haotian/duan8
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
6b135adaa9
duan8/retinaface/retina_r50.cpp
wang-xinyu 004e2d5a5b add retinaface, unfinished
2020-04-01 21:19:57 +08:00

484 lines
18 KiB
C++
Raw Blame History

#include "NvInfer.h"
#include "NvInferPlugin.h"
#include "cuda_runtime_api.h"
#include "common.h"
#include <fstream>
#include <iostream>
#include <map>
#include <sstream>
#include <vector>
#include <chrono>
//#include "plugin_factory.h"
//#include "yololayer.h"
#include <opencv2/opencv.hpp>
#define USE_FP16 // comment out this if want to use FP32
// stuff we know about the network and the input/output blobs
static const int INPUT_H = 360;
static const int INPUT_W = 640;
static const int OUTPUT_SIZE = 2048 * 12 * 20;
const char* INPUT_BLOB_NAME = "data";
const char* OUTPUT_BLOB_NAME = "prob";
using namespace nvinfer1;
static Logger gLogger;
typedef struct {
float bbox[4];
float det_confidence;
float class_id;
float class_confidence;
} Detection;
cv::Mat preprocess_img(cv::Mat& img, int input_dim) {
int w, h, x, y;
if (img.cols > img.rows) {
w = input_dim;
h = input_dim * img.rows / img.cols;
x = 0;
y = (input_dim - h) / 2;
} else {
w = input_dim * img.cols / img.rows;
h = input_dim;
x = (input_dim - 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_dim, input_dim, CV_8UC3, cv::Scalar(128, 128, 128));
re.copyTo(out(cv::Rect(x, y, re.cols, re.rows)));
return out;
}
cv::Rect get_rect(cv::Mat& img, int input_dim, float bbox[4]) {
int l, r, t, b;
if (img.cols > img.rows) {
l = bbox[0] - bbox[2]/2.f;
r = bbox[0] + bbox[2]/2.f;
t = bbox[1] - bbox[3]/2.f - (input_dim - input_dim * img.rows / img.cols) / 2;
b = bbox[1] + bbox[3]/2.f - (input_dim - input_dim * img.rows / img.cols) / 2;
l = l * img.cols / input_dim;
r = r * img.cols / input_dim;
t = t * img.cols / input_dim;
b = b * img.cols / input_dim;
} else {
l = bbox[0] - bbox[2]/2.f - (input_dim - input_dim * img.cols / img.rows) / 2;
r = bbox[0] + bbox[2]/2.f - (input_dim - input_dim * img.cols / img.rows) / 2;
t = bbox[1] - bbox[3]/2.f;
b = bbox[1] + bbox[3]/2.f;
l = l * img.rows / input_dim;
r = r * img.rows / input_dim;
t = t * img.rows / input_dim;
b = b * img.rows / input_dim;
}
return cv::Rect(l, t, r-l, b-t);
}
float iou(float lbox[4], float rbox[4]) {
float interBox[] = {
max(lbox[0] - lbox[2]/2.f , rbox[0] - rbox[2]/2.f), //left
min(lbox[0] + lbox[2]/2.f , rbox[0] + rbox[2]/2.f), //right
max(lbox[1] - lbox[3]/2.f , rbox[1] - rbox[3]/2.f), //top
min(lbox[1] + lbox[3]/2.f , rbox[1] + rbox[3]/2.f), //bottom
};
if(interBox[2] > interBox[3] || interBox[0] > interBox[1])
return 0.0f;
float interBoxS =(interBox[1]-interBox[0])*(interBox[3]-interBox[2]);
return interBoxS/(lbox[2]*lbox[3] + rbox[2]*rbox[3] -interBoxS);
}
bool cmp(Detection& a, Detection& b) {
return a.det_confidence > b.det_confidence;
}
void nms(std::vector<Detection>& res, float *output, float nms_thresh = 0.4) {
std::map<float, std::vector<Detection>> m;
for (int i = 0; i < OUTPUT_SIZE / 7; i++) {
if (output[7 * i + 4] <= 0.5) continue;
Detection det;
memcpy(&det, &output[7 * i], 7 * sizeof(float));
if (m.count(det.class_id) == 0) m.emplace(det.class_id, std::vector<Detection>());
m[det.class_id].push_back(det);
}
for (auto it = m.begin(); it != m.end(); it++) {
//std::cout << it->second[0].class_id << " --- " << std::endl;
auto& dets = it->second;
std::sort(dets.begin(), dets.end(), cmp);
for (size_t m = 0; m < dets.size(); ++m) {
auto& item = dets[m];
res.push_back(item);
for (size_t n = m + 1; n < dets.size(); ++n) {
if (iou(item.bbox, dets[n].bbox) > nms_thresh) {
dets.erase(dets.begin()+n);
--n;
}
}
}
}
}
// Load weights from files shared with TensorRT samples.
