanding proficiency, which comprehensively covers four domains: fundamental perception, advanced cognition, challenging vision tasks, and various expert capacities. We compare Gemini Pro with the state-of-the-art GPT-4V to evaluate its upper limits, along with the latest open-sourced MLLM, Sphinx, which reveals the gap between manual efforts and black-box systems. The qualitative samples indicate that, while GPT-4V and Gemini showcase different answering styles and preferences, they can exhibit comparable visual reasoning capabilities, and Sphinx still trails behind them concerning domain generalizability. Specifically, GPT-4V tends to elaborate detailed explanations and intermediate steps, and Gemini prefers to output a direct and concise answer. The quantitative evaluation on the popular MME benchmark also demonstrates the potential of Gemini to be a strong challenger to GPT-4V. Our early investigation of Gemini also observes some common issues of MLLMs, indicating that there still remains a considerable distance towards artificial general intelligence. Our project for tracking the progress of MLLM is released at https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models.

Fangyuan Liu, Sirui Zhao, Zeyu Zhang et al.

Automated depression estimation is highly vulnerable to signal corruption and ambient noise in real-world deployment. Prevailing deterministic methods produce uncalibrated point estimates, exposing safety-critical clinical systems to the severe risk of overconfident misdiagnoses. To establish a highly resilient and trustworthy assessment paradigm, we propose EviDep, an evidential learning framework that jointly quantifies depression severity alongside aleatoric and epistemic uncertainties via a Normal-Inverse-Gamma distribution. A fundamental vulnerability in multimodal evidential fusion is the uncontrolled accumulation of cross-modal redundancies. This structural flaw artificially inflates diagnostic confidence by double-counting overlapping evidence. To guarantee robust evidence synthesis, EviDep enforces strict information integrity. First, a Frequency-aware Feature Extraction module leverages a wavelet-based Mixture-of-Experts to dynamically isolate task-irrelevant noise, preserving the fidelity of diagnostic signals. Subsequently, a Disentangled Evidential Learning strategy separates the shared consensus from modality-specific nuances. By explicitly decoupling these representations before Bayesian fusion, EviDep systematically mitigates evidence redundancy. Extensive experiments on AVEC 2013, 2014, DAIC-WOZ, and E-DAIC confirm that EviDep achieves state-of-the-art predictive accuracy and superior uncertainty calibration, delivering a robust fail-safe mechanism for trustworthy clinical screening.

Yubo Huang, Weiqiang Wang, Sirui Zhao et al.

Recent years have witnessed remarkable advances in audio-driven talking head generation. However, existing approaches predominantly focus on single-character scenarios. While some methods can create separate conversation videos between two individuals, the critical challenge of generating unified conversation videos with multiple physically co-present characters sharing the same spatial environment remains largely unaddressed. This setting presents two key challenges: audio-to-character correspondence control and the lack of suitable datasets featuring multi-character talking videos within the same scene. To address these challenges, we introduce Bind-Your-Avatar, an MM-DiT-based model specifically designed for multi-talking-character video generation in the same scene. Specifically, we propose (1) A novel framework incorporating a fine-grained Embedding Router that binds `who' and `speak what' together to address the audio-to-character correspondence control. (2) Two methods for implementing a 3D-mask embedding router that enables frame-wise, fine-grained control of individual characters, with distinct loss functions based on observed geometric priors and a mask refinement strategy to enhance the accuracy and temporal smoothness of the predicted masks. (3) The first dataset, to the best of our knowledge, specifically constructed for multi-talking-character video generation, and accompanied by an open-source data processing pipeline, and (4) A benchmark for the dual-talking-characters video generation, with extensive experiments demonstrating superior performance over multiple state-of-the-art methods.

Wenli Zhang, Xianglong Shi, Sirui Zhao et al.

