@lzhengning 您好！

感谢您能够开源这么优秀的项目！

目前我在尝试使用自己的数据集进行训练，但是现在遇到一些问题： 1、数据集预处理部分，那个json文件中的参数我没搞明白，我看过了您在#9 中的回答，那个sub_labels和raw_to_sub的数据是如何获取的，trimesh.proximity.closest_point中的输入是一个mesh数据和一个points，这个points应该选择哪些点的坐标？mesh的顶点吗？ 2、在打标签时，应该在原始数据上进行还是在进行remesh后的mesh中进行？ 3、我目前需要训练的mesh数据每个都包含了10w个左右的三角面片，请问还需要运行MAPS算法吗？如果需要，是否应该先对mesh进行简化后再运行该算法？

期待您的回答，谢谢！

”name: manifold40 Train 0: 0%| | 0/2049 [00:00<?, ?it/s] Compiling Operators(1/1) used: 19.2s eta: 0s

Compiling Operators(1/1) used: 13s eta: 0s

Compiling Operators(1/1) used: 13s eta: 0s

Compiling Operators(1/1) used: 13.8s eta: 0s

Compiling Operators(56/56) used: 78.5s eta: 0s

Compiling Operators(44/44) used: 60.4s eta: 0s

Compiling Operators(1/1) used: 16s eta: 0s Train 0: 0%| | 1/2049 [03:56<134:34:05, 236.55s/it] Compiling Operators(2/2) used: 19.3s eta: 0s

Compiling Operators(1/1) used: 14.4s eta: 0s Train 0: 1%|▋ | 21/2049 [04:52<1:14:44, 2.21s/it][w 0624 09:21:26.817385 20 cudnn_conv_Tx:float32__Ty:float32__Tw:float32__XFORMAT:abcd__WFORMAT:oihw__YFORMAT:abcd__J...hash:798cb5ed49dadaa2_op.cc:200] forward_ algorithm cache is full [w 0624 09:21:27.423196 20 cudnn_conv_backward_w_Tx:float32__Ty:float32__Tw:float32__XFORMAT:abcd__WFORMAT:oihw__YFOR...hash:c41e3d43aa5d4cf7_op.cc:199] backward w algorithm cache is full Train 0: 1%|▊ | 24/2049 [04:56<56:14, 1.67s/it][w 0624 09:21:31.279631 20 cudnn_conv_backward_x_Tx:float32__Ty:float32__Tw:float32__XFORMAT:abcd__WFORMAT:oihw__YFOR...hash:74f24b7a5fa4fe17_op.cc:201] backward x algorithm cache is full Train 0: 65%|███████████████████████████████████████████▌ | 1332/2049 [1:21:43<51:18, 4.29s/it] “ The training process consume too much time. I utilize 1 gpu (12G) to train. An hour later, Train 0 is not over. I noticed that you mentioned that using multiple gpus could speed up the trainning, so I want to know how many gpus do you utilize and how long is the training time?

Thank you very much!

Hi @lzhengning, Thanks for releasing this code.

I would like to test your pretrained segmentation model on .obj meshes without groundtruth, do you think it is easily achievable ? Is there a particular tool you use to label the meshes and get the .json files present in your repository ?

Kind regards.

The results of PD-MeshNet in your paper are very different from the results in the original paper. I also notice your answer to question 15. But the metric of PD-MeshNet is on the primal nodes (on faces), which means that the metric is strict. Therefore, I'm curious about the reason. Thanks! Have a nice day!

你好 我在进行manifold数据集上的分类任务训练时，遇到了下面的错误

name:  manifold40
Train 0:   0%|                                                                                                                                                                           | 0/2049 [00:01<?, ?it/s]
Traceback (most recent call last):
  File "train_cls.py", line 184, in <module>
    train(net, optim, train_dataset, writer, epoch)
  File "train_cls.py", line 33, in train
    mesh_tensor = to_mesh_tensor(meshes)
  File "/mnt/data/wy/SubdivNet/subdivnet/utils.py", line 28, in to_mesh_tensor
    jt.int32(meshes['Fs']))
  File "/mnt/data/wy/SubdivNet/subdivnet/mesh_tensor.py", line 35, in __init__
    assert self.F == self.Fs.max().data[0]
AssertionError

