phermosilla/totaldenoising

Total Denoising: Unsupervised Learning of 3D Point Cloud Cleaning

totaldenoising's Issues

CUDA out of memory during training

Dear Authors,

I tried to train using your given dataset 0 (command: python Train.py --dataset 0). The available memory is around 10GB. However, the training cannot start as it keeps saying CUDA out of memory.

I wonder if 10GB is enough? I am curious how much memory you used on your GeForce RTX 2080 when you trained your model.
Or how can we use multiple GPUs to train? Alternatively, is there a way to change the batch size?

Thank you for your help! I appreciate it.

GPUassert: an illegal memory access was encountered sort_gpu.cu 418

Hello,

I am getting memory access error while testing the model. With the given dataset, it is working well. But when I gave my data(https://drive.google.com/file/d/1K8o3xGeHG_zMvzFuE3Ed0IX8CMqu-Mk7/view?usp=sharing), it is throwing above error.

Can you please help me out.

Thank you
Gopi

About visualization

Dear author, may I ask how the visualization of your paper is displayed? Looking forward to your reply, thank you very much

ply_reader does not have a function save_model

Running GenerateRueMadameDataSet.py gives the following error message:

Traceback (most recent call last):
File "GenerateRueMadameDataSet.py", line 26, in
from ply_reader import read_points_binary_ply, save_model
ImportError: cannot import name 'save_model'

The Rue Madame is now unavailable

It seems that the download link of Paris-rue-Madame is broken.
This means that all datasets used in TotalDenoising are unavailable now.
Can anyone upload the Rue Madame somewhere, so that I can make further research ?

Problem when running: Check failed: size >= 0 (-1092442872 vs. 0)

Dear author, I am trying to run this model on my own dataset for tests, and I have followed the instructions in Readme.md (except that I use the local data). However, After I run python Train.py --dataset 3, I receive the following results:
Epoch: 0 of 1
2023-07-25 15:55:24.120018: I tensorflow/stream_executor/dso_loader.cc:152] successfully opened CUDA library libcublas.so.10.0 locally
2023-07-25 15:55:24.319214: F tensorflow/core/framework/tensor_shape.cc:241] Check failed: size >= 0 (-1092442872 vs. 0)
I am wondering what may cause this and how can I fix it.
Thanks!

question about the loss

Hello,I have traned the model and test the model with the Ruedataset.In the ''log'' documention,I see that the loss increases with the increase of iteration.This is my question.Shouldn't the loss get smaller and smaller?Why it increase?

About the synthetic dataset and pretrained weight

Hi! Thank you for opening source such a great work! I just wonder are you still scheduling to release the synthetic dataset described in the paper (if my understanding is correct, that is the data from ModelNet40 right?). Also, I will be grateful if you can provide pretrained weights for your model. Thank you!

need for Synthetic dataset

Hello,I'm a student and I'm very interested in your research.Thank you for your code and can you provide me the Synthetic dataset mentioned in your paper? I really need it to do more research.Thank you very much.