Chaoyu Li

Chaoyu Li, Eun Woo Im, Pooyan Fazli

Multimodal large language models (MLLMs) have recently shown significant advancements in video understanding, excelling in content reasoning and instruction-following tasks. However, hallucination, where models generate inaccurate or misleading content, remains underexplored in the video domain. Building on the observation that MLLM visual encoders often fail to distinguish visually different yet semantically similar video pairs, we introduce VidHalluc, the largest benchmark designed to examine hallucinations in MLLMs for video understanding. It consists of 5,002 videos, paired to highlight cases prone to hallucinations. VidHalluc assesses hallucinations across three critical dimensions: (1) action, (2) temporal sequence, and (3) scene transition. Comprehensive testing shows that most MLLMs are vulnerable to hallucinations across these dimensions. Furthermore, we propose DINO-HEAL, a training-free method that reduces hallucinations by incorporating spatial saliency from DINOv2 to reweight visual features during inference. Our results show that DINO-HEAL consistently improves performance on VidHalluc, achieving an average improvement of 3.02% in mitigating hallucinations across all tasks. Both the VidHalluc benchmark and DINO-HEAL code are available at https://people-robots.github.io/vidhalluc.

Chaoyu Li, Tianzhi Li, Fei Tao et al.

Vision-language models (VLMs) have advanced video understanding, but their performance is limited by the number of input frames they can process. Existing frame sampling strategies, such as uniform or fixed-budget selection, often fail to adapt to variations in information density or task complexity, resulting in inefficiency and information loss. To address this, we present FrameOracle, a lightweight and plug-and-play module that predicts both (1) which frames are most relevant to a given query and (2) how many frames are needed. FrameOracle is trained using a four-stage curriculum, with the first three stages relying on weak proxy signals such as cross-modal similarity. In the final stage, it leverages stronger supervision from a new dataset we introduce, FrameOracle-41K, the first large-scale VideoQA collection to provide keyframe annotations specifying the minimal set of frames required to answer each question. Extensive experiments across five VLMs and six benchmarks demonstrate that FrameOracle reduces 16-frame inputs to an average of 10.4 frames without any loss in accuracy. When starting from 64-frame candidates, it reduces the input to an average of 13.9 frames while improving accuracy by 1.4%, achieving state-of-the-art efficiency-accuracy trade-offs for scalable video understanding.

Chaoyu Li, Yogesh Kulkarni, Pooyan Fazli

The computational cost of training multimodal large language models (MLLMs) rapidly increases with the number of tokens involved. Existing efficiency methods primarily target inference and rely on token reduction or merging, offering limited benefit during training. In this paper, we propose ReGATE (Reference$-$Guided Adaptive Token Elision), an adaptive token pruning method for accelerating MLLM training. Specifically, ReGATE adopts a teacher-student framework in which the MLLM being trained serves as the student, and a frozen reference large language model (LLM) acts as the teacher. The teacher computes per-token reference losses, which are combined with an exponential moving average (EMA) of the student's own difficulty scores. This adaptive difficulty-based scoring enables the selective processing of crucial tokens while bypassing less informative ones in the forward pass, significantly reducing computational overhead. Experiments demonstrate that ReGATE, when applied to VideoLLaMA2, matches the peak accuracy of standard training on MVBench up to 2$\times$ faster, using only 35% of the tokens. With additional training, it even surpasses the baseline on several multimodal benchmarks, all while reducing the total token count by over 41%. Code and models will be released soon.