Skip to content

Latest commit

 

History

History
729 lines (544 loc) · 25.2 KB

README.md

File metadata and controls

729 lines (544 loc) · 25.2 KB
Raw

Read this in other languages: English, 简体中文.

News:

Tutorial for exporting CenterNet from pytorch to tensorRT is released.

Tutorial Video

An Out-of-the-Box TensorRT-based Framework for High Performance Inference with C++/Python Support

  • C++ Interface: 3 lines of code is all you need to run a YoloX

    // create inference engine on gpu-0
//auto engine = Yolo::create_infer("yolov5m.fp32.trtmodel", Yolo::Type::V5, 0);
auto engine = Yolo::create_infer("yolox_m.fp32.trtmodel", Yolo::Type::X, 0);

// load image
auto image = cv::imread("1.jpg");

// do inference and get the result
auto box = engine->commit(image).get();  // return vector<Box>

  • Python Interface:

    import trtpy

model     = models.resnet18(True).eval().to(device)
trt_model = tp.from_torch(model, input)
trt_out   = trt_model(input)

INTRO

  1. High level interface for C++/Python.
  2. Based on TensorRT8.0.
  3. Simplify the implementation of custom plugin. And serialization and deserialization have been encapsulated for easier usage.
  4. Simplify the compile of fp32, fp16 and int8 for facilitating the deployment with C++/Python in server or embeded device.
  5. Models ready for use also with examples are RetinaFace, Scrfd, YoloV5, YoloX, Arcface, AlphaPose, CenterNet and DeepSORT(C++)

