From 6161f288542a99d222b28c2cbb64d630cc70f85a Mon Sep 17 00:00:00 2001 From: YuzhouPeng <760783896@qq.com> Date: Thu, 28 Jan 2021 12:59:02 +0800 Subject: [PATCH] add unet code (#380) * add code * remove txt --- unet/CMakeLists.txt | 34 +++ unet/README.md | 64 ++++++ unet/common.hpp | 148 +++++++++++++ unet/gen_wts.py | 36 ++++ unet/logging.h | 503 ++++++++++++++++++++++++++++++++++++++++++++ unet/unet.cpp | 372 ++++++++++++++++++++++++++++++++ 6 files changed, 1157 insertions(+) create mode 100644 unet/CMakeLists.txt create mode 100644 unet/README.md create mode 100644 unet/common.hpp create mode 100644 unet/gen_wts.py create mode 100644 unet/logging.h create mode 100644 unet/unet.cpp diff --git a/unet/CMakeLists.txt b/unet/CMakeLists.txt new file mode 100644 index 0000000..568e4fc --- /dev/null +++ b/unet/CMakeLists.txt @@ -0,0 +1,34 @@ +cmake_minimum_required(VERSION 2.6) + +project(unet) + +add_definitions(-std=c++11) + +option(CUDA_USE_STATIC_CUDA_RUNTIME OFF) +set(CMAKE_CXX_STANDARD 11) +set(CMAKE_BUILD_TYPE Debug) + +set(CUDA_NVCC_PLAGS ${CUDA_NVCC_PLAGS};-std=c++11;-g;-G;-gencode;arch=compute_30;code=sm_30) + +# cuda directory +include_directories(${PROJECT_SOURCE_DIR}/include) +include_directories(/usr/local/cuda-10.2/targets/x86_64-linux/include) +link_directories(/usr/local/cuda-10.2/targets/x86_64-linux/lib) + +# tensorrt +include_directories(/home/sycv/workplace/pengyuzhou/TensorRT-7.0.0.11/targets/x86_64-linux-gnu/include) +link_directories(/home/sycv/workplace/pengyuzhou/TensorRT-7.0.0.11/targets/x86_64-linux-gnu/lib) + +set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Ofast -Wfatal-errors -D_MWAITXINTRIN_H_INCLUDED") + +# link library and add exec file +add_executable(unet ${PROJECT_SOURCE_DIR}/unet.cpp) +target_link_libraries(unet nvinfer) +target_link_libraries(unet cudart) + +add_definitions(-O2 -pthread) + +# opencv library +find_package(OpenCV) +include_directories(OpenCV_INCLUDE_DIRS) +target_link_libraries(unet ${OpenCV_LIBS}) \ No newline at end of file diff --git a/unet/README.md b/unet/README.md new file mode 100644 index 0000000..7cb3d5d --- /dev/null +++ b/unet/README.md @@ -0,0 +1,64 @@ +# tensorrt-unet +This is a TensorRT version Unet, inspired by [tensorrtx](https://github.com/wang-xinyu/tensorrtx) and [pytorch-unet](https://github.com/milesial/Pytorch-UNet).
+You can generate TensorRT engine file using this script and customize some params and network structure based on network you trained (FP32/16 precision, input size, different conv, activation function...)
+ +# requirements + +TensorRT 7.0 (you need to install tensorrt first)
+Cuda 10.2
+Python3.7
+opencv 4.4
+cmake 3.18
+# train .pth file and convert .wts + +## create env + +``` +pip install -r requirements.txt +``` + +## train .pth file + +train your dataset by following [pytorch-unet](https://github.com/milesial/Pytorch-UNet) and generate .pth file.
