Li Guo

Pretrained checkpoints for REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation, accepted to ECCV 2026. REVA-PO is a reinforcement learning framework for chest X-ray report generation. It combines: - Response-Weighted Regularization, which adjusts response-level KL regularization using the response advantage and reference-policy entropy. - Validation-Anchored Policy Reset, which periodically synchronizes the current and reference policies with the best validation checkpoint. - A three-stage training pipeline consisting of warm-up training, classifier-guided supervised fine-tuning, and reinforcement learning. This repository contains the PyTorch and DeepSpeed…

Dataset files and annotations used in REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation, accepted to ECCV 2026. REVA-PO is a reinforcement learning framework for chest X-ray report generation. It uses Response-Weighted Regularization and Validation-Anchored Policy Reset to improve training stability and exploration. This repository contains the IU-Xray data package and the MIMIC-CXR annotation files used by the released code. The model checkpoints are available in the REVA-PO weights repository. Install the Hugging Face Hub client: Extract images.zip inside iuxraydataset/: Set the following fields in the REVA-PO configuration files: MIMIC-CXR images are not…