Open wangzhen699 opened 1 year ago
We will be updating certain non-parameter metrics such as FCNR. However, as all other metrics require a heavily pre-trained model and task-specific metrics, it would be difficult to merge them into a single repository. Nevertheless, we recommend using the following repositories to calculate the Dice/Acc on the test set: IterNet for VSD, MNet for OCSD, MCFNet for FIQA, and Lesion CL for DRA. All these repositories have provided pre-trained models and evaluation codes for your convenience.
Of course, you can train your own evaluation model. We found that a model trained on high resolution will give better performance for quality assessment. Experimentally, all evaluation models we used were trained on approximately 512x512.
Thanks for your work,could you share the code about evaluation metrics,such as FIQA ,OCSD,VSD, FCNR and DRA.