我是在ubuntu18.04上用conda搭建环境的，jittor版本1.3.0.14，期待您的答复。

使用了cubes、manifold40、shrec11-split10、shrec11-split16的预训练模型，分类结果：

cubes：

name:  Cubes
Test 0: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 103/103 [01:09<00:00,  1.48it/s]
test acc =  0.00015174506828528073
test acc [voted] =  0.0

manifold40：

name:  Manifold40
Test 0:   5%|███████                                                                                                                          | 112/2057 [00:11<04:29,  7.22it/s]
[w 0723 15:28:27.917150 76 cudnn_conv_Tx:float32__Ty:float32__Tw:float32__XFORMAT:abcd__WFORMAT:oihw__YFORMAT:abcd__J...hash:798cb5ed49dadaa2_op.cc:200] forward_ algorithm cache is full
Test 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2057/2057 [03:12<00:00, 10.67it/s]
test acc =  0.037560777957860614
test acc [voted] =  0.03727714748784441

shrec11-split10：

name:  shrec11-split10
Test 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 99/99 [00:48<00:00,  2.04it/s]
test acc =  0.0
test acc [voted] =  0.0

shrec11-split16：

name:  shrec11-split16
Test 0: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 40/40 [00:28<00:00,  1.43it/s]
test acc =  0.0
test acc [voted] =  0.0

Hello. Thanks for your work. However, when I try to run the test script of coseg-alien, it gives me an error message like this:

name: coseg-alien
0:   0%|                    | 0/37 [00:00<?, ?it/s]/home/xxx/.cache/jittor/default/g++/jit/_opkey0:array_T:int32__JIT:1__JIT_cuda:1__index_t:int32___opkey1:broadcast_to_Tx:int32__DI...hash:bc2f95c82b48131a_op.cc(40): error: calling a constexpr __host__ function("floor") from a __global__ function("func_bc2f95c82b48131a_0") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

1 error detected in the compilation of "/home/xxx/.cache/jittor/default/g++/jit/_opkey0:array_T:int32__JIT:1__JIT_cuda:1__index_t:int32___opkey1:broadcast_to_Tx:int32__DI...hash:bc2f95c82b48131a_op.cc".
[e 0721 14:21:37.874137 48:C15 parallel_compiler.cc:261] [Error] source file location: /home/xxx/.cache/jittor/default/g++/jit/_opkey0:array_T:int32__JIT:1__JIT_cuda:1__index_t:int32___opkey1:broadcast_to_Tx:int32__DI...hash:bc2f95c82b48131a_op.cc
[e 0721 14:21:37.874176 48:C15 parallel_compiler.cc:264] Compile fused operator(18/56) failed: [Op(0x5561911a3de0:0:0:1:i0:o1:s0,array->0x5561911a3e80),Op(0x5561911a2360:0:0:1:i1:o1:s0,broadcast_to->0x5561911a2410),Op(0x5561911a37f0:0:0:1:i2:o1:s0,binary.mod->0x5561911a30a0),] 

Reason: [f 0721 14:21:37.873882 48:C15 log.cc:387] Check failed ret(256) == 0(0) Run cmd failed: cd /home/xxx/.cache/jittor/default/g++ && /home/xxx/.cache/jittor/jtcuda/cuda11.2_cudnn8_linux/bin/nvcc '/home/xxx/.cache/jittor/default/g++/jit/_opkey0:array_T:int32__JIT:1__JIT_cuda:1__index_t:int32___opkey1:broadcast_to_Tx:int32__DI...hash:bc2f95c82b48131a_op.cc'     -std=c++14 -Xcompiler -fPIC  -Xcompiler -march=native  -Xcompiler -fdiagnostics-color=always  -I/home/xxx/anaconda3/envs/subdivnet/lib/python3.7/site-packages/jittor/src -I/home/xxx/anaconda3/envs/subdivnet/include/python3.7m -I/home/xxx/anaconda3/envs/subdivnet/include/python3.7m -DHAS_CUDA -I'/home/xxx/.cache/jittor/jtcuda/cuda11.2_cudnn8_linux/include' -I'/home/xxx/anaconda3/envs/subdivnet/lib/python3.7/site-packages/jittor/extern/cuda/inc'  -lstdc++ -ldl -shared  -x cu --cudart=shared -ccbin='/usr/bin/g++' --use_fast_math  -w  -I'/home/xxx/anaconda3/envs/subdivnet/lib/python3.7/site-packages/jittor/extern/cuda/inc'  -arch=compute_61  -code=sm_61  -o '/home/xxx/.cache/jittor/default/g++/jit/_opkey0:array_T:int32__JIT:1__JIT_cuda:1__index_t:int32___opkey1:broadcast_to_Tx:int32__DI...hash:bc2f95c82b48131a_op.so'
0:   0%|                    | 0/37 [00:02<?, ?it/s]
Traceback (most recent call last):
  File "train_seg.py", line 162, in <module>
    test(net, test_dataset, writer, 0, args)
  File "/home/xxx/anaconda3/envs/subdivnet/lib/python3.7/site-packages/jittor/__init__.py", line 257, in inner
    ret = func(*args, **kw)
  File "train_seg.py", line 64, in test
    preds = np.argmax(outputs.data, axis=1)
RuntimeError: Wrong inputs arguments, Please refer to examples(help(jt.data)).

