czHappy
commented
3 years ago @yarkable Could you please help me to solve the problem? Thank you very much!
Closed czHappy closed 3 years ago
czHappy
commented
3 years ago @yarkable Could you please help me to solve the problem? Thank you very much!
yarkable
commented
3 years ago It seems that you have some problems while importing modules, maybe you can check if you have a __init__.py in every package.
czHappy
commented
3 years ago It seems that you have some problems while importing modules, maybe you can check if you have a
__init__.pyin every package.
This is the result of 'tree' command of the project root directory:
[sk49@g33 ~/new_workspace/cz/MODNet-master]$ tree
.
|-- demo
| |-- image_matting
| | `-- colab
| | |-- inference.py
| | `-- README.md
| `-- video_matting
| |-- custom
| | |-- README.md
| | |-- requirements.txt
| | `-- run.py
| `-- webcam
| |-- README.md
| |-- requirements.txt
| `-- run.py
|-- doc
| `-- gif
| |-- homepage_demo.gif
| |-- image_matting_demo.gif
| `-- video_matting_demo.gif
|-- onnx
| |-- export_onnx.py
| |-- inference_onnx.py
| |-- __init__.py
| |-- modnet_onnx.py
| |-- README.md
| `-- requirements.txt
|-- pretrained
| |-- mobilenetv2_human_seg.ckpt
| |-- modnet_photographic_portrait_matting.ckpt
| |-- modnet_webcam_portrait_matting.ckpt
| `-- README.md
|-- README.md
|-- src
| |-- __init__.py
| |-- models
| | |-- backbones
| | | |-- __init__.py
| | | |-- mobilenetv2.py
| | | |-- __pycache__
| | | | |-- __init__.cpython-36.pyc
| | | | |-- __init__.cpython-37.pyc
| | | | |-- mobilenetv2.cpython-36.pyc
| | | | |-- mobilenetv2.cpython-37.pyc
| | | | |-- wrapper.cpython-36.pyc
| | | | `-- wrapper.cpython-37.pyc
| | | `-- wrapper.py
| | |-- __init__.py
| | |-- modnet.py
| | `-- __pycache__
| | |-- __init__.cpython-36.pyc
| | `-- __init__.cpython-37.pyc
| |-- __pycache__
| | |-- __init__.cpython-36.pyc
| | `-- __init__.cpython-37.pyc
| `-- trainer.py
`-- torchscript
|-- export_torchscript.py
|-- __init__.py
|-- __init__.pyc
|-- modnet_torchscript.py
|-- __pycache__
| |-- export_torchscript.cpython-36.pyc
| |-- export_torchscript.cpython-37.pyc
| |-- __init__.cpython-36.pyc
| |-- __init__.cpython-37.pyc
| |-- modnet_torchscript.cpython-36.pyc
| `-- modnet_torchscript.cpython-37.pyc
`-- README.md
I cloned the whole project and downloaded the models to pretrained folder. Then I use the command in project root directory:
python3 -m torchscript.export_torchscript \
--ckpt-path=pretrained/modnet_photographic_portrait_matting.ckpt \
--output-path=pretrained/modnet_photographic_portrait_matting.torchscriptSo I think the init.py is OK. Could you tell me your environment details or other possible errors in my operations ?
czHappy
commented
3 years ago By the way, I try to use the modnet.pt directly to build executable file, but fail to load the model . The libtorch version is libtorch-cxx11-abi-shared-with-deps-1.3.1 It is cpu version of libtorch for linux.
CMakeLists.txt
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example-app)
find_package(Torch REQUIRED)
add_executable(example-app example-app.cpp)
target_link_libraries(example-app "${TORCH_LIBRARIES}")
set_property(TARGET example-app PROPERTY CXX_STANDARD 14)example-app.cpp:
#include <torch/script.h> // One-stop header.
#include <iostream>
#include <memory>
int main(int argc, const char* argv[]) {
if (argc != 2) {
std::cerr << "usage: example-app <path-to-exported-script-module>\n";
return -1;
}
torch::jit::script::Module module;
try {
// Deserialize the ScriptModule from a file using torch::jit::load().
module = torch::jit::load(argv[1]);
}
catch (const c10::Error& e) {
std::cerr << "error loading the model\n";
return -1;
}
std::cout << "ok\n";
}build successfully but fail to execute ./example-app:
command: ./example-app ../modnet.pt result: error loading the model.
Could you please tell us the environment required and how to use the modnet.pt in cpp?
yarkable
commented
3 years ago Hi, I just clone the project and download official pretrained model, then I type the command
python3 -m torchscript.export_torchscript \
--ckpt-path=pretrained/modnet_photographic_portrait_matting.ckpt \
--output-path=pretrained/modnet_photographic_portrait_matting.torchscriptIt works successfully, maybe you should clone it once again and see whether the error still occurs?