+ +## convert .wts + +run gen_wts from utils folder, and move it to project folder
+ +# generate engine file and infer + +## create build folder in project folder +``` +mkdir build +``` + +## make file, generate exec file +``` +cd build +cmake .. +make +``` + +## generate TensorRT engine file and infer image +``` +unet -s +``` +then a unet exec file will generated, you can use unet -d to infer files in a folder
+``` +unet -d ../samples +``` + +# efficiency +the speed of tensorRT engine is much faster + + pytorch | TensorRT FP32 | TensorRT FP16 + ---- | ----- | ------ + 816x672 | 816x672 | 816x672 + 58ms | 43ms (batchsize 8) | 14ms (batchsize 8) + + +# Further development + +1. add INT8 calibrator
+2. add custom plugin
+etc diff --git a/unet/common.hpp b/unet/common.hpp new file mode 100644 index 0000000..3121f58 --- /dev/null +++ b/unet/common.hpp @@ -0,0 +1,148 @@ +#ifndef UNET_COMMON_H_ +#define UNET_COMMON_H_ + +#include +#include +#include +#include +#include +#include +#include "NvInfer.h" + + +#define CHECK(status) \ + do\ + {\ + auto ret = (status);\ + if (ret != 0)\ + {\ + std::cerr << "Cuda failure: " << ret << std::endl;\ + abort();\ + }\ + } while (0) + +using namespace nvinfer1; + + + + + +// TensorRT weight files have a simple space delimited format: +// [type] [size] +std::map loadWeights(const std::string file) { + std::cout << "Loading weights: " << file << std::endl; + std::map weightMap; + + // Open weights file + std::ifstream input(file); + assert(input.is_open() && "Unable to load weight file. please check if the .wts file path is right!!!!!!"); + + // Read number of weight blobs + int32_t count; + input >> count; + assert(count > 0 && "Invalid weight map file."); + + while (count--) + { + Weights wt{DataType::kFLOAT, nullptr, 0}; + uint32_t size; + + // Read name and type of blob + std::string name; + input >> name >> std::dec >> size; + wt.type = DataType::kFLOAT; + + // Load blob + uint32_t* val = reinterpret_cast(malloc(sizeof(val) * size)); + for (uint32_t x = 0, y = size; x < y; ++x) + { + input >> std::hex >> val[x]; + } + wt.values = val; + + wt.count = size; + weightMap[name] = wt; + } + + return weightMap; +} + +IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map& weightMap, ITensor& input, std::string lname, float eps) { + float *gamma = (float*)weightMap[lname + ".weight"].values; + float *beta = (float*)weightMap[lname + ".bias"].values; + float *mean = (float*)weightMap[lname + ".running_mean"].values; + float *var = (float*)weightMap[lname + ".running_var"].values; + int len = weightMap[lname + ".running_var"].count; + + float *scval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + scval[i] = gamma[i] / sqrt(var[i] + eps); + } + Weights scale{DataType::kFLOAT, scval, len}; + + float *shval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps); + } + Weights shift{DataType::kFLOAT, shval, len}; + + float *pval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + pval[i] = 1.0; + } + Weights power{DataType::kFLOAT, pval, len}; + + weightMap[lname + ".scale"] = scale; + weightMap[lname + ".shift"] = shift; + weightMap[lname + ".power"] = power; + IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power); + assert(scale_1); + return scale_1; +} + + +ILayer* convBlock(INetworkDefinition *network, std::map& weightMap, ITensor& input, int outch, int ksize, int s, int g, std::string lname) { + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + int p = ksize / 2; + IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{ksize, ksize}, weightMap[lname + ".conv.weight"], emptywts); + assert(conv1); + conv1->setStrideNd(DimsHW{s, s}); + conv1->setPaddingNd(DimsHW{p, p}); + conv1->setNbGroups(g); + IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + ".bn", 1e-3); + + // hard_swish = x * hard_sigmoid + auto hsig = network->addActivation(*bn1->getOutput(0), ActivationType::kHARD_SIGMOID); + assert(hsig); + hsig->setAlpha(1.0 / 6.0); + hsig->setBeta(0.5); + auto ew = network->addElementWise(*bn1->getOutput(0), *hsig->getOutput(0), ElementWiseOperation::kPROD); + assert(ew); + return ew; +} + + + +int read_files_in_dir(const char *p_dir_name, std::vector &file_names) { + DIR *p_dir = opendir(p_dir_name); + if (p_dir == nullptr) { + return -1; + } + + struct dirent* p_file = nullptr; + while ((p_file = readdir(p_dir)) != nullptr) { + if (strcmp(p_file->d_name, ".") != 0 && + strcmp(p_file->d_name, "..") != 0) { + //std::string cur_file_name(p_dir_name); + //cur_file_name += "/"; + //cur_file_name += p_file->d_name; + std::string cur_file_name(p_file->d_name); + file_names.push_back(cur_file_name); + } + } + + closedir(p_dir); + return 0; +} + +#endif