search

tensorlayer / SRGAN

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
https://github.com/tensorlayer/tensorlayerx
3.24k stars 813 forks source link

Poor results using subway images on validation #228

Open JonZubieta opened 3 years ago

JonZubieta commented 3 years ago

I have previously enlighten subway images using EnlightenGAN network. During that process, the images lost resolution so now I intend to improve them with SRGAN.

I've trained the SRGAN network using DIV2K_train_LR_bicubic X4 and DIV2K/DIV2K_train_HR datasets and it performs well on validation using VALIDLR_BICUBIC_X4 dataset.

However, the resolution improvement on my subway train dataset is really poor. Any idea on how could I improve the results?

Here goes one photo of my subway validation dataset: Fake_0

Thanks

husterwang commented 1 year ago

Sorry, could I ask that how to test our own images? @JonZubieta