ERDES Benchmark Model Checkpoints

This record contains pre-trained model checkpoints used to benchmark the ERDES dataset for ocular ultrasound retinal detachment classification as described in our paper ERDES: A Benchmark Video Dataset for Retinal Detachment and Macular Status Classification in Ocular Ultrasound paper. Models span 8 architectures across 5 classification tasks plus pooling tasks. The dataset is available here at Zenodo along with the code here in our GitHub repo. We have 5 tasks: Non-RD vs RD, Macula Detached vs Intact, Normal vs RD, Normal vs PVD, PVD vs RD. For Normal vs PVD, we have also trained models with selective pooling strategies which retains 30%, 50%, or 70% of the temporal segments D.Each released model is named as __best.ckpt This record also contains a pre-trained YOLOv8 globe detection model checkpoint for cropping ocular ultrasound videos around the globe region of interest (ROI). Please refer to our GitHub for the cropping pipeline.