syfbme

The result can be a little worse

I want to know if it works in modern society. I doubt if 4 lstm is enough to predict... Looking forward to your apply. Thanks a lot

D is much stronger than G when training VOC2012 causing gradient vanishing

1

I have to set D's learning rate be 1/10 of G. I wonder if anyone meet my issue. In my training, the D convergences early after 2 or 3 epoches...

According to the paper of pix2pix, there are some goods for skip connection.

question about high resolution model

1

Thanks for releasing the hr model. I have a question about the training details: I notice the hr model training script still get 256*256 images. The diff is using a...

I think it will lose some info. Could please tell me why you set params like that. Thanks.

The online degradation is not the same as described in the paper

1

Could you please explain. Thanks. Besides, which one do you prefer

Could you please share the details of training pix2pixhd. Like how many epochs for global and local generator. How to deal with down sample when the picture size in VOC2012...