Diffusion-based audio-driven talking-head generation enables realistic portrait animation, but also introduces risks of misuse, such as fraud and misinformation. Existing protection methods are largely limited to a single modality, and neither image-only nor audio-only attacks can effectively suppress speech-driven facial dynamics. To address this gap, we propose SyncBreaker, a stage-aware multimodal protection framework that jointly perturbs portrait and audio inputs under modality-specific perceptual constraints. Our key contributions are twofold. First, for the image stream, we introduce nullifying supervision with Multi-Interval Sampling (MIS) across diffusion stages to steer the generation toward the static reference portrait by aggregating guidance from multiple denoising intervals. Second, for the audio stream, we propose Cross-Attention Fooling (CAF), which suppresses interval-specific audio-conditioned cross-attention responses. Both streams are optimized independently and combined at inference time to enable flexible deployment. We evaluate SyncBreaker in a white-box proactive protection setting. Extensive experiments demonstrate that SyncBreaker more effectively degrades lip synchronization and facial dynamics than strong single-modality baselines, while preserving input perceptual quality and remaining robust under purification. Code: https://github.com/kitty384/SyncBreaker.

Shifeng Liu, Zhengye Zhang, Sirui Zhao et al.

Recent advances in Multimodal Large Language Models (MLLMs) have created new opportunities for facial expression recognition (FER), moving it beyond pure label prediction toward reasoning-based affect understanding. However, existing MLLM-based FER methods still follow a passive paradigm: they rely on externally prepared facial inputs and perform single-pass reasoning over fixed visual evidence, without the capability for active facial perception. To address this limitation, we propose ActFER, an agentic framework that reformulates FER as active visual evidence acquisition followed by multimodal reasoning. Specifically, ActFER dynamically invokes tools for face detection and alignment, selectively zooms into informative local regions, and reasons over facial Action Units (AUs) and emotions through a visual Chain-of-Thought. To realize such behavior, we further develop Utility-Calibrated GRPO (UC-GRPO), a reinforcement learning algorithm tailored to agentic FER. UC-GRPO uses AU-grounded multi-level verifiable rewards to densify supervision, query-conditional contrastive utility estimation to enable sample-aware dynamic credit assignment for local inspection, and emotion-aware EMA calibration to reduce noisy utility estimates while capturing emotion-wise inspection tendencies. This algorithm enables ActFER to learn both when local inspection is beneficial and how to reason over the acquired evidence. Comprehensive experiments show that ActFER trained with UC-GRPO consistently outperforms passive MLLM-based FER baselines and substantially improves AU prediction accuracy.

Tingjia Shen, Hao Wang, Jiaqing Zhang et al.

Cross-Domain Sequential Recommendation (CDSR) aims to mine and transfer users' sequential preferences across different domains to alleviate the long-standing cold-start issue. Traditional CDSR models capture collaborative information through user and item modeling while overlooking valuable semantic information. Recently, Large Language Model (LLM) has demonstrated powerful semantic reasoning capabilities, motivating us to introduce them to better capture semantic information. However, introducing LLMs to CDSR is non-trivial due to two crucial issues: seamless information integration and domain-specific generation. To this end, we propose a novel framework named URLLM, which aims to improve the CDSR performance by exploring the User Retrieval approach and domain grounding on LLM simultaneously. Specifically, we first present a novel dual-graph sequential model to capture the diverse information, along with an alignment and contrastive learning method to facilitate domain knowledge transfer. Subsequently, a user retrieve-generation model is adopted to seamlessly integrate the structural information into LLM, fully harnessing its emergent inferencing ability. Furthermore, we propose a domain-specific strategy and a refinement module to prevent out-of-domain generation. Extensive experiments on Amazon demonstrated the information integration and domain-specific generation ability of URLLM in comparison to state-of-the-art baselines. Our code is available at https://github.com/TingJShen/URLLM

Shukang Yin, Chaoyou Fu, Sirui Zhao et al.

