The inference and post-processing are very slow,

[KingofAsianPopJC]

Hello, nice job of your contribution. I downloaded your model for segmenting all body parts in CT images, I followed your instructions to configure my environment and programs, and it truly can correctly segment body parts in the CT images from my datasets.

However, a serious problem is that the inference and post-processing are too slow, inferring a single case takes nearly 2 hours !!. I think it is not acceptable. I make sure the machine I used for inference and post-processing is good enough (Nvidia A100), so, I want to ask you, why it is so slow, how can I improve the speed? Is there something I need to change in configuration or codes?

Thanks a lot for your reply !!!

[alexanderjaus]

We also experienced this problem and we are working on it, but as of right now, our inference time is determined by the nnUnetV1. Regarding the post-processing, we implemented it with the possibility of using multi-threading. This however only allows for the parallel processing of individual predictions.

[pooya-mohammadi]

@KingofAsianPopJC I know it's too late. I faced the same issue with nnunetv1, so I modified the inference section which was done mainly with numpy and did it with torch as it is done in nnunetv2. The inference speed is much faster.
You need to install https://github.com/pooya-mohammadi/nnUNet/tree/nnunetv1