7 papers · 1 filter
Visual Attention Faithfulness in Vision-Language Models is Heterogeneous
Xurui Song, Weishi Wang, Zhongqi Yue +5
Whether attention weights faithfully reflect model reasoning has been actively debated in NLP, yet this question remains largely unexplored for the visual modality in Vision-Langua…
When are radiology reports useful for training medical image classifiers?
Herman Bergström, Zhongqi Yue, Fredrik D. Johansson
Medical images used to train machine learning models are often accompanied by radiology reports containing rich expert annotations. However, relying on these reports as inputs for…
Selftok: Discrete Visual Tokens of Autoregression, by Diffusion, and for Reasoning
Bohan Wang, Zhongqi Yue, Fengda Zhang +15
We completely discard the conventional spatial prior in image representation and introduce a novel discrete visual tokenizer: Self-consistency Tokenizer (Selftok). At its design co…
Reasoning Physical Video Generation with Diffusion Timestep Tokens via Reinforcement Learning
Wang Lin, Liyu Jia, Wentao Hu +6
Despite recent progress in video generation, producing videos that adhere to physical laws remains a significant challenge. Traditional diffusion-based methods struggle to extrapol…
Generative Multimodal Pretraining with Discrete Diffusion Timestep Tokens
Kaihang Pan, Wang Lin, Zhongqi Yue +6
Recent endeavors in Multimodal Large Language Models (MLLMs) aim to unify visual comprehension and generation by combining LLM and diffusion models, the state-of-the-art in each ta…
Benchmarking Multimodal CoT Reward Model Stepwise by Visual Program
Minghe Gao, Xuqi Liu, Zhongqi Yue +7
Recent advancements in reward signal usage for Large Language Models (LLMs) are remarkable. However, significant challenges exist when transitioning reward signal to the multimodal…
Mastering Collaborative Multi-modal Data Selection: A Focus on Informativeness, Uniqueness, and Representativeness
Qifan Yu, Zhebei Shen, Zhongqi Yue +7
Instruction tuning fine-tunes pre-trained Multi-modal Large Language Models (MLLMs) to handle real-world tasks. However, the rapid expansion of visual instruction datasets introduc…