Closed QianMuXiao closed 1 week ago
The best way to utilise multiple GPUs for TotalSegmentator is to run it for several subjects in parallel on different GPUs. For TotalSegmentator itself it does not make sense to support MultiGPU since it anyways uses the GPU only for a few seconds.
I often need to use TotalSegmentator to perform preliminary segmentation on large datasets to create rough labels, which is a preprocessing method I frequently use in my current research. However, processing hundreds of patients one by one using TotalSegmentator on a single GPU is very slow and often takes several days. My school provides multi-GPU servers, so I’m wondering if the Python API can support multi-GPU acceleration?