使用libtorch将pytorch 部署到移动端去
使用resnet50举例
使用代码将pytorch模型转为libtorch
import torch
import torchvision.models as models
from PIL import Image
import numpy as np
#加载测试图片调整大小
image = Image.open("test.jpg") #图片发在了build文件夹下
image = image.resize((224, 224),Image.ANTIALIAS)
#进行预处理
image = np.asarray(image)
image = image / 255
image = torch.Tensor(image).unsqueeze_(dim=0)
#变换维度
image = image.permute((0, 3, 1, 2)).float()
#加载使用pytorch自带resnet50模型
model = models.resnet50(pretrained=True)
model = model.eval()
resnet = torch.jit.trace(model, torch.rand(1,3,224,224))
# output=resnet(torch.ones(1,3,224,224))
#使用测试模型转换
output = resnet(image)
max_index = torch.max(output, 1)[1].item()
print(max_index) # ImageNet1000类的类别序
#保存转化后的模型
resnet.save('resnet.pt')
使用c++调用写好的模型
再c++调用模型之前要先写好CMakeLists文件
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example_torch)
set(CMAKE_PREFIX_PATH "XXX/libtorch") //注意这里填自己解压libtorch时的路径
find_package(Torch REQUIRED)
find_package(OpenCV 3.0 QUIET)
if(NOT OpenCV_FOUND)
find_package(OpenCV 2.4.3 QUIET)
if(NOT OpenCV_FOUND)
message(FATAL_ERROR "OpenCV > 2.4.3 not found.")
endif()
endif()
add_executable(${PROJECT_NAME} "main.cpp")
target_link_libraries(${PROJECT_NAME} ${TORCH_LIBRARIES} ${OpenCV_LIBS})
set_property(TARGET ${PROJECT_NAME} PROPERTY CXX_STANDARD 11)```
其中要设置好CMAKE_PREFIX_PATH路径,这个路径就是我们解压libtorch的路径,不然无法链接到libtorch库,其中也设置了OpenCV的配置,具体OpenCV的安装这里介绍了。
然后就是C++调用PyTorch模型的代码
#include <torch/torch.h>
#include <torch/script.h>
#include <iostream>
#include <vector>
#include <opencv2/highgui.hpp>
#include <opencv2/core/core.hpp>
#include <opencv2/opencv.hpp>
void TorchTest(){
std::shared_ptr<torch::jit::script::Module> module = torch::jit::load("../resnet.pt");
assert(module != nullptr);
std::cout << "Load model successful!" << std::endl;
std::vector<torch::jit::IValue> inputs;
inputs.push_back(torch::zeros({1,3,224,224}));
at::Tensor output = module->forward(inputs).toTensor();
auto max_result = output.max(1, true);
auto max_index = std::get<1>(max_result).item<float>();
std::cout << max_index << std::endl;
}
void Classfier(cv::Mat &image){
torch::Tensor img_tensor = torch::from_blob(image.data, {1, image.rows, image.cols, 3}, torch::kByte);
img_tensor = img_tensor.permute({0, 3, 1, 2});
img_tensor = img_tensor.toType(torch::kFloat);
img_tensor = img_tensor.div(255);
std::shared_ptr<torch::jit::script::Module> module = torch::jit::load("../Train/resnet.pt");
torch::Tensor output = module->forward({img_tensor}).toTensor();
auto max_result = output.max(1, true);
auto max_index = std::get<1>(max_result).item<float>();
std::cout << max_index << std::endl;
}
int main() {
// TorchTest();
cv::Mat image = cv::imread("airliner.jpg");
cv::resize(image,image, cv::Size(224,224));
std::cout << image.rows <<" " << image.cols <<" " << image.channels() << std::endl;
Classfier(image);
return 0;
}
其中TorchTest函数只是做了简单的演示,而Classfier通过OpenCV读取图片,并通过libtorch的函数将Mat格式转换成Tensor(注意:这里转换了维度,因为OpenCV的维度是[H,W,C], 而PyTorch模型需要的是[C,H,W]),最后依然能够输出和Python代码一样的答案。
这里比较重要的几个函数有:
MatTensor
torch::jit::load(): 该函数顾名思义就是加载模型的函数。
module->forward(): 模型前向传播的函数,输入值建议使用vector类型
max(): 这个函数是libtorch中的max,返回c++中的tuple类型(第一个值为维度上最大值,第二个值为最大值的序号)所以使用std::get<1>(max_result)来取出序号,这是tuple类型取出方式。
模型转换libtorch不依赖于python,python训练的模型,需要转换为script model才能由libtorch加载,并进行推理。在这一步官网提供了两种方法:
方法一:Tracing
这种方法操作比较简单,只需要给模型一组输入,走一遍推理网络,然后由torch.ji.trace记录一下路径上的信息并保存即可。示例如下:
