search

amzn / convolutional-handwriting-gan

ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation (CVPR20)
https://www.amazon.science/publications/scrabblegan-semi-supervised-varying-length-handwritten-text-generation
MIT License
264 stars 55 forks source link

Couldn't find models gradients values ! #19

Open AhmedAl93 opened 3 years ago

AhmedAl93 commented 3 years ago

Hello, Thanks for your amazing work !

When training on my custom dataset, I wanted to plot the gradients of each model's loss function using tensorboard, but I was confused when I found out that for "_parameters" field is empty for netG, netD & netOCR models, same goes for "_state_dict_hooks" field. Here are some log lines:

`End of epoch 3 / 200 Time Taken: 171 sec learning rate = 0.0002000 Model netOCR _parameters: {} Model netOCR _state_dict_hooks: OrderedDict()

Model netG _parameters: OrderedDict() Model netG G_param: SN Model netG _state_dict_hooks: OrderedDict()

Model netD _parameters: OrderedDict() Model netD D_param: SN Model netD _state_dict_hooks: OrderedDict()`

Can anyone explain this, please?

rlit commented 3 years ago

not sure I see why this would happen, but this implementation writes on default into a 'visdom' server, which can replace tensorboard in most cases - did you try to use that?