YoloX and YoloV5-series Model Test Report

app_yolo.cpp speed testing
  1. Resolution (YoloV5P5, YoloX) = (640x640), (YoloV5P6) = (1280x1280)
  2. max batch size = 16
  3. preprocessing + inference + postprocessing
  4. cuda10.2, cudnn8.2.2.26, TensorRT-8.0.1.6
  5. RTX2080Ti
  6. num of testing: take the average on the results of 100 times but excluding the first time for warmup
  7. Testing log: [workspace/perf.result.std.log (workspace/perf.result.std.log)
  8. code for testing: src/application/app_yolo.cpp
  9. images for testing: 6 images in workspace/inference
    • with resolution 810x1080，500x806，1024x684，550x676，1280x720，800x533 respetively
  10. Testing method: load 6 images. Then do the inference on the 6 images, which will be repeated for 100 times. Note that each image should be preprocessed and postprocessed.
Model Resolution Type Precision Elapsed Time FPS
yolox_x 640x640 YoloX FP32 21.879 45.71
yolox_l 640x640 YoloX FP32 12.308 81.25
yolox_m 640x640 YoloX FP32 6.862 145.72
yolox_s 640x640 YoloX FP32 3.088 323.81
yolox_x 640x640 YoloX FP16 6.763 147.86
yolox_l 640x640 YoloX FP16 3.933 254.25
yolox_m 640x640 YoloX FP16 2.515 397.55
yolox_s 640x640 YoloX FP16 1.362 734.48
yolox_x 640x640 YoloX INT8 4.070 245.68
yolox_l 640x640 YoloX INT8 2.444 409.21
yolox_m 640x640 YoloX INT8 1.730 577.98
yolox_s 640x640 YoloX INT8 1.060 943.15
yolov5x6 1280x1280 YoloV5_P6 FP32 68.022 14.70
yolov5l6 1280x1280 YoloV5_P6 FP32 37.931 26.36
yolov5m6 1280x1280 YoloV5_P6 FP32 20.127 49.69
yolov5s6 1280x1280 YoloV5_P6 FP32 8.715 114.75
yolov5x 640x640 YoloV5_P5 FP32 18.480 54.11
yolov5l 640x640 YoloV5_P5 FP32 10.110 98.91
yolov5m 640x640 YoloV5_P5 FP32 5.639 177.33
yolov5s 640x640 YoloV5_P5 FP32 2.578 387.92
yolov5x6 1280x1280 YoloV5_P6 FP16 20.877 47.90
yolov5l6 1280x1280 YoloV5_P6 FP16 10.960 91.24
yolov5m6 1280x1280 YoloV5_P6 FP16 7.236 138.20
yolov5s6 1280x1280 YoloV5_P6 FP16 3.851 259.68
yolov5x 640x640 YoloV5_P5 FP16 5.933 168.55
yolov5l 640x640 YoloV5_P5 FP16 3.450 289.86
yolov5m 640x640 YoloV5_P5 FP16 2.184 457.90
yolov5s 640x640 YoloV5_P5 FP16 1.307 765.10
yolov5x6 1280x1280 YoloV5_P6 INT8 12.207 81.92
yolov5l6 1280x1280 YoloV5_P6 INT8 7.221 138.49
yolov5m6 1280x1280 YoloV5_P6 INT8 5.248 190.55
yolov5s6 1280x1280 YoloV5_P6 INT8 3.149 317.54
yolov5x 640x640 YoloV5_P5 INT8 3.704 269.97
yolov5l 640x640 YoloV5_P5 INT8 2.255 443.53
yolov5m 640x640 YoloV5_P5 INT8 1.674 597.40
yolov5s 640x640 YoloV5_P5 INT8 1.143 874.91
app_yolo_fast.cpp speed testing. Never stop desiring for being faster
  • Highlight: 0.5 ms faster without any loss in precision compared with the above. Specifically, we remove the Focus and some transpose nodes etc, and implement them in CUDA kenerl function. But the rest remains the same.
  • Test log: workspace/perf.result.std.log
  • Code for testing: src/application/app_yolo_fast.cpp
  • Tips: you can do the modification while refering to the downloaded onnx. Any questions are welcomed through any kinds of contact.
  • Conclusion: the main idea of this work is to optimize the pre-and-post processing. If you go for yolox, yolov5 small version, the optimization might help you.
Model Resolution Type Precision Elapsed Time FPS
yolox_x_fast 640x640 YoloX FP32 21.598 46.30
yolox_l_fast 640x640 YoloX FP32 12.199 81.97
yolox_m_fast 640x640 YoloX FP32 6.819 146.65
yolox_s_fast 640x640 YoloX FP32 2.979 335.73
yolox_x_fast 640x640 YoloX FP16 6.764 147.84
yolox_l_fast 640x640 YoloX FP16 3.866 258.64
yolox_m_fast 640x640 YoloX FP16 2.386 419.16
yolox_s_fast 640x640 YoloX FP16 1.259 794.36
yolox_x_fast 640x640 YoloX INT8 3.918 255.26
yolox_l_fast 640x640 YoloX INT8 2.292 436.38
yolox_m_fast 640x640 YoloX INT8 1.589 629.49
yolox_s_fast 640x640 YoloX INT8 0.954 1048.47
yolov5x6_fast 1280x1280 YoloV5_P6 FP32 67.075 14.91
yolov5l6_fast 1280x1280 YoloV5_P6 FP32 37.491 26.67
yolov5m6_fast 1280x1280 YoloV5_P6 FP32 19.422 51.49
yolov5s6_fast 1280x1280 YoloV5_P6 FP32 7.900 126.57
yolov5x_fast 640x640 YoloV5_P5 FP32 18.554 53.90
yolov5l_fast 640x640 YoloV5_P5 FP32 10.060 99.41
yolov5m_fast 640x640 YoloV5_P5 FP32 5.500 181.82
yolov5s_fast 640x640 YoloV5_P5 FP32 2.342 427.07
yolov5x6_fast 1280x1280 YoloV5_P6 FP16 20.538 48.69
yolov5l6_fast 1280x1280 YoloV5_P6 FP16 10.404 96.12
yolov5m6_fast 1280x1280 YoloV5_P6 FP16 6.577 152.06
yolov5s6_fast 1280x1280 YoloV5_P6 FP16 3.087 323.99
yolov5x_fast 640x640 YoloV5_P5 FP16 5.919 168.95
yolov5l_fast 640x640 YoloV5_P5 FP16 3.348 298.69
yolov5m_fast 640x640 YoloV5_P5 FP16 2.015 496.34
yolov5s_fast 640x640 YoloV5_P5 FP16 1.087 919.63
yolov5x6_fast 1280x1280 YoloV5_P6 INT8 11.236 89.00
yolov5l6_fast 1280x1280 YoloV5_P6 INT8 6.235 160.38
yolov5m6_fast 1280x1280 YoloV5_P6 INT8 4.311 231.97
yolov5s6_fast 1280x1280 YoloV5_P6 INT8 2.139 467.45
yolov5x_fast 640x640 YoloV5_P5 INT8 3.456 289.37
yolov5l_fast 640x640 YoloV5_P5 INT8 2.019 495.41
yolov5m_fast 640x640 YoloV5_P5 INT8 1.425 701.71
yolov5s_fast 640x640 YoloV5_P5 INT8 0.844 1185.47