import torch
import torchvision
# An instance of your model.
model = torchvision.models.resnet18()
# An example input you would normally provide to your model's forward() method.
example = torch.rand(1, 3, 224, 224)
# Use torch.jit.trace to generate a torch.jit.ScriptModule via tracing.
traced_script_module = torch.jit.trace(model, example)
缺点是如果模型中存在控制流比如if-else语句,一组输入只能遍历一个分支,这种情况下就没办法完整的把模型信息记录下来。
意思就是说这个方法只能转换一波流的网络
方法二:Scripting
直接在Torch脚本中编写模型并相应地注释模型,通过torch.jit.script编译模块,将其转换为ScriptModule。示例如下:
class MyModule(torch.nn.Module):
def __init__(self, N, M):
super(MyModule, self).__init__()
self.weight = torch.nn.Parameter(torch.rand(N, M))
def forward(self, input):
if input.sum() > 0:
output = self.weight.mv(input)
else:
output = self.weight + input
return output
my_module = MyModule(10,20)
sm = torch.jit.script(my_module)
forward方法会被默认编译,forward中被调用的方法也会按照被调用的顺序被编译
如果想要编译一个forward以外且未被forward调用的方法,可以添加 @torch.jit.export.
如果想要方法不被编译,可使用@torch.jit.ignore 或者 @torch.jit.unused
##例子如下:
# Same behavior as pre-PyTorch 1.2
@torch.jit.script
def some_fn():
return 2
# Marks a function as ignored, if nothing
# ever calls it then this has no effect
@torch.jit.ignore
def some_fn2():
return 2
# As with ignore, if nothing calls it then it has no effect.
# If it is called in script it is replaced with an exception.
@torch.jit.unused
def some_fn3():
import pdb; pdb.set_trace()
return 4
# Doesn't do anything, this function is already
# the main entry point
@torch.jit.export
def some_fn4():
return 2
不是所有方法都支持的
https://pytorch.org/docs/master/jit_unsupported.html#jit-unsupported
1. 不支持的操作
TorchScript支持的操作是python的子集,大部分torch中用到的操作都可以找到对应实现,但也存在一些尴尬的不支持操作,详细列表可见unsupported-ops,下面列一些我自己遇到的操作:
1)参数/返回值不支持可变个数,例如
def __init__(self, **kwargs):
或者
if output_flag == 0:
return reshape_logits
else:
loss = self.loss(reshape_logits, term_mask, labels_id)
return reshape_logits, loss
2)各种iteration操作
eg1.
layers = [int(a) for a in layers]
报错torch.jit.frontend.UnsupportedNodeError: ListComp aren’t supported
可以改成:
for k in range(len(layers)):
layers[k] = int(layers[k])
eg2.
seq_iter = enumerate(scores)
try:
_, inivalues = seq_iter.__next__()
except:
_, inivalues = seq_iter.next()
eg3.
line = next(infile)
3)不支持的语句
eg1. 不支持continue
torch.jit.frontend.UnsupportedNodeError: continue statements aren’t supported
eg2. 不支持try-catch
torch.jit.frontend.UnsupportedNodeError: try blocks aren’t supported
eg3. 不支持with语句
4)其他常见op/module
eg1. torch.autograd.Variable
解决:使用torch.ones/torch.randn等初始化+.float()/.long()等指定数据类型。
eg2. torch.Tensor/torch.LongTensor etc.