diff --git a/unet/gen_wts.py b/unet/gen_wts.py new file mode 100644 index 0000000..f5fc683 --- /dev/null +++ b/unet/gen_wts.py @@ -0,0 +1,36 @@ +import torch +from torch import nn +import torchvision +import os +import struct +from torchsummary import summary + +def main(): + print('cuda device count: ', torch.cuda.device_count()) + net = torch.load('ori_unet.pth') + net = net.to('cuda:0') + net = net.eval() + print('model: ', net) + #print('state dict: ', net.state_dict().keys()) + tmp = torch.ones(1, 3, 224, 224).to('cuda:0') + print('input: ', tmp) + out = net(tmp) + + print('output:', out) + + summary(net, (3, 224, 224)) + #return + f = open("unet.wts", 'w') + f.write("{}\n".format(len(net.state_dict().keys()))) + for k,v in net.state_dict().items(): + print('key: ', k) + print('value: ', v.shape) + vr = v.reshape(-1).cpu().numpy() + f.write("{} {}".format(k, len(vr))) + for vv in vr: + f.write(" ") + f.write(struct.pack(">f", float(vv)).hex()) + f.write("\n") + +if __name__ == '__main__': + main() diff --git a/unet/logging.h b/unet/logging.h new file mode 100644 index 0000000..7f291b5 --- /dev/null +++ b/unet/logging.h @@ -0,0 +1,503 @@ +/* + * Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef TENSORRT_LOGGING_H +#define TENSORRT_LOGGING_H + +#include "NvInferRuntimeCommon.h" +#include +#include +#include +#include +#include +#include +#include + +using Severity = nvinfer1::ILogger::Severity; + +class LogStreamConsumerBuffer : public std::stringbuf +{ +public: + LogStreamConsumerBuffer(std::ostream& stream, const std::string& prefix, bool shouldLog) + : mOutput(stream) + , mPrefix(prefix) + , mShouldLog(shouldLog) + { + } + + LogStreamConsumerBuffer(LogStreamConsumerBuffer&& other) + : mOutput(other.mOutput) + { + } + + ~LogStreamConsumerBuffer() + { + // std::streambuf::pbase() gives a pointer to the beginning of the buffered part of the output sequence + // std::streambuf::pptr() gives a pointer to the current position of the output sequence + // if the pointer to the beginning is not equal to the pointer to the current position, + // call putOutput() to log the output to the stream + if (pbase() != pptr()) + { + putOutput(); + } + } + + // synchronizes the stream buffer and returns 0 on success + // synchronizing the stream buffer consists of inserting the buffer contents into the stream, + // resetting the buffer and flushing the stream + virtual int sync() + { + putOutput(); + return 0; + } + + void putOutput() + { + if (mShouldLog) + { + // prepend timestamp + std::time_t timestamp = std::time(nullptr); + tm* tm_local = std::localtime(×tamp); + std::cout << "["; + std::cout << std::setw(2) << std::setfill('0') << 1 + tm_local->tm_mon << "/"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_mday << "/"; + std::cout << std::setw(4) << std::setfill('0') << 1900 + tm_local->tm_year << "-"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_hour << ":"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_min << ":"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_sec << "] "; + // std::stringbuf::str() gets the string contents of the buffer + // insert the buffer contents pre-appended by the appropriate prefix into the stream + mOutput << mPrefix << str(); + // set the buffer to empty + str(""); + // flush the stream + mOutput.flush(); + } + } + + void setShouldLog(bool shouldLog) + { + mShouldLog = shouldLog; + } + +private: + std::ostream& mOutput; + std::string mPrefix; + bool mShouldLog; +}; + +//! +//! \class LogStreamConsumerBase +//! \brief Convenience object used to initialize LogStreamConsumerBuffer before std::ostream in LogStreamConsumer +//! +class LogStreamConsumerBase +{ +public: + LogStreamConsumerBase(std::ostream& stream, const std::string& prefix, bool shouldLog) + : mBuffer(stream, prefix, shouldLog) + { + } + +protected: + LogStreamConsumerBuffer mBuffer; +}; + +//! +//! \class LogStreamConsumer +//! \brief Convenience object used to facilitate use of C++ stream syntax when logging messages. +//! Order of base classes is LogStreamConsumerBase and then std::ostream. +//! This is because the LogStreamConsumerBase class is used to initialize the LogStreamConsumerBuffer member field +//! in LogStreamConsumer and then the address of the buffer is passed to std::ostream. +//! This is necessary to prevent the address of an uninitialized buffer from being passed to std::ostream. +//! Please do not change the order of the parent classes. +//! +class LogStreamConsumer : protected LogStreamConsumerBase, public std::ostream +{ +public: + //! \brief Creates a LogStreamConsumer which logs messages with level severity. + //! Reportable severity determines if the