Testing with a single example (point cloud)

I’ve trained PointNet++ for classification on ModelNet40. Unfortunately I am unable to test the trained net on a single input point cloud.

Sample test script:

import torch
import torch_geometric.read as R
import torch_geometric.transforms as T
from model import PointNet2

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# load model weights
model = PointNet2()
model.load_state_dict(torch.load('./weights.pt'))
model.to(device)

# load input and pre-process
off_input = R.read_off('./bench_0174.off')
sampler = T.SamplePoints(1024)
normalise = T.NormalizeScale()
off_preprocessed = normalise(sampler(off_input))

# inference
print(model(off_preprocessed.to(device)))

This script results in AttributeError: 'Data' object has no attribute 'batch'. It seems unlikely to be able to set batch size without going through a dataloader. Any idea on how to proceed?

Full stack trace error in interested:

Traceback (most recent call last):
  File "/home/reddwarf/3d-feature-extraction/pointnet2/test.py", line 18, in <module>
    print(model(off_preprocessed.to(device)))
  File "/home/reddwarf/anaconda3/envs/3df/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/reddwarf/3d-feature-extraction/pointnet2/model.py", line 58, in forward
    sa0_out = (data.x, data.pos, data.batch)
AttributeError: 'Data' object has no attribute 'batch'

[rusty1s]

Just add a batch object to your test data: data.batch = torch.zeros(data.num_nodes, dtype=torch.long)

[rusty1s]

Closing this issue due to inactivity. Please feel free to re-open if there are still concerns.