Btw, I just convert it to TorchScript version for IOS device, we have no problem using modnet.pt.
czHappy
commented
3 years ago What is your torch version and libtorch version, please?
yarkable
commented
3 years ago Btw, here is my code for classification using C++.
int main(int argc, char * argv[]){
if (argc != 2) {
std::cerr << "no module found !\n";
return -1;
}
torch::jit::script::Module module;
try {
module = torch::jit::load(argv[1]);
}
catch (const c10::Error& e){
std::cerr << "error loading the module\n";
return -1;
}
std::cout << "ok!\n";
/////////////////////////////////////////////////
string path = "/pytorch-deployment/assets/3.jpg";
Mat img = imread(path), img_float;
cvtColor(img, img, CV_BGR2RGB);
bitwise_not(img, img);
vector<Mat> mv;
split(img, mv);
img = mv[1];
img.convertTo(img_float, CV_32F, 1.0 / 255);
resize(img_float, img_float , Size(28, 28));
auto img_tensor = torch::from_blob(img_float.data, {1, 28, 28, 1}, at::kFloat).permute({ 0,3,1,2 });
auto img_var = torch::autograd::make_variable(img_tensor, false);
// Create a vector of inputs.
std::vector<torch::jit::IValue> inputs;
inputs.push_back(img_var);
auto output = module.forward(inputs).toTensor();
cout << output << endl;
auto index = output.argmax(1);
cout << "The predicted class is : " << index << endl;
}You can re-clone the project and generate a TorchScript model to see if your code is correct.
yarkable
commented
3 years ago torch 1.6.0 and I don't have a libtorch
czHappy
commented
3 years ago I tried again with torch verion 1.3.1(gpu), the same errors occured. Maybe I should install torch 1.6.0 and try again. And if you don't have a libtorch, how can you use 'torch::jit::script::Module' in your cpp file? Which headers do you include before main function?
yarkable
commented
3 years ago @czHappy Lol, I used to use libtorch. But in this project, I just export it to TorchScript version and give it to the other engineer.🤣
czHappy
commented
3 years ago @yarkable I use torch 1.6.0 (CPU) and modify your export script to produce a cpu torchscript model successfully. It proves that the required version (torch >= 1.2.0) is not accurate, I have tried torch 1.2.0, 1.3.1, 1.4.0 but all of them didn't work. Now I find torch 1.6.0 is OK. Then I use C++ to load the pt file and test forward function, it really works! Anyway, thanks for your excellent job and patient reply! The details are as follows:
CMakeLists.txt
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example-app)
find_package(Torch REQUIRED)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")
add_executable(example-app example-app.cpp)
target_link_libraries(example-app "${TORCH_LIBRARIES}")
set_property(TARGET example-app PROPERTY CXX_STANDARD 14)
# The following code block is suggested to be used on Windows.
# According to https://github.com/pytorch/pytorch/issues/25457,
# the DLLs need to be copied to avoid memory errors.
if (MSVC)
file(GLOB TORCH_DLLS "${TORCH_INSTALL_PREFIX}/lib/*.dll")
add_custom_command(TARGET example-app
POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different
${TORCH_DLLS}
$<TARGET_FILE_DIR:example-app>)
endif (MSVC)example-app.cpp
#include <torch/script.h> // One-stop header.
#include <vector>
#include <iostream>
#include <memory>
int main(int argc, const char* argv[]) {
if (argc != 2) {
std::cerr << "usage: example-app <path-to-exported-script-module>\n";
return -1;
}
torch::jit::script::Module module;
try {
// Deserialize the ScriptModule from a file using torch::jit::load().
module = torch::jit::load(argv[1]);
}
catch (const c10::Error& e) {
std::cerr << "error loading the model\n";
return -1;
}
std::cout << "ok\n";
// Create a vector of inputs.
std::vector<torch::jit::IValue> inputs;
inputs.push_back(torch::ones({1, 3, 640, 480}));
// Execute the model and turn its output into a tensor.
at::Tensor output = module.forward(inputs).toTensor();
//std::cout << output.slice(/*dim=*/1, /*start=*/0, /*end=*/5) << '\n';
std::cout<<output[0][0][0][0]<<std::endl;
}
Just like Minimal Example of official document, https://pytorch.org/cppdocs/installing.html my environment: ubuntu 18.04 LTS GCC 7.5.0 cmake 3.10.2 torch 1.6.0 CPU libtorch 1.6.0 CPU I think GPU torch version is OK, too.
yarkable
commented
3 years ago Good job
When I export the TorchScript version of MODNet, error occurs:
Environment: Centos 7 torch 1.3.1 gpu