messages are severe enough to be logged. + LogStreamConsumer(Severity reportableSeverity, Severity severity) + : LogStreamConsumerBase(severityOstream(severity), severityPrefix(severity), severity <= reportableSeverity) + , std::ostream(&mBuffer) // links the stream buffer with the stream + , mShouldLog(severity <= reportableSeverity) + , mSeverity(severity) + { + } + + LogStreamConsumer(LogStreamConsumer&& other) + : LogStreamConsumerBase(severityOstream(other.mSeverity), severityPrefix(other.mSeverity), other.mShouldLog) + , std::ostream(&mBuffer) // links the stream buffer with the stream + , mShouldLog(other.mShouldLog) + , mSeverity(other.mSeverity) + { + } + + void setReportableSeverity(Severity reportableSeverity) + { + mShouldLog = mSeverity <= reportableSeverity; + mBuffer.setShouldLog(mShouldLog); + } + +private: + static std::ostream& severityOstream(Severity severity) + { + return severity >= Severity::kINFO ? std::cout : std::cerr; + } + + static std::string severityPrefix(Severity severity) + { + switch (severity) + { + case Severity::kINTERNAL_ERROR: return "[F] "; + case Severity::kERROR: return "[E] "; + case Severity::kWARNING: return "[W] "; + case Severity::kINFO: return "[I] "; + case Severity::kVERBOSE: return "[V] "; + default: assert(0); return ""; + } + } + + bool mShouldLog; + Severity mSeverity; +}; + +//! \class Logger +//! +//! \brief Class which manages logging of TensorRT tools and samples +//! +//! \details This class provides a common interface for TensorRT tools and samples to log information to the console, +//! and supports logging two types of messages: +//! +//! - Debugging messages with an associated severity (info, warning, error, or internal error/fatal) +//! - Test pass/fail messages +//! +//! The advantage of having all samples use this class for logging as opposed to emitting directly to stdout/stderr is +//! that the logic for controlling the verbosity and formatting of sample output is centralized in one location. +//! +//! In the future, this class could be extended to support dumping test results to a file in some standard format +//! (for example, JUnit XML), and providing additional metadata (e.g. timing the duration of a test run). +//! +//! TODO: For backwards compatibility with existing samples, this class inherits directly from the nvinfer1::ILogger +//! interface, which is problematic since there isn't a clean separation between messages coming from the TensorRT +//! library and messages coming from the sample. +//! +//! In the future (once all samples are updated to use Logger::getTRTLogger() to access the ILogger) we can refactor the +//! class to eliminate the inheritance and instead make the nvinfer1::ILogger implementation a member of the Logger +//! object. + +class Logger : public nvinfer1::ILogger +{ +public: + Logger(Severity severity = Severity::kWARNING) + : mReportableSeverity(severity) + { + } + + //! + //! \enum TestResult + //! \brief Represents the state of a given test + //! + enum class TestResult + { + kRUNNING, //!< The test is running + kPASSED, //!< The test passed + kFAILED, //!< The test failed + kWAIVED //!< The test was waived + }; + + //! + //! \brief Forward-compatible method for retrieving the nvinfer::ILogger associated with this Logger + //! \return The nvinfer1::ILogger associated with this Logger + //! + //! TODO Once all samples are updated to use this method to register the logger with TensorRT, + //! we can eliminate the inheritance of Logger from ILogger + //! + nvinfer1::ILogger& getTRTLogger() + { + return *this; + } + + //! + //! \brief Implementation of the nvinfer1::ILogger::log() virtual method + //! + //! Note samples should not be calling this function directly; it will eventually go away once we eliminate the + //! inheritance from nvinfer1::ILogger + //! + void log(Severity severity, const char* msg) override + { + LogStreamConsumer(mReportableSeverity, severity) << "[TRT] " << std::string(msg) << std::endl; + } + + //! + //! \brief Method for controlling the verbosity of logging output + //! + //! \param severity The logger will only emit messages that have severity of this level or higher. + //! + void setReportableSeverity(Severity severity) + { + mReportableSeverity = severity; + } + + //! + //! \brief Opaque handle that holds logging information for a particular test + //! + //! This object is an opaque handle to information used by the Logger to print test results. + //! The sample must call Logger::defineTest() in order to obtain a TestAtom that can be used + //! with Logger::reportTest{Start,End}(). + //! + class TestAtom + { + public: + TestAtom(TestAtom&&) = default; + + private: + friend