log_sample after the batch training

[amasawa]

Thanks for your code for the project!

I need to find out the log_sample() after the batch training.

Is it used to reconstruct the sample at each end of the training epoch? And then, put the reconstructed samples into checkpoint/name/sample folder?

[Saranga7]

I am also exploring the log_sample() method of the code. Yes, it is used to reconstruct the sample at each end of the training epoch and both the reconstrucred samples are put into checkpoint/name/sample. On tensorboard, these samples are also logged along with the respective batch of real images.

I understand that autoencoding is being performed on a sampled batch of images. I am, however, unsure where those original images are being used to condition the generation. In other words, I was expecting cond = encoder(x_start) to be used as a condition to the DDIM, but I don't see it happening, and yet somehow, the generated images are reconstructions of of a sampled real image batch.

I am not sure if I have explained the problem clearly, but can anyone please help clarify it to me??