TypeError: __init__() got an unexpected keyword argument 'num_points'

[3bobo]

when I run train_classification.py, there is an error:

Traceback (most recent call last):
  File "train_classification.py", line 58, in <module>
    classifier = PointNetCls(k = num_classes, num_points = opt.num_points)
TypeError: __init__() got an unexpected keyword argument 'num_points'

I noticed that opt.num_points has been used in

pointnet.pytorch/train_classification.py

Line 40 in 2ec315f

dataset = PartDataset(root = 'shapenetcore_partanno_segmentation_benchmark_v0', classification = True, npoints = opt.num_points)

so I comment it and run again, but another error:
ValueError: Expected more than 1 value per channel when training, got input size [1, 512]
and this comes from :

pointnet.pytorch/pointnet.py

Line 47 in 2ec315f

x = F.relu(self.bn4(self.fc1(x)))

how to solve it? thanks for your help

[mayurbhangale]

Seems like num_points is never used in PointNetCls. I solved it by simply not passing it to line 58

pointnet.pytorch/train_classification.py

Line 58 in 2ec315f

classifier = PointNetCls(k = num_classes, num_points = opt.num_points)

to

classifier = PointNetCls(k = num_classes)

[3bobo]

@mayurbhangale thanks for your help, I've changed this line, but I still get the error:
ValueError: Expected more than 1 value per channel when training, got input size [1, 512], which is caused by bathnorm1d from pytorch, so I change this line:

pointnet.pytorch/pointnet.py

Line 47 in 2ec315f

x = F.relu(self.bn4(self.fc1(x)))

to:
x = F.relu(self.fc1(x))
but this time all output data are zero, like this, have you met this?

[mayurbhangale]

Not really, right after cloning, the only change I made is the one mentioned above. Using torch==0.4.1, in case that helps you.
I've also successfully trained classification and segmentation. Sharing weights: https://drive.google.com/open?id=16fXGyOTOxvT-ozYwu3OPL3Qk4vaS6liL

[3bobo]

@mayurbhangale ,thanks for sharing your weights, I will try to test and see the result , but I'm still wondering what's wrong with mine, it will be convenient for further our using.

[DC1991]

Hi @mayurbhangale , how about your test performance on your trained classification model. I tested the trained model (25 epochs, on classification), the accuracy is always 0.0000000 and the loss is about 1.5-2.0.

[alfredtorres]

Hi @mayurbhangale , how about your test performance on your trained classification model. I tested the trained model (25 epochs, on classification), the accuracy is always 0.0000000 and the loss is about 1.5-2.0.

you just run the show_cls.py ? maybe the last line in this .py has something run, the loss around 1.5-2 may be ok.
I try 60 epochs ,get nearly 95% acc.