Announcement_5

Two papers accepted at CVPR’26: OraPO, a data-efficient reinforcement finetuning method for radiology report generation that achieves SOTA performance while only using less than 0.1% of training data, and a flow matching-based scene graph generation method.