Add Yolov3 (COCO80) and Yolov3-tiny (COCO80), TensorRT for tiny not working

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
This commit is contained in:
Micaela Verucchi
2019-12-04 17:01:40 +00:00
parent 2594f59d0d
commit a2db98670a
9 changed files with 1222 additions and 8 deletions
+6
View File
@@ -86,6 +86,12 @@ target_link_libraries(test_yolo_berkeley tkDNN)
add_executable(test_yolo3_coco4 tests/yolo3_coco4/yolo3_coco4.cpp)
target_link_libraries(test_yolo3_coco4 tkDNN)
add_executable(test_yolo3 tests/yolo3/yolo3.cpp)
target_link_libraries(test_yolo3 tkDNN)
add_executable(test_yolo3_tiny tests/yolo3_tiny/yolo3_tiny.cpp)
target_link_libraries(test_yolo3_tiny tkDNN)
add_executable(test_yolo3_berkeley tests/yolo3_berkeley/yolo3_berkeley.cpp)
target_link_libraries(test_yolo3_berkeley tkDNN)
+5 -2
View File
@@ -209,7 +209,9 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Dense *l) {
ILayer* NetworkRT::convert_layer(ITensor *input, Conv2d *l) {
//std::cout<<"convert conv2D\n";
std::cout<<"convert conv2D\n";
printf("%d %d %d %d %d\n", l->kernelH, l->kernelW, l->inputs, l->outputs, l->batchnorm);
void *data_b, *bias_b, *power_b, *mean_b, *variance_b, *scales_b;
if(dtRT == DataType::kHALF) {
@@ -274,7 +276,8 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Conv2d *l) {
}
ILayer* NetworkRT::convert_layer(ITensor *input, Pooling *l) {
//std::cout<<"convert Pooling\n";
std::cout<<"convert Pooling\n";
// printf("%d %d\n", l->winW, l->winH);
PoolingType ptype;
if(l->pool_mode == tkdnnPoolingMode_t::POOLING_MAX) ptype = PoolingType::kMAX;
+19 -3
View File
@@ -26,6 +26,9 @@ Pooling::Pooling( Network *net, int winH, int winW, int strideH, int strideW,
int w = input_dim.w;
int l = input_dim.l;
printf("before: %d %d\n", h, w);
poolOn3d = false;
if(l > 1) {
@@ -38,6 +41,8 @@ Pooling::Pooling( Network *net, int winH, int winW, int strideH, int strideW,
n = l;
}
checkCUDNN( cudnnSetPooling2dDescriptor(poolingDesc, cudnnPoolingMode_t(pool_mode),
CUDNN_NOT_PROPAGATE_NAN, winH, winW, paddingH, paddingW, strideH, strideW) );
@@ -45,12 +50,23 @@ Pooling::Pooling( Network *net, int winH, int winW, int strideH, int strideW,
net->tensorFormat, net->dataType, n, c, h, w) );
//get out dim
checkCUDNN( cudnnGetPooling2dForwardOutputDim(poolingDesc, srcTensorDesc, &n, &c, &h, &w));
//h = (h + winH*this->paddingH)/strideH;
//w = (w + winW*this->paddingW)/strideW;
// checkCUDNN( cudnnGetPooling2dForwardOutputDim(poolingDesc, srcTensorDesc, &n, &c, &h, &w));
//compute w and h as in darknet
int padH = paddingH == 0? winH -1 : paddingH;
int padW = paddingW == 0? winW -1 : paddingW;
h = (h + padH - winH)/strideH +1;
w = (w + padW - winW)/strideW +1;
// h = (h + winH*this->paddingH)/strideH;
// w = (w + winW*this->paddingW)/strideW;
checkCUDNN( cudnnSetTensor4dDescriptor(dstTensorDesc,
net->tensorFormat, net->dataType, n, c, h, w) );
printf("after: %d %d\n", h, w);
output_dim.n = n;
output_dim.c = c;
+5 -3
View File
@@ -15,14 +15,16 @@ Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights) :
Layer(net) {
this->classes = classes;
this->num = num;
this->num = 3;
// load anchors
if(fname_weights != "") {
int seek = 0;
readBinaryFile(fname_weights, num, &mask_h, &mask_d, seek);
seek += num;
readBinaryFile(fname_weights, 3, &mask_h, &mask_d, seek);
seek += 3;
readBinaryFile(fname_weights, 3*num*2, &bias_h, &bias_d, seek);
for(int i=0; i<3*num*2; i++)
printf("%f\n", bias_h[i]);
}
// init default classes name
+2
View File
@@ -38,7 +38,9 @@ void readBinaryFile(std::string fname, int size, dnnType** data_h, dnnType** dat
