PathSelect framework enables efficient WSI processing for vision-language models

Researchers have developed PathSelect, a novel framework for efficiently processing gigapixel Whole-Slide Images (WSIs) with vision-language models. This method reformulates token pruning as a sequential selection process, allowing models to autonomously learn optimal routing strategies instead of relying on static heuristics. PathSelect integrates as an active plugin into the SlideChat base model, freezing both the slide encoder and large language model while introducing a differentiable Soft Top-K operator and a diagonal-attention Denoiser to manage gradients during training. At inference, PathSelect is detached, using a deterministic Hard Top-K operator for adaptive trajectory termination, achieving significant token reduction and low latency. AI

IMPACT This approach significantly reduces computational bottlenecks for processing large medical images, potentially accelerating diagnostic AI development.

RANK_REASON Research paper detailing a novel computational framework for image processing. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Hard Top-K
• Hugging Face
• PathSelect
• SlideBench
• SlideChat
• Soft Top-K
• The Cancer Genome Atlas
• vision-language models

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →