search

Justin-Tan / generative-compression

TensorFlow Implementation of Generative Adversarial Networks for Extreme Learned Image Compression
MIT License
511 stars 108 forks source link

about generator loss graph #12

Closed chenxianghu closed 6 years ago

chenxianghu commented 6 years ago

i use the same code as you for training cityscapes dataset 60 epochs, the generator loss at last 23.163, but your model will decline to about 0.5 i use the defaut configuration in conf.py, only modify the num_epochs, i don't know why?

Epoch 59 | Generator Loss: 23.163 | Discriminator Loss: 0.410 | Rate: 0 examples/s (99773.67 s)

Justin-Tan commented 6 years ago

High generator loss is not necessarily a bad thing because there are a lot of additional terms in the generator loss function I have added since the initial commit. Is the generated image quality poor?

chenxianghu commented 6 years ago

@Justin-Tan The generated images quality almost the same as yours on cityscapes dataset, so i think i don't need to care the Generator Loss according to your reply. Thank you very much!