class Logger; + + TestAtom(bool started, const std::string& name, const std::string& cmdline) + : mStarted(started) + , mName(name) + , mCmdline(cmdline) + { + } + + bool mStarted; + std::string mName; + std::string mCmdline; + }; + + //! + //! \brief Define a test for logging + //! + //! \param[in] name The name of the test. This should be a string starting with + //! "TensorRT" and containing dot-separated strings containing + //! the characters [A-Za-z0-9_]. + //! For example, "TensorRT.sample_googlenet" + //! \param[in] cmdline The command line used to reproduce the test + // + //! \return a TestAtom that can be used in Logger::reportTest{Start,End}(). + //! + static TestAtom defineTest(const std::string& name, const std::string& cmdline) + { + return TestAtom(false, name, cmdline); + } + + //! + //! \brief A convenience overloaded version of defineTest() that accepts an array of command-line arguments + //! as input + //! + //! \param[in] name The name of the test + //! \param[in] argc The number of command-line arguments + //! \param[in] argv The array of command-line arguments (given as C strings) + //! + //! \return a TestAtom that can be used in Logger::reportTest{Start,End}(). + static TestAtom defineTest(const std::string& name, int argc, char const* const* argv) + { + auto cmdline = genCmdlineString(argc, argv); + return defineTest(name, cmdline); + } + + //! + //! \brief Report that a test has started. + //! + //! \pre reportTestStart() has not been called yet for the given testAtom + //! + //! \param[in] testAtom The handle to the test that has started + //! + static void reportTestStart(TestAtom& testAtom) + { + reportTestResult(testAtom, TestResult::kRUNNING); + assert(!testAtom.mStarted); + testAtom.mStarted = true; + } + + //! + //! \brief Report that a test has ended. + //! + //! \pre reportTestStart() has been called for the given testAtom + //! + //! \param[in] testAtom The handle to the test that has ended + //! \param[in] result The result of the test. Should be one of TestResult::kPASSED, + //! TestResult::kFAILED, TestResult::kWAIVED + //! + static void reportTestEnd(const TestAtom& testAtom, TestResult result) + { + assert(result != TestResult::kRUNNING); + assert(testAtom.mStarted); + reportTestResult(testAtom, result); + } + + static int reportPass(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kPASSED); + return EXIT_SUCCESS; + } + + static int reportFail(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kFAILED); + return EXIT_FAILURE; + } + + static int reportWaive(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kWAIVED); + return EXIT_SUCCESS; + } + + static int reportTest(const TestAtom& testAtom, bool pass) + { + return pass ? reportPass(testAtom) : reportFail(testAtom); + } + + Severity getReportableSeverity() const + { + return mReportableSeverity; + } + +private: + //! + //! \brief returns an appropriate string for prefixing a log message with the given severity + //! + static const char* severityPrefix(Severity severity) + { + switch (severity) + { + case Severity::kINTERNAL_ERROR: return "[F] "; + case Severity::kERROR: return "[E] "; + case Severity::kWARNING: return "[W] "; + case Severity::kINFO: return "[I] "; + case Severity::kVERBOSE: return "[V] "; + default: assert(0); return ""; + } + } + + //! + //! \brief returns an appropriate string for prefixing a test result message with the given result + //! + static const char* testResultString(TestResult result) + { + switch (result) + { + case TestResult::kRUNNING: return "RUNNING"; + case TestResult::kPASSED: return "PASSED"; + case TestResult::kFAILED: return "FAILED"; + case TestResult::kWAIVED: return "WAIVED"; + default: assert(0); return ""; + } + } + + //! + //! \brief returns an appropriate output stream (cout or cerr) to use with the given severity + //! + static std::ostream& severityOstream(Severity severity) + { + return severity >= Severity::kINFO ? std::cout : std::cerr; + } + + //! + //! \brief method that implements logging test results + //! + static void reportTestResult(const TestAtom& testAtom, TestResult result) + { + severityOstream(Severity::kINFO) << "&&&& " << testResultString(result) << " " << testAtom.mName << " # " + << testAtom.mCmdline << std::endl; + } + + //! + //! \brief generate a command line string from the given (argc, argv) values + //! + static std::string genCmdlineString(int argc, char const* const* argv) + { + std::stringstream ss; + for (int i = 0; i < argc; i++) + { + if (i > 0) + ss << " "; + ss << argv[i]; + } + return ss.str(); + } + + Severity mReportableSeverity; +}; + +namespace +{ + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kVERBOSE +//! +//! Example usage: +//! +//! LOG_VERBOSE(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_VERBOSE(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kVERBOSE); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINFO +//! +//! Example usage: +//! +//! LOG_INFO(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_INFO(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINFO); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kWARNING +//! +//! Example usage: +//! +//! LOG_WARN(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_WARN(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kWARNING); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kERROR +//! +//! Example usage: +//! +//! LOG_ERROR(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_ERROR(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kERROR); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINTERNAL_ERROR +// ("fatal" severity) +//! +//! Example usage: +//! +//! LOG_FATAL(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_FATAL(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINTERNAL_ERROR); +} + +} // anonymous namespace + +#endif // TENSORRT_LOGGING_H \ No newline at end of file diff --git a/unet/unet.cpp b/unet/unet.cpp new file mode 100644 index 0000000..0e8e0fa --- /dev/null +++ b/unet/unet.cpp @@ -0,0 +1,372 @@ +#include +#include +#include "cuda_runtime_api.h" +#include "logging.h" +#include "common.hpp" +#define DEVICE 0 +#define NET s // s m l x +#define NETSTRUCT(str) createEngine_##str +#define CREATENET(net) NETSTRUCT(net) +#define STR1(x) #x +#define STR2(x) STR1(x) +// #define USE_FP16 // comment out this if want to use FP16 +#define CONF_THRESH 0.5 +#define BATCH_SIZE 1 +// stuff we know about the network and the input/output blobs +static const int INPUT_H = 816; +static const int INPUT_W = 672; +static const int OUTPUT_SIZE = 672*816; + +const char* INPUT_BLOB_NAME = "data"; +const char* OUTPUT_BLOB_NAME = "prob"; + +using namespace nvinfer1; + +static Logger gLogger; + + +cv::Mat preprocess_img(cv::Mat& img) { + int w, h, x, y; + float r_w = INPUT_W / (img.cols*1.0); + float r_h = INPUT_H / (img.rows*1.0); + if (r_h > r_w) { + w = INPUT_W; + h = r_w * img.rows; + x = 0; + y = (INPUT_H - h) / 2; + } else { + w = r_h* img.cols; + h = INPUT_H; + x = (INPUT_W - w) / 2; + y = 0; + } + cv::Mat re(h, w, CV_8UC3); + cv::resize(img, re, re.size(), 0, 0, cv::INTER_CUBIC); + cv::Mat out(INPUT_H, INPUT_W, CV_8UC3, cv::Scalar(128, 128, 128)); + re.copyTo(out(cv::Rect(x, y, re.cols, re.rows))); + return out; +} + + + +ILayer* doubleConv(INetworkDefinition *network, std::map& weightMap, ITensor& input, int outch, int ksize, std::string lname, int midch){ + // Weights emptywts{DataType::kFLOAT, nullptr, 0}; + // int p = ksize / 2; + // if (midch==NULL){ + // midch = outch; + // } + IConvolutionLayer* conv1 = network->addConvolutionNd(input, midch, DimsHW{ksize, ksize}, weightMap[lname + ".double_conv.0.weight"], weightMap[lname + ".double_conv.0.bias"]); + conv1->setStrideNd(DimsHW{1, 1}); + conv1->setPaddingNd(DimsHW{1, 1}); + conv1->setNbGroups(1); + IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + ".double_conv.1", 0); + IActivationLayer* relu1 = network->addActivation(*bn1->getOutput(0), ActivationType::kLEAKY_RELU); + IConvolutionLayer* conv2 = network->addConvolutionNd(*relu1->getOutput(0), outch, DimsHW{3, 3}, weightMap[lname + ".double_conv.3.weight"], weightMap[lname + ".double_conv.3.bias"]); + conv2->setStrideNd(DimsHW{1, 1}); + conv2->setPaddingNd(DimsHW{1, 1}); + conv2->setNbGroups(1); + IScaleLayer* bn2 = addBatchNorm2d(network, weightMap, *conv2->getOutput(0), lname + ".double_conv.4", 0); + IActivationLayer* relu2 = network->addActivation(*bn2->getOutput(0), ActivationType::kLEAKY_RELU); + assert(relu2); + return relu2; +} + +ILayer* down(INetworkDefinition *network, std::map& weightMap, ITensor& input, int outch, int p, std::string lname){ + + IPoolingLayer* pool1 = network->addPoolingNd(input, PoolingType::kMAX, DimsHW{2, 2}); + assert(pool1); + ILayer* dcov1 = doubleConv(network,weightMap,*pool1->getOutput(0),outch,3,lname+".maxpool_conv.1",outch); + assert(dcov1); + return dcov1; +} + +ILayer* up(INetworkDefinition *network, std::map& weightMap, ITensor& input1, ITensor& input2, int resize, int outch, int midch, std::string lname){ + float *deval = reinterpret_cast(malloc(sizeof(float) * resize * 2 * 2)); + for (int i = 0; i < resize * 2 * 2; i++) { + deval[i] = 1.0; + } + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + Weights deconvwts1{DataType::kFLOAT, deval, resize * 2 * 2}; + IDeconvolutionLayer* deconv1 = network->addDeconvolutionNd(input1, resize, DimsHW{2, 2}, deconvwts1, emptywts); + deconv1->setStrideNd(DimsHW{2, 2}); + deconv1->setNbGroups(resize); + weightMap["deconvwts."