Types of your inputs are:
 self   = Var,

The function declarations are:
 inline DataView data()

Failed reason:[f 0721 14:21:38.068584 96 parallel_compiler.cc:316] Error happend during compilation, see error above.

Any ideas on why this happens? I have downloaded the data of coseg-alien via the shell script provided.

I found the datasets download via "sh scripts/<DATASET_NAME>/get_data.sh" are original datasets without the subdivision connectivity. Moreover, it is very hard to remesh them by datagen_maps.py for the remeshing is always interrupted by some problems in the program, especially for the Manifold40.

Therefore, could you please provide the remeshed datasets with subdivision connectivity? Thank you very much!

Hi, I'm trying to reproduce your results on Manifold40.

I downloaded the dataset with the command sh scripts/manifold40/get_data.sh, which downloads the file Manifold40-MAPS-96-3.zip. I noticed in your instructions that it is possible to download a different version of Manifold40 "before remeshing". Which is the correct version to use for evaluation?

I noticed that the version in Manifold40-MAPS-96-3.zip contains approximatively 10 times the number of shapes contained in the original ModelNet40, why is that?

I run the command sh scripts/manifold40/get_pretrained.sh to obtain the checkpoint for the model and then I run the command sh scripts/manifold40/test.sh to perform evaluation. The evaluation runs fine but I get a very low accuracy. Do you have any idea of what I could have missed?

Thanks in advance, Luca.

I'm trying to use your script datagen_maps.py to preprocess some meshes in order to feed them to SubDivNet. The meshes are obtained with marching cubes and I tried different resolutions to get meshes with different number of vertices/faces. The preprocessing fails in the decimate() function here, without any meaningful error (at least that I can understand).

Do you have any suggestion? Is there any special requirement for the meshes in order to be able to preprocess them?

Thanks in advance!

Hello, thanks for your great work and offered code.

i'm working on a school project and i want to use the SUBDIVNET code to train my data, but there are some points i did not understand..

in each .obj file there are 3 indexes:

• raw_labels: each item here represents a label of a face in the original data.
• sub_labels: each item here represents a label of a face in the data after remeshing it (after applying datagen_maps.py on it )
• raw_to_sub: this i don't understand what does it represent, and what is its role

another thing i didn't understand: length of raw_labels and sub_labels is 13776 but number of faces in each object is 16384. so if raw_labels and sub_labels are representing the label of each face why they don't have same length as faces.

so can you please explain to me how can i preprocess step by step my data to before feeding it to the subdivnet? how can i make each object have the exact number of faces as in your input data (16384) and how to label my data, and if you can recommend me a tool i can use!

thank you in advance

Thanks for sharing the code of your nice work! I've been trying to apply the framework to my own dataset for segmentation. As I understood, the dataroot contains the folders 'train', 'test' and 'raw'. For each sample, there are three files: *.obj, *.json (labels), and the *.off (in 'raw'). What is unclear to me are:

1. What are the data in the 'raw' folder? I expect the raw data to be *.obj. What is the purpose of saving the meshes in .off instead of .obj? The remeshing algorithms (MAPS and neural subdivision) have the finest mesh as input and downsample the meshes. However, the meshes in the 'raw' folders have a lower resolution than those in the 'train' and 'test' folders.
2. How do you obtain the labels (*.json)?

Thank you!

Hi, thanks for you great paper and code! When I want to make use of the new dataset Manifold40 you proposed from tsinghua cloud to training my own model, I found that there was a mere mistake in only one sample that is 'bookshelf_0099.obj'. The content of the version I downloaded is:

Object res.obj

Vertices: 245

Faces: 842

with 842 faces, not 500, so I remeshed it in meshlab to make the code works.