dataFile.seekg(seek * sizeof(dnnType), dataFile.cur);
}
// printf("data_h %d size_b %d\n", *data_h,size_b);
if (!dataFile.read((char *) *data_h, size_b)) {
error_s << "Error reading file " << fname;
FatalError(error_s.str());
}
+92
View File
@@ -0,0 +1,92 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
int main() {
// Network layout
tk::dnn::dataDim_t dim(1, 3, 416, 416, 1);
tk::dnn::Network net(dim);
// create yolo3 model
std::string bin_path = "../tests/yolo3";
int classes = 80;
tk::dnn::Yolo *yolo [3];
#include "models/Yolo3.h"
// fill classes names
for(int i=0; i<3; i++) {
yolo[i]->classesNames = {"person" , "bicycle" , "car" , "motorbike" , "aeroplane" , "bus" , "train" , "truck" , "boat" , "traffic light" , "fire hydrant" , "stop sign" , "parking meter" , "bench" , "bird" , "cat" , "dog" , "horse" , "sheep" , "cow" , "elephant" , "bear" , "zebra" , "giraffe" , "backpack" , "umbrella" , "handbag" , "tie" , "suitcase" , "frisbee" , "skis" , "snowboard" , "sports ball" , "kite" , "baseball bat" , "baseball glove" , "skateboard" , "surfboard" , "tennis racket" , "bottle" , "wine glass" , "cup" , "fork" , "knife" , "spoon" , "bowl" , "banana" , "apple" , "sandwich" , "orange" , "broccoli" , "carrot" , "hot dog" , "pizza" , "donut" , "cake" , "chair" , "sofa" , "pottedplant" , "bed" , "diningtable" , "toilet" , "tvmonitor" , "laptop" , "mouse" , "remote" , "keyboard" , "cell phone" , "microwave" , "oven" , "toaster" , "sink" , "refrigerator" , "book" , "clock" , "vase" , "scissors" , "teddy bear" , "hair drier" , "toothbrush"};
}
// Load input
dnnType *data;
dnnType *input_h;
readBinaryFile(input_bin, dim.tot(), &input_h, &data);
//print network model
net.print();
//convert network to tensorRT
tk::dnn::NetworkRT netRT(&net, "yolo3.rt");
// the network have 3 outputs
tk::dnn::dataDim_t out_dim[3];
for(int i=0; i<3; i++) out_dim[i] = yolo[i]->output_dim;
dnnType *cudnn_out[3], *rt_out[3];
tk::dnn::dataDim_t dim1 = dim; //input dim
printCenteredTitle(" CUDNN inference ", '=', 30); {
dim1.print();
TIMER_START
net.infer(dim1, data);
TIMER_STOP
dim1.print();
}
for(int i=0; i<3; i++) cudnn_out[i] = yolo[i]->dstData;
printCenteredTitle(" compute detections ", '=', 30);
TIMER_START
int ndets = 0;
tk::dnn::Yolo::detection *dets = tk::dnn::Yolo::allocateDetections(tk::dnn::Yolo::MAX_DETECTIONS, classes);
for(int i=0; i<3; i++) yolo[i]->computeDetections(dets, ndets, net.input_dim.w, net.input_dim.h, 0.5);
tk::dnn::Yolo::mergeDetections(dets, ndets, classes);
for(int j=0; j<ndets; j++) {
tk::dnn::Yolo::box b = dets[j].bbox;
int x0 = (b.x-b.w/2.);
int x1 = (b.x+b.w/2.);
int y0 = (b.y-b.h/2.);
int y1 = (b.y+b.h/2.);
int cl = 0;
for(int c = 0; c < classes; ++c){
float prob = dets[j].prob[c];
if(prob > 0)
cl = c;
}
std::cout<<cl<<": "<<x0<<" "<<y0<<" "<<x1<<" "<<y1<<"\n";
}
TIMER_STOP
tk::dnn::dataDim_t dim2 = dim;
printCenteredTitle(" TENSORRT inference ", '=', 30); {
dim2.print();
TIMER_START
netRT.infer(dim2, data);
TIMER_STOP
dim2.print();
}
for(int i=0; i<3; i++) rt_out[i] = (dnnType*)netRT.buffersRT[i+1];
for(int i=0; i<3; i++) {
printCenteredTitle((std::string(" YOLO ") + std::to_string(i) + " CHECK RESULTS ").c_str(), '=', 30);
dnnType *out, *out_h;
int odim = out_dim[i].tot();
readBinaryFile(output_bins[i], odim, &out_h, &out);
std::cout<<"CUDNN vs correct"; checkResult(odim, cudnn_out[i], out);
std::cout<<"TRT vs correct"; checkResult(odim, rt_out[i], out);
std::cout<<"CUDNN vs TRT "; checkResult(odim, cudnn_out[i], rt_out[i]);
}
return 0;
}
+789
View File
@@ -0,0 +1,789 @@
[net]