Recent years have seen the success of Multimodal Large Language Models (MLLMs) in the domain of vision understanding. The success of these models can largely be attributed to the dominant scaling law, which states that larger parameter sizes and data volumes contribute to better performance. Notably, data scaling has been primarily driven by automatic data pipelines, which focus on the self-instruction of LLMs. The paradigm has been taken for granted for quite some time, but the study of the effectiveness of scaling with these data has been neglected for a long time. In this context, this work revisits scaling with synthetic data and focuses on developing video-LLMs from a data-centric perspective. Our primary study approach involves fine-tuning pre-trained image-LLMs with video data and examining learning efficiency through data scaling. Results from our preliminary experiments reveal a low learning efficiency phenomenon when simply scaling up video data samples, which, through our probing, can be ascribed to a lack of instruction diversity. Aiming at this issue, we propose a data augmentation method called Sparrow, which synthesizes video-like samples from pure text instruction data. Mixing these synthetic samples with the video data enables a more efficient training scheme. Through comprehensive experiments, we demonstrate that our proposed method achieves performance comparable to or even superior to that of baselines trained with significantly more samples. Meanwhile, we find that incorporating these synthetic samples can enhance the performance of long video understanding without requiring training on long video data. The code and data examples are available at https://github.com/VITA-MLLM/Sparrow.

Chaoyou Fu, Yi-Fan Zhang, Shukang Yin et al. · pku

As a prominent direction of Artificial General Intelligence (AGI), Multimodal Large Language Models (MLLMs) have garnered increased attention from both industry and academia. Building upon pre-trained LLMs, this family of models further develops multimodal perception and reasoning capabilities that are impressive, such as writing code given a flow chart or creating stories based on an image. In the development process, evaluation is critical since it provides intuitive feedback and guidance on improving models. Distinct from the traditional train-eval-test paradigm that only favors a single task like image classification, the versatility of MLLMs has spurred the rise of various new benchmarks and evaluation methods. In this paper, we aim to present a comprehensive survey of MLLM evaluation, discussing four key aspects: 1) the summarised benchmarks types divided by the evaluation capabilities, including foundation capabilities, model self-analysis, and extented applications; 2) the typical process of benchmark counstruction, consisting of data collection, annotation, and precautions; 3) the systematic evaluation manner composed of judge, metric, and toolkit; 4) the outlook for the next benchmark. This work aims to offer researchers an easy grasp of how to effectively evaluate MLLMs according to different needs and to inspire better evaluation methods, thereby driving the progress of MLLM research.

Feng-Qi Cui, Jinyang Huang, Sirui Zhao et al.

Video behavior recognition demands stable and discriminative representations under complex spatiotemporal variations. However, prevailing data augmentation strategies for videos remain largely perturbation-driven, often introducing uncontrolled variations that amplify non-discriminative factors, which finally weaken intra-class distributional structure and representation drift with inconsistent gains across temporal scales. To address these problems, we propose Representation-aware Mixing Augmentation (ReMA), a plug-and-play augmentation strategy that formulates mixing as a controlled replacement process to expand representations while preserving class-conditional stability. ReMA integrates two complementary mechanisms. Firstly, the Representation Alignment Mechanism (RAM) performs structured intra-class mixing under distributional alignment constraints, suppressing irrelevant intra-class drift while enhancing statistical reliability. Then, the Dynamic Selection Mechanism (DSM) generates motion-aware spatiotemporal masks to localize perturbations, guiding them away from discrimination-sensitive regions and promoting temporal coherence. By jointly controlling how and where mixing is applied, ReMA improves representation robustness without additional supervision or trainable parameters. Extensive experiments on diverse video behavior benchmarks demonstrate that ReMA consistently enhances generalization and robustness across different spatiotemporal granularities.

Mingjia Yin, Hao Wang, Wei Guo et al.

