Skip to content

Commit ab9f965

Browse files
DaniAffCHDaniAffCH
authored
C++ Demo - Human Segmentation (#243)
* add human segmentation c++ demo * removed debug print and update README * inverted colors for consistency * adjusted blending weight for visualization
1 parent 6929d47 commit ab9f965

File tree

4 files changed

+279
-2
lines changed

4 files changed

+279
-2
lines changed

‎models/human_segmentation_pphumanseg/CMakeLists.txt

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
cmake_minimum_required(VERSION 3.24)
2+
set(CMAKE_CXX_STANDARD 11)
3+
set(project_name "opencv_zoo_human_segmentation")
4+
5+
PROJECT (${project_name})
6+
7+
set(OPENCV_VERSION "4.9.0")
8+
set(OPENCV_INSTALLATION_PATH "" CACHE PATH "Where to look for OpenCV installation")
9+
find_package(OpenCV ${OPENCV_VERSION} REQUIRED HINTS ${OPENCV_INSTALLATION_PATH})
10+
# Find OpenCV, you may need to set OpenCV_DIR variable
11+
# to the absolute path to the directory containing OpenCVConfig.cmake file
12+
# via the command line or GUI
13+
14+
file(GLOB SourceFile
15+
"demo.cpp")
16+
# If the package has been found, several variables will
17+
# be set, you can find the full list with descriptions
18+
# in the OpenCVConfig.cmake file.
19+
# Print some message showing some of them
20+
message(STATUS "OpenCV library status:")
21+
message(STATUS " config: ${OpenCV_DIR}")
22+
message(STATUS " version: ${OpenCV_VERSION}")
23+
message(STATUS " libraries: ${OpenCV_LIBS}")
24+
message(STATUS " include path: ${OpenCV_INCLUDE_DIRS}")
25+
26+
# Declare the executable target built from your sources
27+
add_executable(${project_name} ${SourceFile})
28+
29+
# Link your application with OpenCV libraries
30+
target_link_libraries(${project_name} PRIVATE ${OpenCV_LIBS})
31+

‎models/human_segmentation_pphumanseg/README.md

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,8 @@ This model is ported from [PaddleHub](https://github.com/PaddlePaddle/PaddleHub)
44

55
## Demo
66

7+
### Python
8+
79
Run the following command to try the demo:
810

911
```shell
@@ -16,6 +18,23 @@ python demo.py --input /path/to/image -v
1618
python demo.py --help
1719
```
1820

21+
### C++
22+
23+
Install latest OpenCV and CMake >= 3.24.0 to get started with:
24+
25+
```shell
26+
# A typical and default installation path of OpenCV is /usr/local
27+
cmake -B build -D OPENCV_INSTALLATION_PATH=/path/to/opencv/installation .
28+
cmake --build build
29+
30+
# detect on camera input
31+
./build/opencv_zoo_human_segmentation
32+
# detect on an image
33+
./build/opencv_zoo_human_segmentation -i=/path/to/image
34+
# get help messages
35+
./build/opencv_zoo_human_segmentation -h
36+
```
37+
1938
### Example outputs
2039

2140
![webcam demo](./example_outputs/pphumanseg_demo.gif)

