diegovalsesia / xmfnet Goto Github PK

This problem happened when I compiled all the third-party modules.

Hi, thanks for your work.

I wonder what is the test_list.txt in eval.py file this line. Is it the same with the test_list2.txt?

Besides, I want to know the what the data split you used to obtain the evaluation results in Table 1 and Table 2 in your paper?

Thanks for providing your good work!!! Could you please provide the pre-trained model of state-of-the-art methods based on ViPC?

Thanks for your work! I have .pcd file and images. But I don't have ground truth. Can I use your model to do point cloud completion on my own data? Can you give me some advice>

I think your experiments may be wrong. You used L1 chamfer distance loss to train the model, but L1 is not derivable. And you use L2 CD loss for evaluation. We usually use L1 for evaluation. In other words, in the code, you have reversed L1 and L2. In fact, we can't achieve such accuracy in your paper by using the trained model provided by you. Moreover, if you input 2048 points, your model only generate 1024 more points (8 branches, generate 128 for each branch). We don't usually generate so few papers. Your paper setting is different from other papers, so it is not comparable with them.

@Emanuele97x Hi, I found that you load the camera_calibration.txt in ViPCDataLoader_ft class when training the model in weakly supervised manner. However, I don't have this file at all.

Could you offer this file? Thanks in advance.

When I run the command cat ShapeNetViPC-Dataset.tar.gz* | tar zx , I only get the corresponding images (ShapeNetViPC-View). So, what can I do to get the corresponding partial and complete point clouds?

Hi, @Emanuele97x. When I train the weakly supervised model, the loss_cd = L2_ChamferLoss_weighted() can not be found. Is there something I'm missing?

I keep failing in setting up the function ‘randpartial’. Could you please upload a version of 'ShapeNetViPC-PartPart' data? Thanks a lot.

Where could I find 'train_list2.txt' and 'test_list2.txt' asked in ViPCDataLoader?
I've tried to use 'train_list.txt' and 'test_list.txt' provided in ViPC dataset, but it seems that their formats are different.

I downloaded your pretrained models, but the results seem to vary widely when testing the category of airplane

As for the impact of auxiliary image input on completion performance in the paper, I reproduced it with the pre-trained model you provided. Different auxiliary images seem to have very little impact on performance. Why is that?

Hello, when we use shapenet-vipc data sets of some categories, there will be data loss. How do you solve it?

Thanks for your great work.
I tested the code with cuda11.1, pytorch 1.10.1, then I got "segmentation fault". Can the code work with cuda11.1?

Cloud you please provide this folder for third party? Thank you.