Cross-domain sequential recommendation (CDSR) aims to uncover and transfer users' sequential preferences across multiple recommendation domains. While significant endeavors have been made, they primarily concentrated on developing advanced transfer modules and aligning user representations using self-supervised learning techniques. However, the problem of aligning item representations has received limited attention, and misaligned item representations can potentially lead to sub-optimal sequential modeling and user representation alignment. To this end, we propose a model-agnostic framework called \textbf{C}ross-domain item representation \textbf{A}lignment for \textbf{C}ross-\textbf{D}omain \textbf{S}equential \textbf{R}ecommendation (\textbf{CA-CDSR}), which achieves sequence-aware generation and adaptively partial alignment for item representations. Specifically, we first develop a sequence-aware feature augmentation strategy, which captures both collaborative and sequential item correlations, thus facilitating holistic item representation generation. Next, we conduct an empirical study to investigate the partial representation alignment problem from a spectrum perspective. It motivates us to devise an adaptive spectrum filter, achieving partial alignment adaptively. Furthermore, the aligned item representations can be fed into different sequential encoders to obtain user representations. The entire framework is optimized in a multi-task learning paradigm with an annealing strategy. Extensive experiments have demonstrated that CA-CDSR can surpass state-of-the-art baselines by a significant margin and can effectively align items in representation spaces to enhance performance.

Sirui Zhao, Zhengye Zhang, Shifeng Liu et al.

Micro-expressions (MEs), brief and low-intensity facial movements revealing concealed emotions, are crucial for affective computing. Despite notable progress in ME recognition, existing methods are largely confined to discrete emotion classification, lacking the capacity for comprehensive ME Understanding (MEU), particularly in interpreting subtle facial dynamics and underlying emotional cues. While Multimodal Large Language Models (MLLMs) offer potential for MEU with their advanced reasoning abilities, they still struggle to perceive such subtle facial affective behaviors. To bridge this gap, we propose a ME Large Language Model (MELLM) that integrates optical flow-based sensitivity to subtle facial motions with the powerful inference ability of LLMs. Specifically, an iterative, warping-based optical-flow estimator, named MEFlowNet, is introduced to precisely capture facial micro-movements. For its training and evaluation, we construct MEFlowDataset, a large-scale optical-flow dataset with 54,611 onset-apex image pairs spanning diverse identities and subtle facial motions. Subsequently, we design a Flow-Guided Micro-Expression Understanding paradigm. Under this framework, the optical flow signals extracted by MEFlowNet are leveraged to build MEU-Instruct, an instruction-tuning dataset for MEU. MELLM is then fine-tuned on MEU-Instruct, enabling it to translate subtle motion patterns into human-readable descriptions and generate corresponding emotional inferences. Experiments demonstrate that MEFlowNet significantly outperforms existing optical flow methods in facial and ME-flow estimation, while MELLM achieves state-of-the-art accuracy and generalization across multiple ME benchmarks. To the best of our knowledge, this work presents two key contributions: MEFlowNet as the first dedicated ME flow estimator, and MELLM as the first LLM tailored for MEU.

Shukang Yin, Sirui Zhao, Hanchao Wang et al.

Token pruning has emerged as a mainstream approach for developing efficient Video Large Language Models (Video LLMs). This work revisits and advances the two predominant token-pruning paradigms: attention-based selection and similarity-based clustering. Our study reveals two critical limitations in existing methods: (1) conventional top-k selection strategies fail to fully account for the attention distribution, which is often spatially multi-modal and long-tailed in magnitude; and (2) direct similarity-based clustering frequently generates fragmented clusters, resulting in distorted representations after pooling. To address these bottlenecks, we propose Tango, a novel framework designed to optimize the utilization of visual signals. Tango integrates a diversity-driven strategy to enhance attention-based token selection, and introduces Spatio-temporal Rotary Position Embedding (ST-RoPE) to preserve geometric structure via locality priors. Comprehensive experiments across various Video LLMs and video understanding benchmarks demonstrate the effectiveness and generalizability of our approach. Notably, when retaining only 10% of the video tokens, Tango preserves 98.9% of the original performance on LLaVA-OV while delivering a 1.88x inference speedup.