Hello!
I am trying to artistically experiment with your great and clear work.
I am looking for smaller articulated grains. From my experiments, I indeed find the synthesis too close from just a usual granular synthesis.
I can train new models without any issue.
In Finetune_Dance_Diffusion.ipynb, changing the sample size to something else than 65536 gives an error from Torch:
RuntimeError: Argument #4: Padding size should be less than the corresponding input dimension, but got: padding (3, 3) at dimension 2 of input [2, 512, 2]
==> How would you change sample-size in Finetune_Dance_Diffusion which runs train_uncond.py?
I have the feeling having smaller grains for training is hapenning somewhere else though. Where can I control that?
Additional question: I need to provide an already existing model (args.ckpt_path in train_uncond.py) even if it is only used for the start. Is there a way to avoid this? Also, I guess the sample size of this model matters if I want to change mine.
Thank you very much!!
Hello!
I am trying to artistically experiment with your great and clear work.
I am looking for smaller articulated grains. From my experiments, I indeed find the synthesis too close from just a usual granular synthesis.
I can train new models without any issue.
In Finetune_Dance_Diffusion.ipynb, changing the sample size to something else than 65536 gives an error from Torch:
RuntimeError: Argument #4: Padding size should be less than the corresponding input dimension, but got: padding (3, 3) at dimension 2 of input [2, 512, 2]==> How would you change sample-size in Finetune_Dance_Diffusion which runs train_uncond.py?
I have the feeling having smaller grains for training is hapenning somewhere else though. Where can I control that?
Additional question: I need to provide an already existing model (args.ckpt_path in train_uncond.py) even if it is only used for the start. Is there a way to avoid this? Also, I guess the sample size of this model matters if I want to change mine.
Thank you very much!!