+lname] = deconvwts1; + + int diffx = input2.getDimensions().d[1]-deconv1->getOutput(0)->getDimensions().d[1]; + int diffy = input2.getDimensions().d[2]-deconv1->getOutput(0)->getDimensions().d[2]; + // IPoolingLayer* pool1 = network->addPooling(dcov1, PoolingType::kMAX, DimsHW{2, 2}); + // pool1->setStrideNd(DimsHW{2, 2}); + // dcov1->add_pading + ILayer* pad1 = network->addPaddingNd(*deconv1->getOutput(0),DimsHW{diffx / 2, diffy / 2},DimsHW{diffx - (diffx / 2), diffy - (diffy / 2)}); + // dcov1->setPaddingNd(DimsHW{diffx / 2, diffx - diffx / 2},DimsHW{diffy / 2, diffy - diffy / 2}); + ITensor* inputTensors[] = {&input2,pad1->getOutput(0)}; + auto cat = network->addConcatenation(inputTensors, 2); + assert(cat); + if (midch==64){ + ILayer* dcov1 = doubleConv(network,weightMap,*cat->getOutput(0),outch,3,lname+".conv",outch); + assert(dcov1); + return dcov1; + }else{ + int midch1 = outch/2; + ILayer* dcov1 = doubleConv(network,weightMap,*cat->getOutput(0),midch1,3,lname+".conv",outch); + assert(dcov1); + return dcov1; + } + + // assert(dcov1); + + // return dcov1; +} + +ILayer* outConv(INetworkDefinition *network, std::map& weightMap, ITensor& input, int outch, std::string lname){ + // Weights emptywts{DataType::kFLOAT, nullptr, 0}; + + IConvolutionLayer* conv1 = network->addConvolutionNd(input, 1, DimsHW{1, 1}, weightMap[lname + ".conv.weight"], weightMap[lname + ".conv.bias"]); + assert(conv1); + conv1->setStrideNd(DimsHW{1, 1}); + conv1->setPaddingNd(DimsHW{0, 0}); + conv1->setNbGroups(1); + return conv1; +} + + + +ICudaEngine* createEngine_l(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig* config, DataType dt) { + INetworkDefinition* network = builder->createNetworkV2(0U); + + // Create input tensor of shape {3, INPUT_H, INPUT_W} with name INPUT_BLOB_NAME + ITensor* data = network->addInput(INPUT_BLOB_NAME, dt, Dims3{ 3, INPUT_H, INPUT_W }); + assert(data); + + std::map weightMap = loadWeights("/home/sycv/workplace/pengyuzhou/tensorrtx/unet/unet_816_672.wts"); + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + + // build network + auto x1 = doubleConv(network,weightMap,*data,64,3,"inc",64); + auto x2 = down(network,weightMap,*x1->getOutput(0),128,1,"down1"); + auto x3 = down(network,weightMap,*x2->getOutput(0),256,1,"down2"); + auto x4 = down(network,weightMap,*x3->getOutput(0),512,1,"down3"); + auto x5 = down(network,weightMap,*x4->getOutput(0),512,1,"down4"); + ILayer* x6 = up(network,weightMap,*x5->getOutput(0),*x4->getOutput(0),512,512,512,"up1"); + ILayer* x7 = up(network,weightMap,*x6->getOutput(0),*x3->getOutput(0),256,256,256,"up2"); + ILayer* x8 = up(network,weightMap,*x7->getOutput(0),*x2->getOutput(0),128,128,128,"up3"); + ILayer* x9 = up(network,weightMap,*x8->getOutput(0),*x1->getOutput(0),64,64,64,"up4"); + ILayer* x10 = outConv(network,weightMap,*x9->getOutput(0),OUTPUT_SIZE,"outc"); + std::cout << "set name out" << std::endl; + x10->getOutput(0)->setName(OUTPUT_BLOB_NAME); + network->markOutput(*x10->getOutput(0)); + + // Build engine + builder->setMaxBatchSize(maxBatchSize); + config->setMaxWorkspaceSize(16 * (1 << 20)); // 16MB +#ifdef USE_FP16 + config->setFlag(BuilderFlag::kFP16); +#endif + std::cout << "Building engine, please wait for a while..." << std::endl; + ICudaEngine* engine = builder->buildEngineWithConfig(*network, *config); + std::cout << "Build engine successfully!" << std::endl; + + // Don't need the network any more + network->destroy(); + + // Release host memory + for (auto& mem : weightMap) + { + free((void*)(mem.second.values)); + } + + return engine; +} + + +void APIToModel(unsigned int maxBatchSize, IHostMemory** modelStream) { + // Create builder + IBuilder* builder = createInferBuilder(gLogger); + IBuilderConfig* config = builder->createBuilderConfig(); + + // Create model to populate the network, then set the outputs and create an engine + // ICudaEngine* engine = (CREATENET(NET))(maxBatchSize, builder, config, DataType::kFLOAT); + ICudaEngine* engine = createEngine_l(maxBatchSize, builder, config, DataType::kFLOAT); + assert(engine != nullptr); + + // Serialize the engine + (*modelStream) = engine->serialize(); + + // Close everything down + engine->destroy(); + builder->destroy(); +} + +void doInference(IExecutionContext& context, float* input, float* output, int batchSize) { + const ICudaEngine& engine = context.getEngine(); + + // Pointers to input and output device buffers to pass to engine. + // Engine requires exactly IEngine::getNbBindings() number of buffers. + assert(engine.getNbBindings() == 