sayakpaul / cmmd-pytorch

PyTorch implementation of CLIP Maximum Mean Discrepancy (CMMD) for evaluating image generation models.
Apache License 2.0
97 stars 5 forks source link

The CMMD value is different from the example #3

Closed feilvvl closed 7 months ago

feilvvl commented 7 months ago

Hi, thank you very much for being able to implement CMMD using Pytorch, it definitely helped me a lot! But when I run the example, the result is different. (7.717 vs 7.696image I wonder if there is a mistake in my environment. My cuda version is release 11.1, torch version is 1.10.2+cu111, can you give me more results about the configuration? For example, torch version, cuda version, etc. If you provide it in detail, I would appreciate it.

sayakpaul commented 7 months ago

I am on PyTorch 2.0 and CUDA 12.0. I ran my tests on A100.

feilvvl commented 7 months ago

I am on PyTorch 2.0 and CUDA 12.0. I ran my tests on A100. That is great! I have tried your configuration and got the right answer 7.696! Thank you a lot!