解决:同上
eg3. requires_grad参数只在torch.tensor中支持,torch.ones/torch.zeros等不可用
eg4. tensor.numpy()
eg5. tensor.bool()
解决:tensor.bool()用tensor>0代替
eg6. self.seg_emb(seg_fea_ids).to(embeds.device)
解决:需要转gpu的地方显示调用.cuda()
总之一句话:除了原生python和pytorch以外的库,比如numpy什么的能不用就不用,尽量用pytorch的各种API。
2. 指定数据类型
1)属性,大部分的成员数据类型可以根据值来推断,空的列表/字典则需要预先指定
from typing import Dict
class MyModule(torch.nn.Module):
my_dict: Dict[str, int]
def __init__(self):
super(MyModule, self).__init__()
# This type cannot be inferred and must be specified
self.my_dict = {}
# The attribute type here is inferred to be `int`
self.my_int = 20
def forward(self):
pass
m = torch.jit.script(MyModule())
2)常量,使用Final关键字
try:
from typing_extensions import Final
except:
# If you don't have `typing_extensions` installed, you can use a
# polyfill from `torch.jit`.
from torch.jit import Final
class MyModule(torch.nn.Module):
my_constant: Final[int]
def __init__(self):
super(MyModule, self).__init__()
self.my_constant = 2
def forward(self):
pass
m = torch.jit.script(MyModule())
3)变量。默认是tensor类型且不可变,所以非tensor类型必须要指明
def forward(self, batch_size:int, seq_len:int, use_cuda:bool):
#方法三:Tracing and Scriptin混合
1)属性,大部分的成员数据类型可以根据值来推断,空的列表/字典则需要预先指定
from typing import Dict
class MyModule(torch.nn.Module):
my_dict: Dict[str, int]
def __init__(self):
super(MyModule, self).__init__()
# This type cannot be inferred and must be specified
self.my_dict = {}
# The attribute type here is inferred to be `int`
self.my_int = 20
def forward(self):
pass
m = torch.jit.script(MyModule())
2)常量,使用Final关键字
try:
from typing_extensions import Final
except:
# If you don't have `typing_extensions` installed, you can use a
# polyfill from `torch.jit`.
from torch.jit import Final
class MyModule(torch.nn.Module):
my_constant: Final[int]
def __init__(self):
super(MyModule, self).__init__()
self.my_constant = 2
def forward(self):
pass
m = torch.jit.script(MyModule())
3)变量。默认是tensor类型且不可变,所以非tensor类型必须要指明
def forward(self, batch_size:int, seq_len:int, use_cuda:bool):
方法三:Tracing and Scriptin混合
一种是在trace模型中调用script,适合模型中只有一小部分需要用到控制流的情况,使用实例如下:
import torch
@torch.jit.script
def foo(x, y):
if x.max() > y.max():
r = x
else:
r = y
return r
def bar(x, y, z):
return foo(x, y) + z
traced_bar = torch.jit.trace(bar, (torch.rand(3), torch.rand(3), torch.rand(3)))
另一种情况是在script module中用tracing生成子模块,对于一些存在script module不支持的python feature的layer,就可以把相关layer封装起来,用trace记录相关layer流,其他layer不用修改。使用示例如下:
import torch
import torchvision
class MyScriptModule(torch.nn.Module):
def __init__(self):
super(MyScriptModule, self).__init__()
self.means = torch.nn.Parameter(torch.tensor([103.939, 116.779, 123.68])
.resize_(1, 3, 1, 1))
self.resnet = torch.jit.trace(torchvision.models.resnet18(),
torch.rand(1, 3, 224, 224))
def forward(self, input):
return self.resnet(input - self.means)
my_script_module = torch.jit.script(MyScriptModule())
2.保存序列化模型
