There is a mistake when I run the train_pcrnet.py

[Yellowshuohahaha]

Hi, vinits5. I meet some mistakes when I run the train_pcrnet.py. It notes that 'subprocess.CalledProcessError: Command '['where', 'cl']' returned non-zero exit status 1'. Do you know how can I solve it?

[vinits5]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

[Yellowshuohahaha]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

Thank you very much! I will have a look.

[Yellowshuohahaha]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

[shenqildr]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

[Yellowshuohahaha]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Thanks to your reply! Could you please share the code ? And I will try to train the model to see the performance.

[shenqildr]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Thanks to your reply! Could you please share the code ? And I will try to train the model to see the performance.

You can search for computing chamfer in github, there are many implementations of pytorch.

[wzm2256]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Hi, did you try to use the EMD loss? I guess the EMD loss will produce better results.

[Yellowshuohahaha]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Hi, did you try to use the EMD loss? I guess the EMD loss will produce better results.

It seems didn't provide the EMDloss? I see the paper says the EMDloss can produce better results, but I havn't try it.

[wzm2256]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Hi, did you try to use the EMD loss? I guess the EMD loss will produce better results.

It seems didn't provide the EMDloss? I see the paper says the EMDloss can produce better results, but I havn't try it.

It seems to include a EMD loss: https://github.com/vinits5/pcrnet_pytorch/blob/master/pcrnet/losses/emd.py as a stand alone package that requries installation.

I might be able to test the code in one or two weeks. I will share the result then.

[Yellowshuohahaha]

Hey Yellowshuohahaha, I recommend you to take a look at this link.
The error is due to the version mismatch of Pytorch.
If the given link resolves the issue then, please let me know.

The problem seems to be solved. But there is another error... I use pytorch 1.4, but it didn't work.

I have met the same problem. It seems that the chamfer distances function fails. So I rewrite the chamfer distances module by myself and it works. However, when I retrain the model it fails to reach the same performance as the pretrained model.

Hi, did you try to use the EMD loss? I guess the EMD loss will produce better results.

It seems didn't provide the EMDloss? I see the paper says the EMDloss can produce better results, but I havn't try it.

It seems to include a EMD loss: https://github.com/vinits5/pcrnet_pytorch/blob/master/pcrnet/losses/emd.py as a stand alone package that requries installation.

I might be able to test the code in one or two weeks. I will share the result then.

That would be great! I test the code using chamsfer distance a month ago, but the result are not good. The two point clouds cannot be aligned very well, I will also use emdloss to try, thank you!

[wzm2256]

@Yellowshuohahaha Sorry for the delay.. I run the code several month ago, but I forgot to share my results.

In general, the model does not converge to the same results everytime, and you may want to see https://github.com/zhulf0804/PCReg.PyTorch for the numerical comparison.

I run the script with default settings, and one of the results is as follows:

It is clear to see a gradient explosion in around epoch 100.
It might be beneficial to use smaller learning rate than the default value.