Setup and Configuration

Linux
  1. VSCode (highly recommended!)
  2. Configure your path for cudnn, cuda, tensorRT8.0 and protobuf.
  3. Configure the compute capability matched with your nvidia graphics card in Makefile/CMakeLists.txt
  4. Configure your library path in .vscode/c_cpp_properties.json
  5. CUDA version: CUDA10.2
  6. CUDNN version: cudnn8.2.2.26. Note that dev(.h file) and runtime(.so file) should be downloaded.
  7. tensorRT version：tensorRT-8.0.1.6-cuda10.2
  8. protobuf version（for onnx parser）：protobufv3.11.4
  • CMake:
    • mkdir build && cd build
    • cmake ..
    • make yolo -j8
  • Makefile:
    • make yolo -j8
Linux: Compile for Python
  • compile and install
    • Makefile：
      • set use_python := true in Makefile
    • CMakeLists.txt:
      • set(HAS_PYTHON ON) in CMakeLists.txt
    • Type in make pyinstall -j8
    • Complied files are in python/trtpy/libtrtpyc.so
Windows

  1. Please check the lean/README.md for the detailed dependency

  2. In TensorRT.vcxproj, replace the <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.0.props" /> with your own CUDA path

  3. In TensorRT.vcxproj, replace the <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 10.0.targets" /> with your own CUDA path

  4. In TensorRT.vcxproj, replace the <CodeGeneration>compute_61,sm_61</CodeGeneration> with your compute capability.

  5. Configure your dependency or download it to the foler /lean. Configure VC++ dir (include dir and refence)

  6. Configure your env, debug->environment

  7. Compile and run the example, where 3 options are available.

Windows: Compile for Python
  1. Compile trtpyc.pyd. Choose python in visual studio to compile
  2. Copy dll and execute 'python/copy_dll_to_trtpy.bat'
  3. Execute the example in python dir by 'python test_yolov5.py'
  • if installation is needed, switch to target env(e.g. your conda env) then 'python setup.py install', which has to be followed by step 1 and step 2.
  • the compiled files are in python/trtpy/libtrtpyc.pyd
Other Protobuf Version
  • in onnx/make_pb.sh, replace the path protoc=/data/sxai/lean/protobuf3.11.4/bin/protoc in protoc with the protoc of your own version
#cd the path in terminal to /onnx
cd onnx

#execuete the command to make pb files
bash make_pb.sh
  • CMake:
    • replace the set(PROTOBUF_DIR "/data/sxai/lean/protobuf3.11.4") in CMakeLists.txt with the same path of your protoc.
mkdir build && cd build
cmake ..
make yolo -j64
  • Makefile:
    • replace the path lean_protobuf := /data/sxai/lean/protobuf3.11.4 in Makefile with the same path of protoc
make yolo -j64
TensorRT 7.x support
  • The default is tensorRT8.x
  1. Replace onnx_parser_for_7.x/onnx_parser to src/tensorRT/onnx_parser
    • bash onnx_parser/use_tensorrt_7.x.sh
  2. Configure Makefile/CMakeLists.txt path to TensorRT7.x
  3. Execute make yolo -j64
TensorRT 8.x support
  • The default is tensorRT8.x
  1. Replace onnx_parser_for_8.x/onnx_parser to src/tensorRT/onnx_parser
    • bash onnx_parser/use_tensorrt_8.x.sh
  2. Configure Makefile/CMakeLists.txt path to TensorRT8.x
  3. Execute make yolo -j64