‎models/human_segmentation_pphumanseg/demo.cpp

Lines changed: 226 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,226 @@
1+
#include "opencv2/opencv.hpp"
2+
3+
#include <map>
4+
#include <vector>
5+
#include <string>
6+
#include <iostream>
7+
8+
using namespace std;
9+
using namespace cv;
10+
using namespace dnn;
11+
12+
std::vector<std::pair<int, int>> backend_target_pairs = {
13+
{DNN_BACKEND_OPENCV, DNN_TARGET_CPU},
14+
{DNN_BACKEND_CUDA, DNN_TARGET_CUDA},
15+
{DNN_BACKEND_CUDA, DNN_TARGET_CUDA_FP16},
16+
{DNN_BACKEND_TIMVX, DNN_TARGET_NPU},
17+
{DNN_BACKEND_CANN, DNN_TARGET_NPU}
18+
};
19+
20+
class PPHS
21+
{
22+
private:
23+
Net model;
24+
string modelPath;
25+
26+
Scalar imageMean = Scalar(0.5,0.5,0.5);
27+
Scalar imageStd = Scalar(0.5,0.5,0.5);
28+
Size modelInputSize = Size(192, 192);
29+
Size currentSize;
30+
31+
const String inputNames = "x";
32+
const String outputNames = "save_infer_model/scale_0.tmp_1";
33+
34+
int backend_id;
35+
int target_id;
36+
37+
public:
38+
PPHS(const string& modelPath,
39+
int backend_id = 0,
40+
int target_id = 0)
41+
: modelPath(modelPath), backend_id(backend_id), target_id(target_id)
42+
{
43+
this->model = readNet(modelPath);
44+
this->model.setPreferableBackend(backend_id);
45+
this->model.setPreferableTarget(target_id);
46+
}
47+
48+
Mat preprocess(const Mat image)
49+
{
50+
this->currentSize = image.size();
51+
Mat preprocessed = Mat::zeros(this->modelInputSize, image.type());
52+
resize(image, preprocessed, this->modelInputSize);
53+
54+
// image normalization
55+
preprocessed.convertTo(preprocessed, CV_32F, 1.0 / 255.0);
56+
preprocessed -= imageMean;
57+
preprocessed /= imageStd;
58+
59+
return blobFromImage(preprocessed);;
60+
}
61+
62+
Mat infer(const Mat image)
63+
{
64+
Mat inputBlob = preprocess(image);
65+
66+
this->model.setInput(inputBlob, this->inputNames);
67+
Mat outputBlob = this->model.forward(this->outputNames);
68+
69+
return postprocess(outputBlob);
70+
}
71+
72+
Mat postprocess(Mat image)
73+
{
74+
reduceArgMax(image,image,1);
75+
image = image.reshape(1,image.size[2]);
76+
image.convertTo(image, CV_32F);
77+
resize(image, image, this->currentSize, 0, 0, INTER_LINEAR);
78+
image.convertTo(image, CV_8U);
79+
80+
return image;
81+
}
82+
83+
};
84+
85+
86+
vector<uint8_t> getColorMapList(int num_classes) {
87+
num_classes += 1;
88+
89+
vector<uint8_t> cm(num_classes*3, 0);
90+
91+
int lab, j;
92+
93+
for (int i = 0; i < num_classes; ++i) {
94+
lab = i;
95+
j = 0;
96+
97+
while(lab){
98+
cm[i] |= (((lab >> 0) & 1) << (7 - j));
99+
cm[i+num_classes] |= (((lab >> 1) & 1) << (7 - j));
100+
cm[i+2*num_classes] |= (((lab >> 2) & 1) << (7 - j));
101+
++j;
102+
lab >>= 3;
103+
}
104+
105+
}
106+
107+
cm.erase(cm.begin(), cm.begin()+3);
108+
109+
return cm;
110+
};
111+
112+
Mat visualize(const Mat& image, const Mat& result, float fps = -1.f, float weight = 0.4)
113+
{
114+
const Scalar& text_color = Scalar(0, 255, 0);
115+
Mat output_image = image.clone();
116+
117+
vector<uint8_t> color_map = getColorMapList(256);
118+
119+
Mat cmm(color_map);
120+
121+
cmm = cmm.reshape(1,{3,256});
122+
123+
if (fps >= 0)
124+
{
125+
putText(output_image, format("FPS: %.2f", fps), Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, text_color, 2);
126+
}
127+
128+
Mat c1, c2, c3;
129+
130+
LUT(result, cmm.row(0), c1);
131+
LUT(result, cmm.row(1), c2);
132+
LUT(result, cmm.row(2), c3);
133+
134+
