iglict / octfield Goto Github PK

The OccNet and ConvOccNet are all trained on 13 categories of ShapeNet. According to your code, your network is trained on each category separately.
For the results of ConvOccNet as shown in Table 2, why are the Chamfer Distances of ConvONet even worse than the results trained on 13 categories as reported in the paper of ConvOccNet? And ConvOccNet is trained using sparse and noisy point clouds (3000 points) as input. However, your network is trained with clean point clouds as shown in Figure 4-(a). How can you do the comparison like this? Is it a fair comparison?

Hi, authors, thanks for your great work. I found that the implementation of models.RecursiveEncoder is missing. It looks like the models.py file is missing.

可以提供数据集划分的txt文件吗

pkg_part.py里生成.h5文件代码中又有点问题，这里的data_path是存放什么数据的

这里的npz文件是之前的哪一步生成的，之前的sample_tool生成的npz文件键值对应不上

Originally posted by @zouwenqin in #5 (comment)

Thank you for your work and for open-sourcing the code! I am trying to run your pipeline, but I am running into a couple of issues. I have followed the instructions from the repo to set up the code as well as the jittor instructions to enable CUDA support.

First, it seems like train_part.py isn't using CUDA, and it's running very slowly on my machine (1% in ~8 hours). If I add jt.flags.use_cuda = 1 I get the follwoing error:

python3.7: symbol lookup error: /home/ubuntu/.cache/jittor/jt1.3.4/g++7.5.0/py3.7.13/Linux-5.4.0-10xfe/IntelRCoreTMi9x9f/default/cu11.2.152_sm_86/jit/cudnn_conv3d_Tx_bool__Ty_float32__Tw_float32__JIT_1__JIT_cuda_1__index_t_int32__hash_b2c2838050d33db1_op.so: undefined symbol: _ZN6jittor11getDataTypeIbEE15cudnnDataType_tv

On the other hand, I can run python3.7 -m jittor.test.test_cudnn_op and python3.7 -m jittor.test.test_resnet without any errors.

Any ideas? Thanks for the help!

Hi, Thanks for your great work.
I want to know how you compute metrics in the Shape Reconstruction experiment?
Spercifically, to compute CD or EMD, how many points do you sample from reconstructed and gt mesh? Do you use L1 or L2 distance? "CD = one_direction + another_direction" or "CD = 0.5 * (one_direction + another_direction)"?

Hi,
I find the report about your excellent work in 计图开源：基于层次结构的隐式几何表示和建模
And I notice that you did some experiments about IMNet: Learning implicit fields for generative shape modeling.
Would you please release the code about IMNet?

Thanks in advanced.

编译后的imp_samplingv1.cpython-39-x86_64-linux-gnu.so在ｕｔｉｌ导入的时候有问题

用ｌｄｄ　－ｒ　查看ｓｏ文件的依赖，编译后的文件如下：

这个imp_sampling指的是编译后的.so文件吗，imp_sampling文件夹下没有SamplesGenerator的库，imp_sampling.cpp文件里有相关函数，不过还需要编译后用cython等工具处理，python文件才能用吧，这个util.py文件直接导入好像有问题

Hi,
Thank you for releasing the code! :)
I'm trying to run the code and I've successfully compiled imp_sampling. However, I cannot find the file "generateSamplesWind" after compilation and I cannot find any code related to generateSamplesWind in imp_sampling/build/Makefile.