// TensorRT weight files have a simple space delimited format:
// [type] [size] <data x size in hex>
std::map<std::string, Weights> loadWeights(const std::string file)
{
std::cout << "Loading weights: " << file << std::endl;
std::map<std::string, Weights> weightMap;
// Open weights file
std::ifstream input(file);
assert(input.is_open() && "Unable to load weight file.");
// 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<uint32_t*>(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<std::string, Weights>& 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;
std::cout << "len " << len << std::endl;
float *scval = reinterpret_cast<float*>(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<float*>(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<float*>(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;
}
IActivationLayer* bottleneck(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int inch, int outch, int stride, std::string lname) {
Weights emptywts{DataType::kFLOAT, nullptr, 0};
IConvolutionLayer* conv1 = network->addConvolution(input, outch, DimsHW{1, 1}, weightMap[lname + "conv1.weight"], emptywts);
assert(conv1);
IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + "bn1", 1e-5);
IActivationLayer* relu1 = network->addActivation(*bn1->getOutput(0), ActivationType::kRELU);
assert(relu1);
IConvolutionLayer* conv2 = network->addConvolution(*relu1->getOutput(0), outch, DimsHW{3, 3}, weightMap[lname + "conv2.weight"], emptywts);
assert(conv2);
conv2->setStride(DimsHW{stride, stride});
conv2->setPadding(DimsHW{1, 1});
IScaleLayer* bn2 = addBatchNorm2d(network, weightMap, *conv2->getOutput(0), lname + "bn2", 1e-5);
IActivationLayer* relu2 = network->addActivation(*bn2->getOutput(0), ActivationType::kRELU);
assert(relu2);
IConvolutionLayer* conv3 = network->addConvolution(*relu2->getOutput(0), outch * 4, DimsHW{1, 1}, weightMap[lname + "conv3.weight"], emptywts);
assert(conv3);
IScaleLayer* bn3 = addBatchNorm2d(network, weightMap, *conv3->getOutput(0), lname + "bn3", 1e-5);
IElementWiseLayer* ew1;
if (stride != 1 || inch != outch * 4) {
IConvolutionLayer* conv4 = network->addConvolution(input, outch * 4, DimsHW{1, 1}, weightMap[lname + "downsample.0.weight"], emptywts);
assert(conv4);
conv4->setStride(DimsHW{stride, stride});
IScaleLayer* bn4 = addBatchNorm2d(network, weightMap, *conv4->getOutput(0), lname + "downsample.1", 1e-5);
ew1 = network->addElementWise(*bn4->getOutput(0), *bn3->getOutput(0), ElementWiseOperation::kSUM);
} else {
ew1 = network->addElementWise(input, *bn3->getOutput(0), ElementWiseOperation::kSUM);
}
IActivationLayer* relu3 = network->addActivation(*ew1->getOutput(0), ActivationType::kRELU);
assert(relu3);
return relu3;
}
// Creat the engine using only the API and not any parser.
ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType dt)
{
INetworkDefinition* network = builder->createNetwork();
// Create input tensor of shape { 1, 1, 32, 32 } with name INPUT_BLOB_NAME
ITensor* data = network->addInput(INPUT_BLOB_NAME, dt, Dims3{3, INPUT_H, INPUT_W});
assert(data);
std::map<std::string, Weights> weightMap = loadWeights("../retinaface.wts");
Weights emptywts{DataType::kFLOAT, nullptr, 0};
// ------------- backbone resnet50 ---------------
IConvolutionLayer* conv1 = network->addConvolution(*data, 64, DimsHW{7, 7}, weightMap["body.conv1.weight"], emptywts);
assert(conv1);
conv1->setStride(DimsHW{2, 2});
conv1->setPadding(DimsHW{3, 3});
IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), "body.bn1", 1e-5);
// Add activation layer using the ReLU algorithm.
IActivationLayer* relu1 = network->addActivation(*bn1->getOutput(0), ActivationType::kRELU);
assert(relu1);
// Add max pooling layer with stride of 2x2 and kernel size of 2x2.
IPoolingLayer* pool1 = network->addPooling(*relu1->getOutput(0), PoolingType::kMAX, DimsHW{3, 3});
assert(pool1);
pool1->setStride(DimsHW{2, 2});
pool1->setPadding(DimsHW{1, 1});
IActivationLayer* x = bottleneck(network, weightMap, *pool1->getOutput(0), 64, 64, 1, "body.layer1.0.");
x = bottleneck(network, weightMap, *x->getOutput(0), 256, 64, 1, "body.layer1.1.");
x = bottleneck(network, weightMap, *x->getOutput(0), 256, 64, 1, "body.layer1.2.");
x = bottleneck(network, weightMap, *x->getOutput(0), 256, 128, 2, "body.layer2.0.");
x = bottleneck(network, weightMap, *x->getOutput(0), 512, 128, 1, "body.layer2.1.");
x = bottleneck(network, weightMap, *x->getOutput(0), 512, 128, 1, "body.layer2.2.");
x = bottleneck(network, weightMap, *x->getOutput(0), 512, 128, 1, "body.layer2.3.");
IActivationLayer* layer2 = x;
x = bottleneck(network, weightMap, *x->getOutput(0), 512, 256, 2, "body.layer3.0.");
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 256, 1, "body.layer3.1.");
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 256, 1, "body.layer3.2.");
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 256, 1, "body.layer3.3.");
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 256, 1, "body.layer3.4.");
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 256, 1, "body.layer3.5.");
IActivationLayer* layer3 = x;
x = bottleneck(network, weightMap, *x->getOutput(0), 1024, 512, 2, "body.layer4.0.");
x = bottleneck(network, weightMap, *x->getOutput(0), 2048, 512, 1, "body.layer4.1.");
x = bottleneck(network, weightMap, *x->getOutput(0), 2048, 512, 1, "body.layer4.2.");
IActivationLayer* layer4 = x;
//IPoolingLayer* pool2 = network->addPooling(*x->getOutput(0), PoolingType::kAVERAGE, DimsHW{7, 7});
//assert(pool2);
//pool2->setStride(DimsHW{1, 1});
//
//IFullyConnectedLayer* fc1 = network->addFullyConnected(*pool2->getOutput(0), 1000, weightMap["fc.weight"], weightMap["fc.bias"]);
//assert(fc1);
layer4->getOutput(0)->setName(OUTPUT_BLOB_NAME);
std::cout << "set name out" << std::endl;
network->markOutput(*layer4->getOutput(0));
// Build engine
builder->setMaxBatchSize(maxBatchSize);
builder->setMaxWorkspaceSize(1 << 20);
#ifdef USE_FP16
builder->setFp16Mode(true);
#endif
ICudaEngine* engine = builder->buildCudaEngine(*network);
std::cout << "build out" << 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);
