Info about test in Pointnet

[manuelloit]

Hello. I was seeing the excellent work on pointnet: congratulations. Sorry for the questions, but it's a new field for me and I'm studying it. I ran the train on a single dataset (../data/modelnet40_ply_hdf5_2048 with train_files.txt with only data / modelnet40_ply_hdf5_2048 / ply_data_train0.h5 inside) then launching the "python train.py" command as instructions. I would like to test the trained network; correct me if I'm wrong, to do this I have to do "python evaluate.py --visu". In ../data/modelnet40_ply_hdf5_2048 there are also .json files; what are they for? I have, in fact, modified the files ply_datatest _ id2file.json keeping, inside, the only ply for "car"; in doing so, I wanted to do the test with "python evaluate.py --visu" for the car point clouds only, but I have, as results, a test on all the objects. How can I test only on a category of objects? Do I need to access h5 test files to indicate the only clouds to test and, if so, kindly, how? Moreover I wanted to see point clouds that, I presume, are in h5 files; howewer I found *.off files downloading them from site containing ModelNet40. If you need clarification, I'm here. Thanks a lot

[moxigua]

you can download the dataset from https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip ,and move it to your data folder