What is the minimum number of images required during inference time?

[mln-wave]

What is the minimum number of images required during inference time?

[lalalune]

You can infer from 1, however these images are hardcoded from the eval files. Currently I believe it iterates and renders all files in the eval.txt of the dataset you use. I am working on a PR to explicitly infer from a single image but its not ready yet, however you can see the start of that here where I'm generalizing data loading: #23

[SteveJunGao]

During inference we don't need any images to run inference as we're doing unconditional generation.

The only part that needs the images is to compute FID metric, for that part I took 20% of the whole dataset

[mln-wave]

oh ok. thankyou!