Pytorch implementation of PCT: Point Cloud Transformer

请问一下这种情况该怎么解决，运行的时候报了这个错误，已经安装了ninja还是不可以

roup_points_gpu.cuda.o FAILED: group_points_gpu.cuda.o /usr/bin/nvcc --generate-dependencies-with-compile --dependency-output group_points_gpu.cuda.o.d -DTORCH_EXTENSION_NAME=ext -DTORCH_API_INCLUDE_EXTENSION_H -DPYBIND11_COMPILER_TYPE="gcc" -DPYBIND11_STDLIB="libstdcpp" -DPYBIND11_BUILD_ABI="cxxabi1011" -I/home/lin/program/PCT_Pytorch-main/pointnet2_ops_lib/pointnet2_ops/ext-src/include -isystem /home/lin/anaconda3/lib/python3.8/site-packages/torch/include -isystem /home/lin/anaconda3/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -isystem /home/lin/anaconda3/lib/python3.8/site-packages/torch/include/TH -isystem /home/lin/anaconda3/lib/python3.8/site-packages/torch/include/THC -isystem /home/lin/anaconda3/include/python3.8 -D_GLIBCXX_USE_CXX11_ABI=0 -D__CUDA_NO_HALF_OPERATORS -D__CUDA_NO_HALF_CONVERSIONS -D__CUDA_NO_BFLOAT16_CONVERSIONS_ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_37,code=compute_37 -gencode=arch=compute_37,code=sm_37 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_62,code=sm_62 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -O3 -Xfatbin -compress-all -std=c++14 -c /home/lin/program/PCT_Pytorch-main/pointnet2_ops_lib/pointnet2_ops/_ext-src/src/group_points_gpu.cu -o group_points_gpu.cuda.o nvcc fatal : Unknown option '-generate-dependencies-with-compile' ninja: build stopped: subcommand failed.

ModuleNotFoundError: No module named '_ext'

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "main.py", line 9, in from model import Pct File "/home/cx/NewDisk/xxl/PCT_Pytorch-main/model.py", line 4, in from util import sample_and_group File "/home/cx/NewDisk/xxl/PCT_Pytorch-main/util.py", line 3, in from pointnet2_ops_lib.pointnet2_ops import pointnet2_utils File "/home/cx/NewDisk/xxl/PCT_Pytorch-main/pointnet2_ops_lib/pointnet2_ops/init.py", line 1, in from . import pointnet2_modules File "/home/cx/NewDisk/xxl/PCT_Pytorch-main/pointnet2_ops_lib/pointnet2_ops/pointnet2_modules.py", line 6, in from . import pointnet2_utils File "/home/cx/NewDisk/xxl/PCT_Pytorch-main/pointnet2_ops_lib/pointnet2_ops/pointnet2_utils.py", line 30, in with_cuda=True, File "/home/cx/NewDisk/anaconda3/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 661, in load is_python_module) File "/home/cx/NewDisk/anaconda3/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 841, in _jit_compile return _import_module_from_library(name, build_directory, is_python_module) File "/home/cx/NewDisk/anaconda3/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 1048, in _import_module_from_library file, path, description = imp.find_module(module_name, [path]) File "/home/cx/NewDisk/anaconda3/lib/python3.6/imp.py", line 297, in find_module raise ImportError(_ERR_MSG.format(name), name=name) ImportError: No module named '_ext'

I got this when training modelnet40 dataset; RuntimeError: CUDA out of memory. Tried to allocate 16.00 MiB (GPU 0; 3.94 GiB total capacity; 2.98 GiB already allocated; 14.25 MiB free; 3.00 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

And it dosent work, even if I reduce the batchsize(2). Can I have any other options, besides a new GPU with more memory? How much memory do you use when training? Thanks a lot!

Hi @Strawberry-Eat-Mango @uyzhang, Thanks for sharing the PyTorch implementation of PCT!

I follow the instructions in this repo and train PCT from scratch on ModelNet40. First, I prepare the dataset. Then configure the training parameters, shown in Fig.1, and keep others default (6 2080ti devices are used, but it's weird that the last two occupy fewer memories, shown in Fig.2). Third, 'scheduler.step()' was added as other issues suggested to eliminate a warning.