2); + void* buffers[2]; + + // In order to bind the buffers, we need to know the names of the input and output tensors. + // Note that indices are guaranteed to be less than IEngine::getNbBindings() + const int inputIndex = engine.getBindingIndex(INPUT_BLOB_NAME); + const int outputIndex = engine.getBindingIndex(OUTPUT_BLOB_NAME); + + // Create GPU buffers on device + CHECK(cudaMalloc(&buffers[inputIndex], batchSize * 3 * INPUT_H * INPUT_W * sizeof(float))); + CHECK(cudaMalloc(&buffers[outputIndex], batchSize * OUTPUT_SIZE * sizeof(float))); + + // Create stream + cudaStream_t stream; + CHECK(cudaStreamCreate(&stream)); + + // DMA input batch data to device, infer on the batch asynchronously, and DMA output back to host + CHECK(cudaMemcpyAsync(buffers[inputIndex], input, batchSize * 3 * INPUT_H * INPUT_W * sizeof(float), cudaMemcpyHostToDevice, stream)); + context.enqueue(batchSize, buffers, stream, nullptr); + CHECK(cudaMemcpyAsync(output, buffers[outputIndex], batchSize * OUTPUT_SIZE * sizeof(float), cudaMemcpyDeviceToHost, stream)); + //流同步：通过cudaStreamSynchronize()来协调。 + cudaStreamSynchronize(stream); + + // Release stream and buffers + cudaStreamDestroy(stream); + CHECK(cudaFree(buffers[inputIndex])); + CHECK(cudaFree(buffers[outputIndex])); +} + +struct Detection{ + float mask[INPUT_W*INPUT_H*1]; + }; + +float sigmoid(float x) +{ + return (1 / (1 + exp(-x))); +} + +void process_cls_result(Detection &res, float *output) { + for(int i=0;i(modelStream->data()), modelStream->size()); + modelStream->destroy(); + return 0; + } else if (argc == 3 && std::string(argv[1]) == "-d") { + std::ifstream file(engine_name, std::ios::binary); + if (file.good()) { + file.seekg(0, file.end); + size = file.tellg(); + file.seekg(0, file.beg); + trtModelStream = new char[size]; + assert(trtModelStream); + file.read(trtModelStream, size); + file.close(); + } + } else { + std::cerr << "arguments not right!" << std::endl; + std::cerr << "./unet -s // serialize model to plan file" << std::endl; + std::cerr << "./unet -d ../samples // deserialize plan file and run inference" << std::endl; + return -1; + } + + std::vector file_names; + if (read_files_in_dir(argv[2], file_names) < 0) { + std::cout << "read_files_in_dir failed." << std::endl; + return -1; + } + + // prepare input data --------------------------- + static float data[BATCH_SIZE * 3 * INPUT_H * INPUT_W]; + //for (int i = 0; i < 3 * INPUT_H * INPUT_W; i++) + // data[i] = 1.0; + static float prob[BATCH_SIZE * OUTPUT_SIZE]; + IRuntime* runtime = createInferRuntime(gLogger); + assert(runtime != nullptr); + ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size); + assert(engine != nullptr); + IExecutionContext* context = engine->createExecutionContext(); + assert(context != nullptr); + delete[] trtModelStream; + + int fcount = 0; + for (int f = 0; f < (int)file_names.size(); f++) { + fcount++; + if (fcount < BATCH_SIZE && f + 1 != (int)file_names.size()) continue; + for (int b = 0; b < fcount; b++) { + cv::Mat img = cv::imread(std::string(argv[2]) + "/" + file_names[f - fcount + 1 + b]); + if (img.empty()) continue; + cv::Mat pr_img = preprocess_img(img); // letterbox BGR to RGB + // cv::imwrite("s_o" + file_names[f - fcount + 1 + b] + "_unet.jpg", pr_img); + int i = 0; + for (int row = 0; row < INPUT_H; ++row) { + uchar* uc_pixel = pr_img.data + row * pr_img.step; + for (int col = 0; col < INPUT_W; ++col) { + data[b * 3 * INPUT_H * INPUT_W + i] = (float)uc_pixel[2] / 255.0; + data[b * 3 * INPUT_H * INPUT_W + i + INPUT_H * INPUT_W] = (float)uc_pixel[1] / 255.0; + data[b * 3 * INPUT_H * INPUT_W + i + 2 * INPUT_H * INPUT_W] = (float)uc_pixel[0] / 255.0; + uc_pixel += 3; + ++i; + } + } + } + + // Run inference + auto start = std::chrono::system_clock::now(); + doInference(*context, data, prob, BATCH_SIZE); + auto end = std::chrono::system_clock::now(); + std::cout << std::chrono::duration_cast(end - start).count() << "ms" << std::endl; + + + + std::vector batch_res(fcount); + for (int b = 0; b < fcount; b++) { + auto& res = batch_res[b]; + process_cls_result(res, &prob[b * OUTPUT_SIZE]); + } + + std::cout << fcount << std::endl; + + for (int b = 0; b < fcount; b++) { + auto& res = batch_res[b]; + float * mask = res.mask; + cv::Mat mask_mat = cv::Mat(INPUT_H,INPUT_W,CV_8UC1); + uchar *ptmp = NULL; + for(int i =0; i< INPUT_H ;i++){ + ptmp = mask_mat.ptr(i); + for(int j=0;j CONF_THRESH){ + ptmp[j] = 255; + } + else{ + ptmp[j]=0; + } + } + } + + cv::imwrite("s_" + file_names[f - fcount + 1 + b] + "_unet.jpg", mask_mat); + } + fcount = 0; + } + + // Destroy the engine + context->destroy(); + engine->destroy(); + runtime->destroy(); + + return 0; +}