search

pfriedri / wdm-3d

PyTorch implementation for "WDM: 3D Wavelet Diffusion Models for High-Resolution Medical Image Synthesis" (DGM4MICCAI 2024)
https://pfriedri.github.io/wdm-3d-io
MIT License
50 stars 5 forks source link

About how to evaluate by FID #3

Closed DianaNerualNetwork closed 2 months ago

DianaNerualNetwork commented 3 months ago

Dear Author,

Thanks for your impressive work! It is so amazing! I wonder to know how to use that pretrained weight to evaluate our generation? Did that any related code released? Or any detail description for that?

pfriedri commented 3 months ago

@DianaNerualNetwork thank you for your interest in our work.

This part of the code is indeed not yet in the repository, thanks for pointing that out. I'll add it as soon as possible.

DianaNerualNetwork commented 3 months ago

Thanks for your kindness, we have confused about that for a week, we try to evaluate it by 3D to 2D. It is hard to find some related code to evaluate 3D FID

pfriedri commented 3 months ago

@DianaNerualNetwork I'm currently cleaning up our evaluation code and we'll release it soon, but if you just want to know how to compute 3D FID scores, you can have a look at https://github.com/batmanlab/HA-GAN/blob/master/evaluation/fid_score.py (our implementation is very similar)

pfriedri commented 2 months ago

@DianaNerualNetwork I just added the code. I hope this answers your question.