search

jianzhangcs / panini

Panini-Net: GAN Prior based Degradation-Aware Feature Interpolation for Face Restoration, AAAI 2022 (PyTorch Code)
112 stars 13 forks source link

Paper results and Repo are clearly different - results of pretained model are very bad #5

Open ofirkris opened 2 years ago

ofirkris commented 2 years ago

Hello, I've tested Panini VS GPEN and GFPgan and found Panini worst on all cases and not matching the quality shown in paper results.

See comparison attached with Panini on the right with worst results: From left to right: Original,GFPgan,GPEN,Panini

family photos_0249

family photos_0242 family photos_0138

jianzhangcs commented 2 years ago

Thank you for your attention and questions! If you want to repair degraded images of the real world, using the pre-training model we provide may not yield good results. This work does not focus on real world degradation, but studies how to adapt to multiple degradation, which is currently artificially designed, and the parameters are not fine-tuned specifically for real world degradation, so our generalization performance on real world degradation is poor. If you want to reproduce the results of the paper, you can use the test examples we provide or use our degrade function to generate the test images.

BbChip0103 commented 2 years ago

Same case here. I tried this model to my data and honestly disappointed...