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 https://github.com/fxia22/pointnet.pytorch/blob/2ec315f174ed5ffc05d127d5cf14c94ce873c1e0/train_classification.py#L40 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 : https://github.com/fxia22/pointnet.pytorch/blob/2ec315f174ed5ffc05d127d5cf14c94ce873c1e0/pointnet.py#L47 how to solve it? thanks for your help

Hi, I am using PointNet on Pytorch to train my custom dataset. However, after training around 10-20 batches of examples, the loss goes into NAN.

I highly doubt that it is a exploding gradient issue. (still trying to diagnose the problem)

Posting this problem here in expectation of any help or direction that you may provide.

Things I have tried is -

1. Reducing the Learning Rate to 0.0001
2. setting drop_last=True in DataLoader
3. reducing the batch size from 32 to 10

Hi, I think the nn.Conv1d and nn.Linear are the same for pointwise convolution( if kernel_size = 1 for conv1d), so could you please provide some reasons about why you choose nn.Conv1d for the first three layers and choose nn.Linear for the last two layers ?

No offense, but I think you don't implement PointNet correctly. For example, the Feature-transform-net isn't found here, and this pointnet.pytorch has much fewer layers than the raw author's design.

Hello,

I have a doubt of the network STN3D, in the paper the authors says that after the max pooling there are only two layers and the output matrix is initialized as identity matrix. But I am not able to see this implementation on the code. In the file https://github.com/fxia22/pointnet.pytorch/blob/master/pointnet.py in the class stn3d what I can see is a third full connected layer that is not described on the paper but I am guessing that this stage is used in order to generate the output matrix of 3x3 (that I think this is a correct) but my issue is related to this lines:

 x = self.fc3(x)

        iden = Variable(torch.from_numpy(np.array([1,0,0,0,1,0,0,0,1]).astype(np.float32))).view(1,9).repeat(batchsize,1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        return x

Here, I am not able too see the init of this layer as a identity matrix. That I see is that in every iteration you are adding to the matrix calculated by the network an identity matrix but this is not the same that is described on the paper, right?

Another doubt that I have is related to the convolutional layers, why are you using 1d convolutional layers? In the tensorflow implementation the convolutional layers are 2d. Is it for any specific reason?

Thanks in advance

Hello,

I am interested in furthering PointNet and using it for regression. Is this possible? Has anyone attempted this yet? Any thoughts or comments would be great!

Thanks!

Hi Fei Xia,

I was wondering whether this line (14-15) of code in the PointNet Architecture is missing an extra layer as of the official paper release:

class STN3d(nn.Module):
    def __init__(self):
        super(STN3d, self).__init__()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        *---HERE-> = torch.nn.Conv1d(64,64,1)*
        self.conv2 = torch.nn.Conv1d(64, 128, 1)

Not a very big deal but just want to clarify on this.

Cheers, Tim

Hi @fxia22,

I would like to know, if train_segmentation.py generates the segemented point cloud, since it would be good to view a segmented point cloud.

In addition to that, it would also be helpful to view the critical point and the upper bound shapes as mentioned in the pointnet paper. Do let me know your views.

Hi,

We run the classification model on ModelNet40 dataset (w/o feature transform), but couldn't get > 0.8 accuracy even after 200 epochs (we could reproduce the shapenet result, though). So I'm asking for some suggestions.

(1) Did you train ModelNet40 with the same learning rate schedule as in train_classification.py?

(2) We prepared the data loader by randomly subsetting ModelNet40 vertices. As mentioned in https://github.com/fxia22/pointnet.pytorch/issues/35#issuecomment-471158774, you used a data processing pipeline to sample points from the mesh. Could you provide a high-level overview of how it's done or point to a reference that describes ModelNet40 data processing?

Thanks you very much!

Hi, your loss implemention is like this ||(A*(A_T-I))||, but the loss should be like ||(I-A*A_T)||, therefore, it should be like this
loss = torch.mean(torch.norm(torch.bmm(trans, trans.transpose(2, 1) ) - I, dim=(1, 2))), am I right?

Thanks for your great work! I have some doubts about the usage of log_softmax function. As written in the 104th line of the file pointnet.py return F.log_softmax(x, dim=0), trans Actually, Dim0 means the dimension of batch. However, the log_softmax should be applied along the channels. So I think the parameter dim should be set as -1. Am I right or I miss something important? Looking forward for your response!

Hi, @fxia22 When i try training this repo, i found here only can use .ply format, so i add a python file "convert.py" in ModelNet40 folder (use after unzip ModelNet40 dataset.).

This file had already been tested on my machine (i7-11700 8C16T), and i can successful run cls mission after convert ModelNet40 dataset.

In detail, this file also check am i successful convert or not, this function approch by check() in 'convert.py', but it cost so much time (about 900 secs on my machine, the other part about 23 secs.), if you don't wont it happen, you can disable that.

Also, i add a run 'convert.py' command in your README.md, and add 4 .txt you already gave in this page https://github.com/fxia22/pointnet.pytorch/issues/52

If you have any need for convert, pls let me know.

Thanks for releasing the package.

Hi, is it possible to have a checkpoint for the results that are shown in the Read.me? Or, if already exists, can anyone point me to the right direction?

Dear Sir:

I'm using Python3 and I'm trying to train the two .py files in /utils, I used:

python3 train_classification.py --dataset='' --nepoch='' --dataset_type='<modelnet40 | shapenet>'

instead but the system told me that train_classification.py: error: argument --nepoch: invalid int value: ''

How can I fix this? Thank you very much!

Best, cellist

File "show_seg.py", line 30, in data_augmentation=False) File "/home/gene/YuRen/pointnet.pytorch/pointnet/dataset.py", line 107, in init self.num_seg_classes = self.seg_classes[list(self.cat.keys())[0]] IndexError: list index out of range

As above, I am facing the problem I have no idea to deal with, please help!! PS:The training progress is fine, so I think there might be some part of code I need to change when I am running on my server.

I ran this command python utils/show_seg.py and reported the following error:

Namespace(class_choice='Chair', dataset='./shapenetcore_partanno_segmentation_benchmark_v0', idx=0, model='./seg/seg_model_Chair_29.pth')
{'Chair': 0}
{'Airplane': 4, 'Bag': 2, 'Cap': 2, 'Car': 4, 'Chair': 4, 'Earphone': 3, 'Guitar': 3, 'Knife': 2, 'Lamp': 4, 'Laptop': 2, 'Motorbike': 6, 'Mug': 2, 'Pistol': 3, 'Rocket': 3, 'Skateboard': 3, 'Table': 3} 4
model 0/704
torch.Size([2500, 3]) torch.Size([2500])
Traceback (most recent call last):
  File "utils/show_seg.py", line 45, in <module>
    classifier = PointNetDenseCls(k= state_dict['conv4.weight'].size()[0])
KeyError: 'conv4.weight'

What should I do? Looking forward to your reply.