Bad results when continue training

[ShaniGam]

First, thank you for this implementation! I want to continue train a model that I trained before. I run the following command: python train.py --mixing --init_size 64 ./images/ --ckpt ./checkpoint/140000.model I don't get any errors when loading but the generated images don't look like the ones I got before, but instead I get really bad results.

[rosinality]

Could you use train-step-X.model checkpoint?

[ShaniGam]

I tried that as well, same results.

[rosinality]

Then can you check init_size is correct for that checkpoints? I think size will be larger than 64 if you use 140,000 iter checkpoints.

[ShaniGam]

I'm sure it's the correct size because I stopped the experiment and ran it again right away. Besides, if it generates good images that are > 64, shouldn't generate good 64 images?

[rosinality]

As generators will be trained for higher resolutions, it will show less quality at lower resolutions.

[chenchenver]

Hello, have you solved the problem of continuing training?

[ShaniGam]

Unfortunately no but I stopped working on that.

[cientgu]

I meet the same problem, i think there might be something wrong when loading the g_optimizer and d_optimizer.