https://github.com/SeanNaren/warp-ctc
下载下来&＃xff0c;传到服务器上
解压
更名
cd warp-ctc
mkdir build; cd build
cmake ..
make
上面这些过程缺啥装啥&＃xff0c;失败了,把缺的东西装上&＃xff0c;重复下面这个过程&＃xff0c;直到成功
cd build
make clean
cmake .. make
下一步开始安装
cd pytorch_binding
python setup.py install
报下面这个错&＃xff1a;
generating build/warpctc_pytorch/_warp_ctc/__warp_ctc.c
(already up-to-date)
not modified: &＃39;build/warpctc_pytorch/_warp_ctc/__warp_ctc.c&＃39;
running install
running bdist_egg
running egg_info
writing warpctc_pytorch.egg-info/PKG-INFO
writing dependency_links to warpctc_pytorch.egg-info/dependency_links.txt
writing top-level names to warpctc_pytorch.egg-info/top_level.txt
reading manifest file &＃39;warpctc_pytorch.egg-info/SOURCES.txt&＃39;
writing manifest file &＃39;warpctc_pytorch.egg-info/SOURCES.txt&＃39;
installing library code to build/bdist.linux-x86_64/egg
running install_lib
running build_py
copying warpctc_pytorch/_warp_ctc/__init__.py -> build/lib.linux-x86_64-3.6/warpctc_pytorch/_warp_ctc
running build_ext
building &＃39;warpctc_pytorch._warp_ctc.__warp_ctc&＃39; extension
gcc -pthread -B /home1/fzp/anaconda3/compiler_compat -Wl,--sysroot&＃61;/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-pib/include -I/home1/fzp/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/TH -I/home1/fzp/anaconda3/li-I/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/include -I/home1/fzp/anaconda3/include/python3.6m -c build/warpctc_pyto__warp_ctc.o -std&＃61;c&＃43;&＃43;11 -fPIC -std&＃61;c99 -DWARPCTC_ENABLE_GPU
cc1: warning: command line option ‘-std&＃61;c&＃43;&＃43;11’ is valid for C&＃43;&＃43;/ObjC&＃43;&＃43; but not for C [enabled by default]
gcc -pthread -B /home1/fzp/anaconda3/compiler_compat -Wl,--sysroot&＃61;/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-pib/include -I/home1/fzp/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/TH -I/home1/fzp/anaconda3/li-I/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/include -I/home1/fzp/anaconda3/include/python3.6m -c /home1/fzp/attenti4-3.6/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.o -std&＃61;c&＃43;&＃43;11 -fPIC -std&＃61;c99 -DWARPCTC_EN
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C&＃43;&＃43; [enabled by default]
cc1plus: warning: command line option ‘-std&＃61;c99’ is valid for C/ObjC but not for C&＃43;&＃43; [enabled by default]
/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.cpp: In function ‘int gpu_ctc(THCudaTensor*,
/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.cpp:92:49: error: cannot convert ‘THCudaTensotTensor*, int)’
int probs_size &＃61; THFloatTensor_size(probs, 2);
^
/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.cpp:105:61: error: invalid conversion from ‘s
void* gpu_workspace &＃61; THCudaMalloc(state, gpu_size_bytes);
^
/home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.cpp:105:61: error: too few arguments to funct
In file included from /home1/fzp/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/THC/THC.h:4:0,
from /home1/fzp/attention-OCR/warp-ctc-pytorch_bindings/pytorch_binding/src/binding.cpp:9:
/home1/fzp/anaconda3/lib/python3.6/site-packages/torch/utils/ffi/../../lib/include/THC/THCGeneral.h:209:21: note: declared h
THC_API cudaError_t THCudaMalloc(THCState *state, void **ptr, size_t size);
^
error: command &＃39;gcc&＃39; failed with exit status 1
解决方案&＃xff1a;
7.0.5和7.1.3版本的cudnn比较合适我的机器
(crnn) [xxx&＃64;xxxxxx crnn.pytorch-master]$ conda install cudnn&＃61;&＃61;7.1.3
Fetching package metadata .........................
Solving package specifications: .
UnsatisfiableError: The following specifications were found to be in conflict:
- cudnn &＃61;&＃61;7.1.3 -> cudatoolkit 8.0*
- libtorch-gpu -> cudatoolkit &＃61;&＃61;7.5
Use "conda info " to see the dependencies for each package.
报错&＃xff0c;版本不对应
再卸载再安装&＃xff1a;
conda uninstall cudatoolkit
Fetching package metadata .........................
Solving package specifications: .
Package plan for package removal in environment /export/gpudata/fujingling/conda/envs/crnn:
The following packages will be REMOVED:
cudatoolkit: 7.5-2 http://conda.jdfin.local/conda/free
libtorch-gpu: 0.1.12-0 http://conda.jdfin.local/conda/free
nccl: 1.3.4-cuda7.5_1 http://conda.jdfin.local/conda/free
一个一个的装回去
conda install cudatoolkit&＃61;&＃61;8.0
Fetching package metadata .........................
Solving package specifications: .
Package plan for installation in environment /export/gpudata/fujingling/conda/envs/crnn:
The following NEW packages will be INSTALLED:
cudatoolkit: 8.0-3 http://conda.jdfin.local/conda/free
Proceed ([y]/n)? y
装cudnn
conda install cudnn&＃61;&＃61;7.1.3(省略日志)
装pytorch
conda install pytorch&＃61;&＃61;0.4
Fetching package metadata .........................
Solving package specifications: .
Package plan for installation in environment /export/gpudata/fujingling/conda/envs/crnn:
The following NEW packages will be INSTALLED:
intel-openmp: 2019.0-118 http://repos.jd.com/conda/main
libgcc-ng: 8.2.0-hdf63c60_1 http://repos.jd.com/conda/main
libgfortran: 3.0.0-1 http://conda.jdfin.local/conda/free
libstdcxx-ng: 8.2.0-hdf63c60_1 http://repos.jd.com/conda/main
nccl: 1.3.5-cuda9.0_0 http://repos.jd.com/conda/main
ninja: 1.7.2-0 http://conda.jdfin.local/conda/free
openblas: 0.2.14-4 http://conda.jdfin.local/conda/free
pytorch: 0.4.0-py27hdf912b8_0 http://repos.jd.com/conda/main
The following packages will be UPDATED:
cudatoolkit: 8.0-3 http://conda.jdfin.local/conda/free --> 9.0-h13b8566_0 http://repos.jd.com/conda/main
mkl: 2017.0.3-0 http://conda.jdfin.local/conda/free --> 2019.0-118 http://repos.jd.com/conda/main
The following packages will be DOWNGRADED:
cudnn: 7.1.3-cuda8.0_0 http://repos.jd.com/conda/main --> 7.1.2-cuda9.0_0 http://repos.jd.com/conda/main
numpy: 1.13.1-py27_0 http://conda.jdfin.local/conda/free --> 1.10.2-py27_0 http://conda.jdfin.local/conda/free
Proceed ([y]/n)? y
最后经过一番折腾&＃xff0c;我发现&＃xff0c;把warp-ctc/pytorch_binding/warpctc_pytorch 这个文件夹移动到crnn工程的上一级目录下就好了
京公网安备 11010802041100号