// Create model to populate the network, then set the outputs and create an engine
ICudaEngine* engine = createEngine(maxBatchSize, builder, 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));
cudaStreamSynchronize(stream);
// Release stream and buffers
cudaStreamDestroy(stream);
CHECK(cudaFree(buffers[inputIndex]));
CHECK(cudaFree(buffers[outputIndex]));
}
int main(int argc, char** argv)
{
std::cout << "beginning" << std::endl;
if (argc != 2) {
std::cerr << "arguments not right!" << std::endl;
std::cerr << "./retina_r50 -s // serialize model to plan file" << std::endl;
std::cerr << "./retina_r50 -d // deserialize plan file and run inference" << std::endl;
return -1;
}
// create a model using the API directly and serialize it to a stream
char *trtModelStream{nullptr};
size_t size{0};
if (std::string(argv[1]) == "-s") {
IHostMemory* modelStream{nullptr};
APIToModel(1, &modelStream);
assert(modelStream != nullptr);
std::ofstream p("retina_r50.engine");
if (!p)
{
std::cerr << "could not open plan output file" << std::endl;
return -1;
}
p.write(reinterpret_cast<const char*>(modelStream->data()), modelStream->size());
modelStream->destroy();
return 1;
} else if (std::string(argv[1]) == "-d") {
std::ifstream file("retina_r50.engine", 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 {
return -1;
}
// prepare input data ---------------------------
float data[3 * INPUT_H * INPUT_W];
for (int i = 0; i < 3 * INPUT_H * INPUT_W; i++)
data[i] = 1.0;
//cv::Mat img = cv::imread("../dog.jpg");
//cv::Mat pr_img = preprocess_img(img, INPUT_H);
//cv::imwrite("123.jpg", pr_img);
//for (int i = 0; i < INPUT_H * INPUT_W; i++) {
// data[i] = pr_img.at<cv::Vec3b>(i)[2] / 255.0;
// data[i + INPUT_H * INPUT_W] = pr_img.at<cv::Vec3b>(i)[1] / 255.0;
// data[i + 2 * INPUT_H * INPUT_W] = pr_img.at<cv::Vec3b>(i)[0] / 255.0;
//}
//PluginFactory pf;
IRuntime* runtime = createInferRuntime(gLogger);
assert(runtime != nullptr);
//ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size, &pf);
ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size, nullptr);
assert(engine != nullptr);
IExecutionContext* context = engine->createExecutionContext();
assert(context != nullptr);
// Run inference
static float prob[OUTPUT_SIZE];
for (int i = 0; i < 10; i++) {
auto start = std::chrono::system_clock::now();
doInference(*context, data, prob, 1);
//std::vector<Detection> res;
//nms(res, prob);
//for (size_t j = 0; j < res.size(); j++) {
// float *p = (float*)&res[j];
// for (size_t k = 0; k < 7; k++) {
// std::cout << p[k] << ", ";
// }
// std::cout << std::endl;
// cv::Rect r = get_rect(img, INPUT_W, res[j].bbox);
// cv::rectangle(img, r, cv::Scalar(0x27, 0xC1, 0x36), 2);
// cv::putText(img, std::to_string((int)res[j].class_id), cv::Point(r.x, r.y - 1), cv::FONT_HERSHEY_PLAIN, 1.2, cv::Scalar(0xFF, 0xFF, 0xFF), 2);
//}
auto end = std::chrono::system_clock::now();
std::cout << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms" << std::endl;
//cv::imwrite("res.jpg", img);
}
// Destroy the engine
context->destroy();
engine->destroy();
runtime->destroy();
// Print histogram of the output distribution
std::cout << "\nOutput:\n\n";
for (unsigned int i = 0; i < OUTPUT_SIZE; i++)
{
std::cout << prob[i] << ", ";
if (i % 10 == 0) std::cout << i / 10 << std::endl;
}
std::cout << std::endl;
return 0;
}
Reference in New Issue View Git Blame Copy Permalink