Guide for Different Tasks/Model Support

YoloV5 Support
  • if pytorch >= 1.7, and the model is 5.0+, the model is suppored by the framework
  • if pytorch < 1.7 or yolov5(2.0, 3.0 or 4.0), minor modification should be done in opset.
  • if you want to achieve the inference with lower pytorch, dynamic batchsize and other advanced setting, please check our blog (http://zifuture.com:8090)(now in Chinese) and scan the QRcode via Wechat to join us.
  1. Download yolov5
git clone [email protected]:ultralytics/yolov5.git
  1. Modify the code for dynamic batchsize
# line 55 forward function in yolov5/models/yolo.py 
# bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
# x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
# modified into:

bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
bs = -1
ny = int(ny)
nx = int(nx)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()

# line 70 in yolov5/models/yolo.py
#  z.append(y.view(bs, -1, self.no))
# modified into：
z.append(y.view(bs, self.na * ny * nx, self.no))

# line 52 in yolov5/export.py
# torch.onnx.export(dynamic_axes={'images': {0: 'batch', 2: 'height', 3: 'width'},  # shape(1,3,640,640)
#                                'output': {0: 'batch', 1: 'anchors'}  # shape(1,25200,85)  修改为
# modified into:
torch.onnx.export(dynamic_axes={'images': {0: 'batch'},  # shape(1,3,640,640)
                                'output': {0: 'batch'}  # shape(1,25200,85)
  1. Export to onnx model
cd yolov5
python export.py --weights=yolov5s.pt --dynamic --include=onnx --opset=11
  1. Copy the model and execute it
cp yolov5/yolov5s.onnx tensorRT_cpp/workspace/
cd tensorRT_cpp
make yolo -j32
YoloX Support
  1. Download YoloX
git clone [email protected]:Megvii-BaseDetection/YOLOX.git
cd YOLOX
  1. Modify the code The modification ensures a successful int8 compilation and inference, otherwise Missing scale and zero-point for tensor (Unnamed Layer* 686) will be raised.
# line 206 forward fuction in yolox/models/yolo_head.py. Replace the commented code with the uncommented code
# self.hw = [x.shape[-2:] for x in outputs] 
self.hw = [list(map(int, x.shape[-2:])) for x in outputs]


# line 208 forward function in yolox/models/yolo_head.py. Replace the commented code with the uncommented code
# [batch, n_anchors_all, 85]
# outputs = torch.cat(
#     [x.flatten(start_dim=2) for x in outputs], dim=2
# ).permute(0, 2, 1)
proc_view = lambda x: x.view(-1, int(x.size(1)), int(x.size(2) * x.size(3)))
outputs = torch.cat(
    [proc_view(x) for x in outputs], dim=2
).permute(0, 2, 1)


# line 253 decode_output function in yolox/models/yolo_head.py Replace the commented code with the uncommented code
#outputs[..., :2] = (outputs[..., :2] + grids) * strides
#outputs[..., 2:4] = torch.exp(outputs[..., 2:4]) * strides
#return outputs
xy = (outputs[..., :2] + grids) * strides
wh = torch.exp(outputs[..., 2:4]) * strides
return torch.cat((xy, wh, outputs[..., 4:]), dim=-1)

# line 77 in tools/export_onnx.py
model.head.decode_in_inference = True
  1. Export to onnx
# download model
wget https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_m.pth

# export
python tools/export_onnx.py -c yolox_m.pth -f exps/default/yolox_m.py --output-name=yolox_m.onnx --dynamic --no-onnxsim
  1. Execute the command
cp YOLOX/yolox_m.onnx tensorRT_cpp/workspace/
cd tensorRT_cpp
make yolo -j32
Retinaface Support
  1. Download Pytorch_Retinaface Repo
git clone [email protected]:biubug6/Pytorch_Retinaface.git
cd Pytorch_Retinaface

