guochengqian / pu-gcn Goto Github PK

python evaluate.py --gt "/home/cvgws-04/Interns_CEVI/upsample/PU-Net_pytorch/datas/test_data/our_collected_data/MC_5k" --pred "/home/cvgws-04/Interns_CEVI/upsample/PU-Net_pytorch/outputs/punet_reproduced_batchsize_12/preds" --save_path "/home/cvgws-04/Interns_CEVI/upsample/PU-Net_pytorch/outputs/punet_reproduced_batchsize_12"

I am trying run evaluate.py provided in your repository.
All ground truth .xyz are in -gt path and the predicted are in -pred path.

I am getting the following error:
Traceback (most recent call last): File "evaluate.py", line 12, in <module> from Common.ops import normalize_point_cloud File "/home/cvgws-04/Interns_CEVI/upsample/PU-GCN/Common/ops.py", line 4, in <module> from tf_ops.grouping.tf_grouping import knn_point_2 File "/home/cvgws-04/Interns_CEVI/upsample/PU-GCN/tf_ops/grouping/tf_grouping.py", line 10, in <module> grouping_module = tf.load_op_library( File "/home/cvgws-04/miniconda3/lib/python3.8/site-packages/tensorflow/python/framework/load_library.py", line 58, in load_op_library lib_handle = py_tf.TF_LoadLibrary(library_filename) tensorflow.python.framework.errors_impl.NotFoundError: /home/cvgws-04/Interns_CEVI/upsample/PU-GCN/tf_ops/grouping/tf_grouping_so.so: undefined symbol: _ZN10tensorflow14kernel_factory17OpKernelRegistrar12InitInternalEPKNS_9KernelDefEN4absl11string_viewESt10unique_ptrINS0_15OpKernelFactoryESt14default_deleteIS8_EE

Please let me know how to resolve.
Thanks in advance.

Hello, I want to know how to test pugan?Thank you very much!

I have downloaded the model that has been provided in this repository and plan to train it using the PU1K data that you have linked to. The ".h5" file provided in the training set, however, is empty, I have tried to view it using HDFview and pandas, in both cases no data is found.
I assume the "h5" file is actually empty as I receive a 'No such file or directory' message when I try to train the model in an Nvidia/cuda:10-devel-ubuntu18.04 docker container.
Please let me know if there is any alternative means to access the training data for your model.

-- This program requires the CGAL library, and will not be compiled.
-- Configuring done
-- Generating done
-- Build files have been written to: /data/home/scy0124/run/PU-GCN-master/evaluation_code
/var/spool/slurmd/job27981/slurm_script: line 14: ./evaluation: No such file or directory

Hello, I want to know where is the rean-scanned dataset? I'm sorry I can't find it in your PU1K dataset and PU-GAN code and dataset.

Can I use your pretrained model to upsample a random point cloud?

First of all, thank you for your code!
And I want to make my own dateset for training, but my data size is too small to train.
So I wonder how to crop a certain number of patches from a model?

Hi, @guochengqian ,

When will the code be released?

THX!

Hello, I am running this excellent project on kitti raw point cloud using provided pretrained model.
Strange hollows exisit in the upsampled clouds. Is it deal to cropping strategies or something else?
To reproduce:
bash test_realscan.sh pretrain/pu1k-punet ./data/real_scan_kitti_pugcn 0 --model punet --upsampler original bash test_realscan.sh pretrain/pu1k-mpu ./data/real_scan_kitti_pugcn 0 --model mpu --upsampler duplicate bash test_realscan.sh pretrain/pu1k-pugcn ./data/real_scan_kitti_pugcn 0 --model pugcn --k 20

using the 0th frame of 2011-09-26_0048 from kitti raw sequence.

Thank you!

Hello, I want to know how to test the effect of additive noise? Thank you very much!
I use the following code but it doesn't work.
source test_pu1k.sh pretrain/pu1k-pugcn/ 0 --model pugcn --k 20 --jitter_sigma 0.01

Hi, thanks for your work. While I have some questions about metric calculation and PU-GAN dataset. The first one is the hausdorff distance, your implementation here seems not consistent with the definition, which means it should be calculated as follow: hd_value = np.max(A, B) but not hd_value = np.max(A+B). I also check the code in PU-GAN (here) and DIS-PU (here), it seems you all calculate this metric mistakenly. The second is the point-to-surface distance, it seems that you only calculate the unidirectional point-to-surface distance here , but not the symmetrical point-to-surface and surface-to-point distance, just as chamfer distance. And I think the symmetric calculation is more convincing. The third is that I find the test set of PU-GAN dataset only contains mesh file, so does it mean each paper using PU-GAN dataset generates the input point cloud file and groundtruth point cloud file by their own? If so, how do you generate the input and groundtruth test pair for PU-GAN dataset? I also notice your PU1K dataset provide test point cloud file, but the loss of test point cloud file in PU-GAN really makes me confused. Thanks a lot, hope for your reply.

作者您好，首先非常感谢您的论文和代码。
请问可视化阶段这种局部放大的效果是怎么做出来的呢？

Both of these lines in Upsampling/model.py return index out of bounds errors

        patches = pc_util.extract_knn_patch(pc[np.asarray(seed_list), :], pc, self.opts.patch_num_point) 

(line 275 )

                    pred_pc = pred_pc[idx, 0:3]

(line 338)

First of all, thank you so much for the code
I have a problem using the code, so I'm asking you a question.
In the evaluation_code, the p2f value between my output point cloud and ground truth point cloud is Nan.
I took a print and found that pred_map_points[i] = shortest_paths.point(location.first,location.second); was Nan. I wonder if you've ever experienced this before. Also, what should I do if I had one?

I wanted to train the models on my own dataset, which is in form of meshes. Could you please share how did you convert your data into .h5 format? Is there a script for it?

Dear Guochenq Qian,

I have set up the environment and installed all necessary packages however I get the following error:
ModuleNotFoundError: No module named 'tf_lib'
I looked through the whole repository but could not find the tf_lib package which should include all methods.
Could it be that this script has not been uploaded yet?

Kind regards,
David

Hello, I only found the test data in . However, there is only 27 files with .off format. How can I get the test data in PU-GAN dataset?
Thank you very much for your time!

I executed test_pu1k.sh for pugcn.
the input is 2048 and expected output is 8192,
your provided code in test_pu1k.sh is upsampling 16 times giving 32768.
i checked the up_ratio argument it is set to 4
please provide your suggestions

Hi, thanks for amazing work, could you provide the pre-trained model you have in your paper ? Thanks !
And another question is that did you use monte-carlo sampling to generate the test samples (not ground turth) or use poission-disk sampling ?

您好，我想请问您放在google drive 里面的PU1K数据集压缩包，我解压出来之后看到数据集的后缀名中包含add_pugan的字样，我想请问下，这个数据集的表达意思是不是您在这个数据集中不仅仅是添加了PU1K数据集而且还添加了PUGAN数据集呢？

I change the up_ratio to 16, but the training reported an error: "Unable to open object (object 'poisson_4096' doesn't exist)"