The paper introduces VT-Contrast, a representation-level temporal counterfactual objective designed to improve temporal understanding in Video Language Models (VideoLMs). By supervising late-layer last-frame video tokens and contrasting order-preserving views with reordered counterfactuals graded by Kendall tau distance, VT-Contrast addresses the mismatch between ordered video input and text-based supervision. The method requires no architectural changes, is compatible with various VideoLM training tasks, and demonstrates improved performance on temporal understanding benchmarks.
By Yumeng Shi, Quanyu Long, Yin Wu, Wenya Wang
arXiv:2608.21030v1 Announce Type: cross
Abstract: Video multimodal large language models have advanced significantly, yet fine-grained motion-temporal understanding remains fragile. The core bottlene...
By Chenghua Zhu, Zhaolu Kang, Qifan Shi, Siyan Wu, Kehan Jiang, Lei Wei, Lianyu Hu, Guangyuan Dong, Mingbo Yang, Rui Lu, Guibo Luo
arXiv:2608. 15869v1 Announce Type: cross Abstract: Multimodal large language models increasingly use visual chain-of-thought (Visual CoT) to reason about spatial, temporal, and embodied environments.
By Xiaoyu Zhu, Xinke Deng, Suresh Taddewadikar, Arnab Kumar Mondal, Zhongyu Jiang, Ian Fasel, Joerg Liebelt
arXiv:2606. 29431v1 Announce Type: new Abstract: Despite the impressive capabilities of Large Vision-Language Models (LVLMs), they remain susceptible to hallucination, generating content inconsistent with the input image.
By Yichen Guo, Kai Tang, Fenglai Lin, Yiding Sun, Dongshuo Zhang, Wenya Wang, Lin William Cong, Shanghang Zhang
TimeBlind is a diagnostic benchmark designed to evaluate fine‑grained spatio‑temporal compositionality in video large language models (LLMs). It categorizes temporal understanding into three levels—atomic event recognition, event property characterization, and reasoning about event interdependencies—and uses a minimal‑pairs paradigm where video pairs share identical static content but differ only in temporal structure. Across 20 state‑of‑the‑art MLLMs tested on 600 curated instances, the best model achieved only 48.2% instance accuracy, far below human performance of 98.2%, highlighting a reliance on static visual shortcuts rather than true temporal reasoning.
By Baiqi Li, Kangyi Zhao, Ce Zhang, Chancharik Mitra, Jean de Dieu Nyandwi, Gedas Bertasius
arXiv:2609.39563v1 Announce Type: new
Abstract: Existing video language models encode sampled RGB frames independently, so a long video must either exhaust the token budget or drop the changes betwee...
By Can Zhang, Xiaotian Han, Junyuan Shang, Yuchen Ding, Zhenyu Zhang, Shuohuan Wang, Dianhai Yu, Ruirui Li
LongVU‑TTT is a causal test‑time training method for long‑video multimodal large language models that inserts a convolutional resampler with fast‑weight updates between the vision encoder and the LLM. The fast weights adapt per video and contextualize frame features before compression, while a hybrid selector keeps explicit visual evidence for downstream reasoning. Experiments show that TTT‑Conv outperforms TTT‑MLP and bidirectional Mamba2 on MLVU, and beats attention‑ and fixed‑state recurrent resamplers on three benchmarks, achieving competitive results on five video‑understanding tasks after reducing 512 frames to 128 LLM frames.
By Mahmoud Ahmed, Sameh Abdulah, Olatunji Ruwase, Sam Ade Jacobs, Mathis Bode, Mohamed Elhoseiny
arXiv:2606. 02522v1 Announce Type: cross Abstract: Video multimodal large language models (MLLMs) have made rapid progress on general and long-form video understanding, yet their ability to preserve brief answer-critical visual evidence remains underexplored.
By Xiaolin Liu, Yilun Zhu, Xiangyu Zhao, Xuehui Wang, Yan Li, Xin Li, Haoyu Cao, Xing Sun, Shaofeng Zhang, Xu Yang, Zhihang Zhong, Xue Yang
arXiv:2603.17825v2 Announce Type: replace
Abstract: In this work, we study the role of Massive Activations (MAs), which are rare, high-magnitude spikes confined to a few fixed hidden dimensions in vi...
By Xianhang Cheng, Yujian Zheng, Zhenyu Xie, Tingting Liao, Hao Li
The paper introduces Recency Forcing, a technique that addresses the long‑horizon degradation in autoregressive video generation caused by KV eviction mismatch. By applying a timestep‑dependent bias—Temporal Response Bias—derived from a positional response measure, the method reduces the influence of distant frames during inference without altering context length or training objectives. An exact reformulation, Biased Attention Reparameterization, enables this bias to be applied as a standard FlashAttention call with zero overhead, achieving state‑of‑the‑art long‑horizon generation quality on VBench datasets.
By Tri Cao, Hung Nguyen, Phong Nguyen, Khoi Nguyen
arXiv:2605. 05895v2 Announce Type: replace-cross Abstract: Modern AI-generated videos are photorealistic at the single-frame level, leaving inter-frame dynamics as the main remaining axis for detection.
By Minsuk Jang, Yujin Yang, Hee-Seon Kim, Minseok Son, Younghun Kim, Changick Kim
arXiv:2603. 16870v3 Announce Type: replace-cross Abstract: Recent advances in video generation have revealed an unexpected phenomenon: diffusion-based video models exhibit non-trivial reasoning capabilities.
By Ruisi Wang, Zhongang Cai, Fanyi Pu, Junxiang Xu, Wanqi Yin, Maijunxian Wang, Ran Ji, Chenyang Gu, Bo Li, Ziqi Huang, Hokin Deng, Dahua Lin, Ziwei Liu, Lei Yang