Fig. 1

Fig. 2

Everything looks good: the loss reduces and the accuracy increases during training. One exceptional point is that the accuracy increases slowly, e.g. Accuracy achieves only 29% after 200 epochs of training, and finally the accuracy stops at 45.91% in 250 epochs of training, shown in Fig.3 . Hence, the results are far from expectations.

Fig. 3

What are possible reasons? Can anyone give some tips or suggestions?

Appreciate!

Dear authors,

Thank you for sharing your code and contributions. I would like to use your code in my project, while there is no LICENSE present in your repository. May I ask in what ways we could use or modify the code, and it wouild be great if you could add the license to the repository.

Thank you so much!

Thanks for this implementation!

I ran the test code with

python main.py --exp_name=test --num_points=1024 --use_sgd=True --eval=True --model_path=checkpoints/best/models/model.t7 --test_batch_size=4

(changed test_batch_size to 4 from 8 due to GPU memory constraints)

But the accuracy is really bad:

Test :: test acc: 0.483387, test avg acc: 0.455517

The only thing i did slightly differently to the original code is that, I downloaded the h5 data manually instead of using the code, as otherwise i met certificate issues.

Can anyone else replicate the 93% accuracy as reported in the repo?

update setting --test_batch_size=1 achieves test acc: 0.932739, test avg acc: 0.899901 another observation, if test_batch_size > 1, prediction result is not reproducible / deterministic.

I'm using pytorch 1.6.0

RuntimeError: Error(s) in loading state_dict for DataParallel: Missing key(s) in state_dict: "module.pt_last.pos_xyz.weight", "module.pt_last.pos_xyz.bias". Unexpected key(s) in state_dict: "module.pt_last.conv2.weight", "module.pt_last.bn2.weight", "module.pt_last.bn2.bias", "module.pt_last.bn2.running_mean", "module.pt_last.bn2.running_var", "module.pt_last.bn2.num_batches_tracked".

Run to pointnet2_ When using utils.py, prompt userw arning:“Uable to load pointnet2_ OPD CPP extension. JIT compling”, and then the program is stuck here, and the run cursor keeps flashing. What's wrong? Thank you

I'm new to this filed, when I tried to install pointnet2_ops_lib, I got this:

(transformer) user002@user002-OptiPlex-7010:~/下载/PCT_Pytorch-main$ pip install pointnet2_ops_lib/. Processing ./pointnet2_ops_lib Preparing metadata (setup.py) ... error error: subprocess-exited-with-error

× python setup.py egg_info did not run successfully. │ exit code: 1 ╰─> [12 lines of output] Traceback (most recent call last): File "", line 36, in File "", line 34, in File "/home/user002/下载/PCT_Pytorch-main/pointnet2_ops_lib/setup.py", line 34, in include_dirs=[osp.join(this_dir, _ext_src_root, "include")], File "/home/user002/miniconda3/envs/transformer/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 779, in CUDAExtension library_dirs += library_paths(cuda=True) File "/home/user002/miniconda3/envs/transformer/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 869, in library_paths if (not os.path.exists(_join_cuda_home(lib_dir)) and File "/home/user002/miniconda3/envs/transformer/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 1783, in _join_cuda_home raise EnvironmentError('CUDA_HOME environment variable is not set. ' OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root. [end of output]

note: This error originates from a subprocess, and is likely not a problem with pip. error: metadata-generation-failed

× Encountered error while generating package metadata. ╰─> See above for output.

note: This is an issue with the package mentioned above, not pip. hint: See above for details.

How can I solve it?

Hi, thanks for your work! I found that there is some difference between your code and origin jittor implemention. In the class Point_Transformer_Last module. you don't input the xyz information, where the original implemention takes a position embedding. Have you compared the two results? Appreciate!

Use SGD CUDA kernel failed : invalid device function void furthest_point_sampling_kernel_wrapper(int, int, int, const float*, float*, int*) at L:228 in pointnet2_ops/_ext-src/src/sampling_gpu.cu 段错误 (核心已转储)

UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`.  Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate "https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate", UserWarning)

Is it ok?