Mat pseudo_img;
135+
merge(std::vector<Mat>{c1,c2,c3}, pseudo_img);
136+
137+
addWeighted(output_image, weight, pseudo_img, 1 - weight, 0, output_image);
138+
139+
return output_image;
140+
};
141+
142+
string keys =
143+
"{ help h | | Print help message. }"
144+
"{ model m | human_segmentation_pphumanseg_2023mar.onnx | Usage: Path to the model, defaults to human_segmentation_pphumanseg_2023mar.onnx }"
145+
"{ input i | | Path to input image or video file. Skip this argument to capture frames from a camera.}"
146+
"{ backend_target t | 0 | Choose one of the backend-target pair to run this demo:\n"
147+
"0: (default) OpenCV implementation + CPU,\n"
148+
"1: CUDA + GPU (CUDA),\n"
149+
"2: CUDA + GPU (CUDA FP16),\n"
150+
"3: TIM-VX + NPU,\n"
151+
"4: CANN + NPU}"
152+
"{ save s | false | Specify to save results.}"
153+
"{ vis v | true | Specify to open a window for result visualization.}"
154+
;
155+
156+
157+
int main(int argc, char** argv)
158+
{
159+
CommandLineParser parser(argc, argv, keys);
160+
161+
parser.about("Human Segmentation");
162+
if (parser.has("help"))
163+
{
164+
parser.printMessage();
165+
return 0;
166+
}
167+
168+
string modelPath = parser.get<string>("model");
169+
string inputPath = parser.get<string>("input");
170+
uint8_t backendTarget = parser.get<uint8_t>("backend_target");
171+
bool saveFlag = parser.get<bool>("save");
172+
bool visFlag = parser.get<bool>("vis");
173+
174+
if (modelPath.empty())
175+
CV_Error(Error::StsError, "Model file " + modelPath + " not found");
176+
177+
PPHS humanSegmentationModel(modelPath, backend_target_pairs[backendTarget].first, backend_target_pairs[backendTarget].second);
178+
179+
VideoCapture cap;
180+
if (!inputPath.empty())
181+
cap.open(samples::findFile(inputPath));
182+
else
183+
cap.open(0);
184+
185+
if (!cap.isOpened())
186+
CV_Error(Error::StsError, "Cannot opend video or file");
187+
188+
Mat frame;
189+
Mat result;
190+
static const std::string kWinName = "Human Segmentation Demo";
191+
TickMeter tm;
192+
193+
while (waitKey(1) < 0)
194+
{
195+
cap >> frame;
196+
197+
if (frame.empty())
198+
{
199+
if(inputPath.empty())
200+
cout << "Frame is empty" << endl;
201+
break;
202+
}
203+
204+
tm.start();
205+
result = humanSegmentationModel.infer(frame);
206+
tm.stop();
207+
208+
Mat res_frame = visualize(frame, result, tm.getFPS());
209+
210+
if(visFlag || inputPath.empty())
211+
{
212+
imshow(kWinName, res_frame);
213+
if(!inputPath.empty())
214+
waitKey(0);
215+
}
216+
if(saveFlag)
217+
{
218+
cout << "Results are saved to result.jpg" << endl;
219+
220+
imwrite("result.jpg", res_frame);
221+
}
222+
}
223+
224+
return 0;
225+
}
226+

‎models/human_segmentation_pphumanseg/demo.py

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -83,8 +83,8 @@ def visualize(image, result, weight=0.6, fps=None):
8383
vis_result (np.ndarray): The visualized result.
8484
"""
8585
color_map = get_color_map_list(256)
86-
color_map = [color_map[i:i + 3] for i in range(0, len(color_map), 3)]
87-
color_map = np.array(color_map).astype(np.uint8)
86+
color_map = np.array(color_map).reshape(256, 3).astype(np.uint8)
87+
8888
# Use OpenCV LUT for color mapping
8989
c1 = cv.LUT(result, color_map[:, 0])
9090
c2 = cv.LUT(result, color_map[:, 1])
@@ -158,3 +158,4 @@ def visualize(image, result, weight=0.6, fps=None):
158158
cv.imshow('PPHumanSeg Demo', frame)
159159

160160
tm.reset()
161+

0 commit comments

Comments
 (0)