# Testing
# batch=1
# subdivisions=1
# Training
batch=32
subdivisions=32
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.001
burn_in=1000
max_batches = 500200
policy=steps
steps=400000,450000
scales=.1,.1
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
# Downsample
[convolutional]
batch_normalize=1
filters=64
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=32
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
# Downsample
[convolutional]
batch_normalize=1
filters=128
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
# Downsample
[convolutional]
batch_normalize=1
filters=256
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
# Downsample
[convolutional]
batch_normalize=1
filters=512
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
# Downsample
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
[shortcut]
from=-3
activation=linear
######################
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 6,7,8
anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
classes=80
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
[route]
layers = -4
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[upsample]
stride=2
[route]
layers = -1, 61
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 3,4,5
anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
classes=80
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
[route]
layers = -4
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[upsample]
stride=2
[route]
layers = -1, 36
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 0,1,2
anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
classes=80
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
+122
View File
@@ -0,0 +1,122 @@
#include<iostream>
#include "tkdnn.h"
const char *input_bin = "../tests/yolo3_tiny/layers/input.bin";
const char *c0_bin = "../tests/yolo3_tiny/layers/c0.bin";
const char *c2_bin = "../tests/yolo3_tiny/layers/c2.bin";
const char *c4_bin = "../tests/yolo3_tiny/layers/c4.bin";
const char *c6_bin = "../tests/yolo3_tiny/layers/c6.bin";
const char *c8_bin = "../tests/yolo3_tiny/layers/c8.bin";
const char *c10_bin = "../tests/yolo3_tiny/layers/c10.bin";
const char *c12_bin = "../tests/yolo3_tiny/layers/c12.bin";
const char *c13_bin = "../tests/yolo3_tiny/layers/c13.bin";
const char *c14_bin = "../tests/yolo3_tiny/layers/c14.bin";
const char *c15_bin = "../tests/yolo3_tiny/layers/c15.bin";
const char *c18_bin = "../tests/yolo3_tiny/layers/c18.bin";
const char *c21_bin = "../tests/yolo3_tiny/layers/c21.bin";
const char *c22_bin = "../tests/yolo3_tiny/layers/c22.bin";
const char *g16_bin = "../tests/yolo3_tiny/layers/g16.bin";
const char *g23_bin = "../tests/yolo3_tiny/layers/g23.bin";
// const char *output_bin = "../tests/yolo3_tiny/layers/output.bin";
const char *output_bin = "../tests/yolo3_tiny/debug/layer23_out.bin";
int main() {
int classes = 80;
// Network layout
tk::dnn::dataDim_t dim(1, 3, 416, 416, 1);
tk::dnn::Network net(dim);
tk::dnn::Conv2d c0 (&net, 16, 3, 3, 1, 1, 1, 1, c0_bin, true);
tk::dnn::Activation a0 (&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p1 (&net, 2, 2, 2, 2, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c2 (&net, 32, 3, 3, 1, 1, 1, 1, c2_bin, true);
tk::dnn::Activation a2 (&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p3 (&net, 2, 2, 2, 2, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c4 (&net, 64, 3, 3, 1, 1, 1, 1, c4_bin, true);
tk::dnn::Activation a4 (&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p5 (&net, 2, 2, 2, 2, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c6 (&net, 128, 3, 3, 1, 1, 1, 1, c6_bin, true);
