stevenygd / PointFlow

PointFlow : 3D Point Cloud Generation with Continuous Normalizing Flows
https://www.guandaoyang.com/PointFlow/
MIT License
720 stars 101 forks source link

The time spent on each epoch is increasing #41

Closed Fly-Pluche closed 1 month ago

Fly-Pluche commented 8 months ago

I added the flow block in the PointFlow to my model. Then, the bug happened. I tried torch.cuda.empty_cache() and del recon_loss , but not work. 71aab4587a4534d74ded156788f4a94

YTYang627 commented 6 months ago

I added the flow block in the PointFlow to my model. Then, the bug happened. I tried torch.cuda.empty_cache() and del recon_loss , but not work. 71aab4587a4534d74ded156788f4a94

Have you solved this problem? I met the same problem.

stevenygd commented 1 month ago

Have you visualize the number of solver steps? It could be the server steps increasing as the training go, which can happen as the points flow in more and more curve trajectories.