5 papers
Distill the Visual Evidence, Not Just the Answer: Cross-World On-Policy Distillation for Vision-Language Models
Yuanhao Sun, Huawei Ji, Jiaxin Ding +2
A central goal of vision-language model (VLM) distillation is to transfer both the teacher's language capabilities and its visual understanding. However, existing methods primarily…
HAP: Head-Adaptive Visual Token Pruning via Cross-Modal Alignment
Yuanhao Sun, Huawei Ji, Yuan Jin +3
Recent Vision-Language Models encode high-resolution images into long visual token sequences, incurring prohibitive prefill costs. To compress them, existing methods score each vis…
ENCORE: Entropy-Guided Cropping and Attention Regularization for Robust Vision--Language Understanding
Yuanhao Sun, Huawei Ji, Jiaxin Ding +2
Vision-Language Models (VLMs) perform well on diverse vision-language tasks, but transformer-based visual encoders split images into fixed-resolution sub-images, compromising objec…
VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models
Huawei Ji, Yuanhao Sun, Yuan Jin +4
Visual token pruning methods effectively mitigate the quadratic computational growth caused by processing high-resolution images and video frames in vision-language models (VLMs).…
AceParse: A Comprehensive Dataset with Diverse Structured Texts for Academic Literature Parsing
Huawei Ji, Cheng Deng, Bo Xue +6
With the development of data-centric AI, the focus has shifted from model-driven approaches to improving data quality. Academic literature, as one of the crucial types, is predomin…