search

cswry / SeeSR

[CVPR2024] SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution
Apache License 2.0
390 stars 25 forks source link

overfitting on small dataset? #14

Open rainsoulsrx opened 8 months ago

rainsoulsrx commented 8 months ago

Hi, I use 500 human face images(only contain 6 id) to train this model, but when I test the trained data, the result is strange, I test using A people, but the result is B people,why?

cswry commented 8 months ago

Helo, this is an interesting phenomenon, as we haven't encountered this issue during training on a large-scale dataset. Can you provide detailed training specifics?

rainsoulsrx commented 8 months ago

Helo, this is an interesting phenomenon, as we haven't encountered this issue during training on a large-scale dataset. Can you provide detailed training specifics?

Yes. I use about 500 pair datases, phone image and corresponding camera image after Optical flow registration. the dataset only contain 6 ids. training config I use what you provide, just replace the training data