unknownue / pdflow Goto Github PK

hello:) I'm trying to run your model on my data using your model weights but nothing changes. On the right is the original part of the plane on the left is the result of your model after 1 iteration. I also tried to run up to 2000 iterations, there were no changes. Could you tell me how you can improve the performance of your model for a large number of points (200 thousand +-), does it make sense to train on your own data or will it also not change anything?

Dear authors,

Can you please provide suggestions on how to replicate other methods' results in Table 1? For example, PointFilter was originally trained on another dataset, did you retrain it on PUNet dataset with random noise added on-the-fly? If so, can you please suggest the training settings, including the batch size, query ball radius, number of points per patch you used, and how did you obtain the normals? Thank you

您好，请问那个定性结果里面的渐变颜色是怎么生成的（The color of each point indicates its reconstruction error measured by P2M distance.）

A tiny problem. It seems that the uniformity score should be divided by the length of test samples.

Hello everyone, I can't run the training, writes about the absence of some files. I checked all the libraries, they are on my computer + I arranged the training data as follows. What could be the matter ? Why is my training not starting ?

我看您论文里面写的是药用optimizer和scheduler，但代码里面您注释掉了，是不需要用吗？
还有就是论文里面写道使用ReduceLROnPlateau scheduler但是使用emd损失作为指导，我想问下，如果我不使用这个损失的话，这里的scheduler应该如何设置，就是说我如果不使用去噪，只是单纯提取点云的分布

Dear authors,

Which library did you use to draw this figure, i.e., lighter blue means cleaner & darker blue means noisier:

Thank you so much for your help

Hi, I need to know what gpu you have used in implement this paper, and the training time. Thanks very much.