  1. Download model from the Training of README.md in https://github.com/biubug6/Pytorch_Retinaface#training .Then unzip it to the /weights . Here, we use mobilenet0.25_Final.pth

  2. Modify the code

# line 24 in models/retinaface.py
# return out.view(out.shape[0], -1, 2) is modified into 
return out.view(-1, int(out.size(1) * out.size(2) * 2), 2)

# line 35 in models/retinaface.py
# return out.view(out.shape[0], -1, 4) is modified into
return out.view(-1, int(out.size(1) * out.size(2) * 2), 4)

# line 46 in models/retinaface.py
# return out.view(out.shape[0], -1, 10) is modified into
return out.view(-1, int(out.size(1) * out.size(2) * 2), 10)

# The following modification ensures the output of resize node is based on scale rather than shape such that dynamic batch can be achieved.
# line 89 in models/net.py
# up3 = F.interpolate(output3, size=[output2.size(2), output2.size(3)], mode="nearest") is modified into
up3 = F.interpolate(output3, scale_factor=2, mode="nearest")

# line 93 in models/net.py
# up2 = F.interpolate(output2, size=[output1.size(2), output1.size(3)], mode="nearest") is modified into
up2 = F.interpolate(output2, scale_factor=2, mode="nearest")

# The following code removes softmax (bug sometimes happens). At the same time, concatenate the output to simplify the decoding.
# line 123 in models/retinaface.py
# if self.phase == 'train':
#     output = (bbox_regressions, classifications, ldm_regressions)
# else:
#     output = (bbox_regressions, F.softmax(classifications, dim=-1), ldm_regressions)
# return output
# the above is modified into:
output = (bbox_regressions, classifications, ldm_regressions)
return torch.cat(output, dim=-1)

# set 'opset_version=11' to ensure a successful export
# torch_out = torch.onnx._export(net, inputs, output_onnx, export_params=True, verbose=False,
#     input_names=input_names, output_names=output_names)
# is modified into:
torch_out = torch.onnx._export(net, inputs, output_onnx, export_params=True, verbose=False, opset_version=11,
    input_names=input_names, output_names=output_names)
  1. Export to onnx
python convert_to_onnx.py
  1. Execute
cp FaceDetector.onnx ../tensorRT_cpp/workspace/mb_retinaface.onnx
cd ../tensorRT_cpp
make retinaface -j64
DBFace Support 
make dbface -j64
Scrfd Support
Arcface Support 
auto arcface = Arcface::create_infer("arcface_iresnet50.fp32.trtmodel", 0);
auto feature = arcface->commit(make_tuple(face, landmarks)).get();
cout << feature << endl;  // 1x512
  • In the example of Face Recognition, workspace/face/library is the set of faces registered.
  • workspace/face/recognize is the set of face to be recognized.
  • the result is saved in workspace/face/result和workspace/face/library_draw
CenterNet Support

check the great details in tutorial/2.0

Bert Support(Chinese Classification)

the INTRO to Interface

Python Interface：Get onnx and trtmodel from pytorch model more easily
  • Just one line of code to export onnx and trtmodel. And save them for usage in the future.
import trtpy

model = models.resnet18(True).eval()
trtpy.from_torch(
    model, 
    dummy_input, 
    max_batch_size=16, 
    onnx_save_file="test.onnx", 
    engine_save_file="engine.trtmodel"
)
Python Interface：TensorRT Inference
  • YoloX TensorRT Inference
import trtpy

yolo   = tp.Yolo(engine_file, type=tp.YoloType.X)   # engine_file is the trtmodel file
image  = cv2.imread("inference/car.jpg")
bboxes = yolo.commit(image).get()
  • Seamless Inference from Pytorch to TensorRT
import trtpy

model     = models.resnet18(True).eval().to(device) # pt model
trt_model = tp.from_torch(model, input)
trt_out   = trt_model(input)
C++ Interface：YoloX Inference 
// create infer engine on gpu 0
auto engine = Yolo::create_infer("yolox_m.fp32.trtmodel"， Yolo::Type::X, 0);

