Reproduce “GroundSeek-CXR: Retrospective Evaluation-Guided Active Observation Framework for Multi-Label Chest Radiography Diagnosis” 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
Yihang Fu, Xinyu Liu, Qing Xu, Xiaofeng Yang, Zhen Chen
GroundSeek-CXR trains a chest X-ray model to inspect selected regions and decide when to stop. A frozen teacher evaluates observation and stop choices by continuing them to final multilabel diagnoses, and score-gap-weighted preferences train the student. On 949 abnormal VinDr-CXR test images, it reaches 57.72% sample-level F1, versus 53.24% for Immediate DPO; it uses more observations overall.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.