The train_partseg file training time is too long

[keal1]

train_partseg.py. The training time is too long — is this normal? Each epoch takes nearly three hours.

[qq456cvb]

Slower than I'd expect, but it lines up with the FPS/kNN bottleneck, which runs in pure PyTorch. To speed it up: increase batch_size, lower num_point, raise num_workers, and ideally swap FPS/kNN for CUDA versions. On a modern GPU a part-seg epoch should be minutes rather than hours, so it's also worth confirming training is really on the GPU (valid gpu index, torch.cuda.is_available() returns True).