如果上一步的坑都踩完,那么模型保存就非常简单了,只需要调用save并传递一个文件名即可,需要注意的是如果想要在gpu上训练模型,在cpu上做inference,一定要在模型save之前转化,再就是记得调用model.eval(),形如
gpu_model.eval()
cpu_model = gpu_model.cpu()
sample_input_cpu = sample_input_gpu.cpu()
traced_cpu = torch.jit.trace(traced_cpu, sample_input_cpu)
torch.jit.save(traced_cpu, "cpu.pth")
traced_gpu = torch.jit.trace(traced_gpu, sample_input_gpu)
torch.jit.save(traced_gpu, "gpu.pth")
3.C++ load训练好的模型
要在C ++中加载序列化的PyTorch模型,必须依赖于PyTorch C ++ API(也称为LibTorch)。libtorch的安装非常简单,只需要在pytorch官网下载对应版本,解压即可。会得到一个结构如下的文件夹。
libtorch/
bin/
include/
lib/
share/
然后就可以构建应用程序了,一个简单的示例目录结构如下:
example-app/
CMakeLists.txt
example-app.cpp
example-app.cpp和CMakeLists.txt的示例代码分别如下:
#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";
}
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(custom_ops)
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)
至此,就可以运行以下命令从example-app/文件夹中构建应用程序啦:
mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=/path/to/libtorch ..
cmake --build . --config Release
其中/path/to/libtorch是之前下载后的libtorch文件夹所在的路径。这一步如果顺利能够看到编译完成100%的提示,下一步运行编译生成的可执行文件,会看到“ok”的输出,可喜可贺!
4. 执行Script Module
终于到最后一步啦!下面只需要按照构建输入传给模型,执行forward就可以得到输出啦。一个简单的示例如下:
// Create a vector of inputs.
std::vector<torch::jit::IValue> inputs;
inputs.push_back(torch::ones({1, 3, 224, 224}));
// 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';
前两行创建一个torch::jit::IValue的向量,并添加单个输入. 使用torch::ones()创建输入张量,等效于C ++ API中的torch.ones。 然后,运行script::Module的forward方法,通过调用toTensor()将返回的IValue值转换为张量。C++对torch的各种操作还是比较友好的,通过torch::或者后加_的方法都可以找到对应实现,例如
torch::tensor(input_list[j]).to(at::kLong).resize_({batch, 128}).clone()
//torch::tensor对应pytorch的torch.tensor; at::kLong对应torch.int64;resize_对应resize
最后check一下确保c++端的输出和pytorch是一致的就大功告成啦~
踩了无数坑,薅掉了无数头发,很多东西也是自己一点点摸索的,如果有错误欢迎指正!
参考资料:
PyTorch C++ API - PyTorch master document
Torch Script - PyTorch master documentation
#实际例子
在python代码中直接将tensor保存为pt文件,然后使用如下代码将其转化为c++可以读取的格式:
class Container(torch.nn.Module):
def __init__(self, my_values):
super().__init__()
for key in my_values:
setattr(self, key, my_values[key])
my_values = {
'img': torch.load('img.pt'),
'proposal_list': torch.load('proposal_list.pt'),
'cls_score': torch.load('cls_score.pt'),
'bbox_pred': torch.load('bbox_pred.pt'),
'det_bboxes': torch.load('det_bboxes.pt'),
'det_labels': torch.load('det_labels.pt'),
}
Save arbitrary values supported by TorchScript
https://pytorch.org/docs/master/jit.html#supported-type
container = torch.jit.script(Container(my_values))
container.save("results.pt")
在c++使用如下代码加载results.pt文件并与c++结果比对:
torch::jit::script::Module results = torch::jit::load("results.pt");
Tensor std_img = results.attr("img").toTensor().to(device);
//auto feature_maps = results.attr("img").toTuple();
//auto cls_out = results.attr("img").toTensorVector();
std::cout << torch::sum(torch::abs(std_img - img)) << std::endl;
参考
https://zhuanlan.zhihu.com/p/263626686
https://zhuanlan.zhihu.com/p/146453159
https://zhuanlan.zhihu.com/p/72750321