How to train on my own dataset?

[weijiawang96]

Hi,

I found that the code automatically downloads the PCPNet dataset when I run the training script. I used your recommended command:
python normals_pcpnetdata_train.py --model_name='network_k48new_epoch{}.pt' --k_train=48 --iterations=8

But then it seems that the downloading command hangs, since the pclouds.zip remains to be 0 bytes and the code never proceeds. It seems that the downloading step is a default setting when we build the PCPNetDataset.

Now I have my own training dataset, and I want to skip downloading PCPNet dataset and use mine. How may I train using my own dataset?

Thanks

[Joeyyyyy98]

hello，I want to ask that when you run the command “python normals_pcpnetdata_train.py”，did you met the problem :ImportError: /root/anaconda3/envs/a/lib/python3.7/site-packages/quat_to_mat-0.0.0-py3.7-linux-x86_64.egg/quat_to_mat.cpython-37m-x86_64-linux-gnu.so: undefined symbol: THPVariableClass