tk::dnn::Activation a6 (&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p7(&net, 2, 2, 2, 2, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c8(&net, 256, 3, 3, 1, 1, 1, 1, c8_bin, true);
tk::dnn::Activation a8(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p9(&net, 2, 2, 2, 2, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c10(&net, 512, 3, 3, 1, 1, 1, 1, c10_bin, true);
tk::dnn::Activation a10(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Pooling p11(&net, 2, 2, 1, 1, tk::dnn::POOLING_MAX);
tk::dnn::Conv2d c12(&net, 1024, 3, 3, 1, 1, 1, 1, c12_bin, true);
tk::dnn::Activation a12(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Conv2d c13(&net, 256, 1, 1, 1, 1, 0, 0, c13_bin, true);
tk::dnn::Activation a13(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Conv2d c14(&net, 512, 3, 3, 1, 1, 1, 1, c14_bin, true);
tk::dnn::Activation a14(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Conv2d c15(&net, 255, 1, 1, 1, 1, 0, 0, c15_bin, false);
tk::dnn::Yolo yolo0 (&net, classes, 2, g16_bin);
tk::dnn::Layer *m17_layers[1] = { &a13 };
tk::dnn::Route m17 (&net, m17_layers, 1);
tk::dnn::Conv2d c18(&net, 128, 1, 1, 1, 1, 0, 0, c18_bin, true);
tk::dnn::Activation a18(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Upsample u19 (&net, 2);
tk::dnn::Layer *m20_layers[2] = { &u19, &a8 };
tk::dnn::Route m20 (&net, m20_layers, 2);
tk::dnn::Conv2d c21(&net, 256, 3, 3, 1, 1, 1, 1, c21_bin, true);
tk::dnn::Activation a21(&net, tk::dnn::ACTIVATION_LEAKY);
tk::dnn::Conv2d c22(&net, 255, 1, 1, 1, 1, 0, 0, c22_bin, false);
tk::dnn::Yolo yolo1 (&net, classes, 2, g23_bin);
// Load input
dnnType *data;
dnnType *input_h;
readBinaryFile(input_bin, dim.tot(), &input_h, &data);
//print network model
net.print();
// convert network to tensorRT
tk::dnn::NetworkRT netRT(&net, "yolo3_tiny.rt");
dnnType *out_data, *out_data2; // cudnn output, tensorRT output
tk::dnn::dataDim_t dim1 = dim; //input dim
printCenteredTitle(" CUDNN inference ", '=', 30); {
dim1.print();
TIMER_START
out_data = net.infer(dim1, data);
TIMER_STOP
dim1.print();
}
// tk::dnn::dataDim_t dim2 = dim;
// printCenteredTitle(" TENSORRT inference ", '=', 30); {
// dim2.print();
// TIMER_START
// out_data2 = netRT.infer(dim2, data);
// TIMER_STOP
// dim2.print();
// }
printCenteredTitle(" CHECK RESULTS ", '=', 30);
dnnType *out, *out_h;
int out_dim = net.getOutputDim().tot();
readBinaryFile(output_bin, out_dim, &out_h, &out);
std::cout<<"CUDNN vs correct"; checkResult(out_dim, out_data, out);
// std::cout<<"TRT vs correct"; checkResult(out_dim, out_data2, out);
// std::cout<<"CUDNN vs TRT "; checkResult(out_dim, out_data, out_data2);
return 0;
}
+182
View File
@@ -0,0 +1,182 @@
[net]
# Testing
batch=1
subdivisions=1
# Training
# batch=64
# subdivisions=2
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.001
burn_in=1000
max_batches = 500200
policy=steps
steps=400000,450000
scales=.1,.1
[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=1
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
###########
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 3,4,5
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=80
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
[route]
layers = -4
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[upsample]
stride=2
[route]
layers = -1, 8
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 0,1,2
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=80
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1