Hi, I had some issues getting the training to work on Ubuntu 18.04 and a recent PyTorch version which PointNet2 seems to require.
Updated the instructions in the README and requirements
Fixed some typos.
Removed synthetic data in LineMod training. Or do you have a LineMod dataset available with the render folder?
Replaced neupeak.utils.webcv2 with cv2
Replaced AT_CHECK with TORCH_CHECK
Replaced CyclicLR with the torch implementation. I hope that the parameters are chosen correctly. For example the torch implementation defaults cycle_momentum=True during the initialization which I changed to False since the optimizer would require a momentum to be set. Is it a good idea to use momentum?
Hi, I had some issues getting the training to work on Ubuntu 18.04 and a recent PyTorch version which PointNet2 seems to require.
cycle_momentum=True
during the initialization which I changed toFalse
since the optimizer would require a momentum to be set. Is it a good idea to use momentum?