chaofengc / IQA-PyTorch

👁️ 🖼️ 🔥PyTorch Toolbox for Image Quality Assessment, including LPIPS, FID, NIQE, NRQM(Ma), MUSIQ, TOPIQ, NIMA, DBCNN, BRISQUE, PI and more...
https://iqa-pytorch.readthedocs.io/
Other
1.89k stars 166 forks source link

About the configuration file for the TOPIQ dataset training. #160

Open XiaoCaoAskedForHelp opened 4 months ago

XiaoCaoAskedForHelp commented 4 months ago

Only four dataset configuration files are provided in the project, can I provide configuration files for training on other datasets, such as CSIQ, LIVE, SPAQ, and FBLIVE?

chaofengc commented 3 months ago

Thank you for your interest and suggestion. We recommend the train/test protocol here: Performance Evaluation Protocol due to dataset quality and generalization performance.

Training on small datasets like CSIQ and LIVE may not be very useful, and models trained with SPAQ and FBLIVE do not generalize well. Nevertheless, if you have configs for CSIQ, LIVE, SPAQ, or FBLIVE, please submit a pull request with the files and any necessary documentation.