Konpat

> Is there any way I can only train the semantic encoder of DIFFAE, keeping the DDIM part fixed? I think you mean training only the **semantic encoder** while keeping...

The semantic encoder is trained end-to-end which means the training signal is propagated from the reconstruction loss function, through the diffusion model, UNET, then arrives at the semactic encoder. This...

z_sem should only encode semantic information leaving the stochasitc part be the job of the X_T.

We didn't try running the code on any of the mentioned hardware. You may begin the investigation on the error report first. Which library is the culprit? I believe most...

There is `torch.no_grad` context in the code path for generation. You may want to disable these at `diffusion/base.py`. For example: https://github.com/phizaz/diffae/blob/865f1926ce0d994e4a8dc2b5b250d57f519cadc1/diffusion/base.py#L794

You cannot do the inversion on your own dataset UNTIL you train a DiffAE on that dataset first (please take a look in README on how to train on your...

I myself am not sure about all the places that needed to change. Sorry about that. However, you don't need to change 1) because it is just a default value...

> So it seems like during training the denoising process is not deterministic but then after training you use the deterministic functions mentioned before. Your observation is correct. This is...

`torch.rand` is a **uniform** random which is not what the diffusion model trained for. Please use `torch.randn`.

I'm not sure what's the usecase here. Can you tell me what's the big picture? This doesn't seem like the usecase mentioned in the paper.