Questions about continue trainning

[LaLaLailalai]

Hi Bruno,

Do you know how to continue training after it stops in the middle (e.g., 50/200epochs)?

I tried to continue to train the model by myself, but I only found the load weights codes in generate.py.

[LaLaLailalai]

By the way, how to set the parameters of evaluation to reproduce the exact FID result?

For example, if I would like to reproduce the NTU120 xsub FID, I should run the command :
"python generate.py --model model_path --n_classes number_classes --label class_index --gen_qtd how_many_samples "

How to set gen_qtd for each dataset?

[DegardinBruno]

Hey @LaLaLailalai, sorry for my late reply!

Do you know how to continue training after it stops in the middle (e.g., 50/200epochs)?

You can insert the same code with strict=True, where you are ensuring the exact same arch:

# Load Models
generator.load_state_dict(torch.load(opt.model), strict=True)

And then for epoch in range(50, opt.n_epochs): to start at epoch 50. We do not use any scheduler or epoch-dependent parameter, so it must work without a problem!

By the way, how to set the parameters of evaluation to reproduce the exact FID result?

Just change n_classes to 120 and mlp_dim to 8 (same as our paper). We also used gen_qtd with 1000. The results may vary a bit due to randomness, but not much! Remember to set trunc_mode to - in generating for FID and MMD evaluation.

For any further questions, feel free to ask!