// load image
auto image = cv::imread("1.jpg");

// do inference and get the result
auto box = engine->commit(image).get();
C++ Interface：Comple Model in FP32/FP16 
TRT::compile(
  TRT::Mode::FP32,   // compile model in fp32
  3,                          // max batch size
  "plugin.onnx",              // onnx file
  "plugin.fp32.trtmodel",     // save path
  {}                         //  redefine the shape of input when needed
);
  • For fp32 compilation, all you need is offering onnx file whose input shape is allowed to be redefined.
C++ Interface：Compile in int8
  • The in8 inference performs slightly worse than fp32 in precision(about -5% drop down), but stunningly faster. In the framework, we offer int8 inference
// define int8 calibration function to read data and handle it to tenor.
auto int8process = [](int current, int count, vector<string>& images, shared_ptr<TRT::Tensor>& tensor){
    for(int i = 0; i < images.size(); ++i){
    // int8 compilation requires calibration. We read image data and set_norm_mat. Then the data will be transfered into the tensor.
        auto image = cv::imread(images[i]);
        cv::resize(image, image, cv::Size(640, 640));
        float mean[] = {0, 0, 0};
        float std[]  = {1, 1, 1};
        tensor->set_norm_mat(i, image, mean, std);
    }
};


// Specify TRT::Mode as INT8
auto model_file = "yolov5m.int8.trtmodel";
TRT::compile(
  TRT::Mode::INT8,            // INT8
  3,                          // max batch size
  "yolov5m.onnx",             // onnx
  model_file,                 // saved filename
  {},                         // redefine the input shape
  int8process,                // the recall function for calibration
  ".",                        // the dir where the image data is used for calibration
  ""                          // the dir where the data generated from calibration is saved(a.k.a where to load the calibration data.)
);
  • We integrate into only one int8process function to save otherwise a lot of issues that might happen in tensorRT official implementation.
C++ Interface：Inference

  • We introduce class Tensor for easier inference and data transfer between host to device. So that as a user, the details wouldn't be annoying.

  • class Engine is another facilitator.

// load model and get a shared_ptr. get nullptr if fail to load.
auto engine = TRT::load_infer("yolov5m.fp32.trtmodel");

// print model info
engine->print();

// load image
auto image = imread("demo.jpg");

// get the model input and output node, which can be accessed by name or index
auto input = engine->input(0);   // or auto input = engine->input("images");
auto output = engine->output(0); // or auto output = engine->output("output");

// put the image into input tensor by calling set_norm_mat()
float mean[] = {0, 0, 0};
float std[]  = {1, 1, 1};
input->set_norm_mat(i, image, mean, std);

// do the inference. Here sync(true) or async(false) is optional
engine->forward(); // engine->forward(true or false)

// get the outut_ptr, which can used to access the output
float* output_ptr = output->cpu<float>();
C++ Interface：Plugin
  • You only need to define kernel function and inference process. The details of code(e.g the serialization, deserialization and injection of plugin etc) are under the hood.
  • Easy to implement a new plugin in FP32 and FP16. Refer to HSwish.cu for details.
template<>
__global__ void HSwishKernel(float* input, float* output, int edge) {

    KernelPositionBlock;
    float x = input[position];
    float a = x + 3;
    a = a < 0 ? 0 : (a >= 6 ? 6 : a);
    output[position] = x * a / 6;
}

int HSwish::enqueue(const std::vector<GTensor>& inputs, std::vector<GTensor>& outputs, const std::vector<GTensor>& weights, void* workspace, cudaStream_t stream) {

    int count = inputs[0].count();
    auto grid = CUDATools::grid_dims(count);
    auto block = CUDATools::block_dims(count);
    HSwishKernel <<<grid, block, 0, stream >>> (inputs[0].ptr<float>(), outputs[0].ptr<float>(), count);
    return 0;
}


RegisterPlugin(HSwish);

About Us