Reproduce “Collaborative Edge-to-Server Inference for Vision-Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Soochang Song, Yongjune Kim
The authors present a server-edge visual question answering system that sends a coarse image first and requests a detail-preserving crop when the answer appears uncertain. It uses min-entropy as an uncertainty proxy and attention to locate relevant regions. On TextVQA, it matches always-crop accuracy with 50% and 24% lower additional communication cost for two open-source models; uncertainty does not guarantee correctness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.