 |
OpenCV
4.2.0
Open Source Computer Vision
|
.2.0+dfsg_contrib_modules_dnn_superres_tutorials_benchmark_sr_benchmark
Benchmarking
The super-resolution module contains sample codes for benchmarking, in order to compare different models and algorithms. Here is presented a sample code for performing benchmarking, and then a few benchmarking results are collected. It was performed on an Intel i7-9700K CPU on an Ubuntu 18.04.02 OS.
Source Code of the sample
14 using namespace dnn_superres;
16 static void showBenchmark(vector<Mat> images,
string title,
Size imageSize,
17 const vector<String> imageTitles,
18 const vector<double> psnrValues,
19 const vector<double> ssimValues)
25 int len =
static_cast<int>(images.size());
27 int cols = 2, rows = 2;
29 Mat fullImage = Mat::zeros(
Size((cols * 10) + imageSize.
width * cols, (rows * 10) + imageSize.
height * rows),
34 for (
int i = 0; i < len; i++) {
43 resize(images[i], tmp,
Size(ROI.width, ROI.height));
58 ss <<
"PSNR: " << psnrValues[i];
71 ss <<
"SSIM: " << ssimValues[i];
84 tmp.
copyTo(fullImage(ROI));
92 static Vec2d getQualityValues(
Mat orig,
Mat upsampled)
94 double psnr =
PSNR(upsampled, orig);
95 Scalar q = quality::QualitySSIM::compute(upsampled, orig,
noArray());
96 double ssim =
mean(
Vec3d((q[0]), q[1], q[2]))[0];
97 return Vec2d(psnr, ssim);
100 int main(
int argc,
char *argv[])
105 cout <<
"usage: Arg 1: image path | Path to image" << endl;
106 cout <<
"\t Arg 2: algorithm | edsr, espcn, fsrcnn or lapsrn" << endl;
107 cout <<
"\t Arg 3: path to model file 2 \n";
108 cout <<
"\t Arg 4: scale | 2, 3, 4 or 8 \n";
112 string path = string(argv[1]);
113 string algorithm = string(argv[2]);
114 string model = string(argv[3]);
115 int scale = atoi(argv[4]);
119 cerr <<
"Couldn't load image: " << img <<
"\n";
126 Mat cropped = img(
Rect(0, 0, width, height));
135 vector <Mat> allImages;
140 sr.setModel(algorithm,
scale);
141 sr.upsample(img_downscaled, img_new);
143 vector<double> psnrValues = vector<double>();
144 vector<double> ssimValues = vector<double>();
147 Vec2f quality = getQualityValues(cropped, img_new);
149 psnrValues.push_back(quality[0]);
150 ssimValues.push_back(quality[1]);
152 cout << sr.getAlgorithm() <<
":" << endl;
153 cout <<
"PSNR: " << quality[0] <<
" SSIM: " << quality[1] << endl;
154 cout <<
"----------------------" << endl;
159 quality = getQualityValues(cropped, bicubic);
161 psnrValues.push_back(quality[0]);
162 ssimValues.push_back(quality[1]);
164 cout <<
"Bicubic " << endl;
165 cout <<
"PSNR: " << quality[0] <<
" SSIM: " << quality[1] << endl;
166 cout <<
"----------------------" << endl;
171 quality = getQualityValues(cropped, nearest);
173 psnrValues.push_back(quality[0]);
174 ssimValues.push_back(quality[1]);
176 cout <<
"Nearest neighbor" << endl;
177 cout <<
"PSNR: " << quality[0] <<
" SSIM: " << quality[1] << endl;
178 cout <<
"----------------------" << endl;
183 quality = getQualityValues(cropped, lanczos);
185 psnrValues.push_back(quality[0]);
186 ssimValues.push_back(quality[1]);
188 cout <<
"Lanczos" << endl;
189 cout <<
"PSNR: " << quality[0] <<
" SSIM: " << quality[1] << endl;
190 cout <<
"-----------------------------------------------" << endl;
192 vector <Mat> imgs{img_new, bicubic, nearest, lanczos};
193 vector <String> titles{sr.getAlgorithm(),
"Bicubic",
"Nearest neighbor",
"Lanczos"};
194 showBenchmark(imgs,
"Quality benchmark",
Size(bicubic.
cols, bicubic.
rows), titles, psnrValues, ssimValues);
Explanation
Read and downscale the image
Mat cropped = img(
Rect(0, 0, width, height));
Mat img_downscaled;
Resize the image by the scaling factor. Before that a cropping is necessary, so the images will align.
Set the model
DnnSuperResImpl sr;
sr.readModel(path);
sr.setModel(algorithm,
scale);
sr.upsample(img_downscaled, img_new);
Instantiate a dnn super-resolution object. Read and set the algorithm and scaling factor.
Perform benchmarking
double psnr =
PSNR(img_new, cropped);
Calculate PSNR and SSIM. Use OpenCVs PSNR (core opencv) and SSIM (contrib) functions to compare the images. Repeat it with other upscaling algorithms, such as other DL models or interpolation methods (eg. bicubic, nearest neighbor).
Benchmarking results
Dataset benchmarking
General100 dataset
2x scaling factor
| Avg inference time in sec (CPU) | Avg PSNR | Avg SSIM |
ESPCN | 0.008795 | 32.7059 | 0.9276 |
EDSR | 5.923450 | 34.1300 | 0.9447 |
FSRCNN | 0.021741 | 32.8886 | 0.9301 |
LapSRN | 0.114812 | 32.2681 | 0.9248 |
Bicubic | 0.000208 | 32.1638 | 0.9305 |
Nearest neighbor | 0.000114 | 29.1665 | 0.9049 |
Lanczos | 0.001094 | 32.4687 | 0.9327 |
3x scaling factor
| Avg inference time in sec (CPU) | Avg PSNR | Avg SSIM |
ESPCN | 0.005495 | 28.4229 | 0.8474 |
EDSR | 2.455510 | 29.9828 | 0.8801 |
FSRCNN | 0.008807 | 28.3068 | 0.8429 |
LapSRN | 0.282575 | 26.7330 | 0.8862 |
Bicubic | 0.000311 | 26.0635 | 0.8754 |
Nearest neighbor | 0.000148 | 23.5628 | 0.8174 |
Lanczos | 0.001012 | 25.9115 | 0.8706 |
4x scaling factor
| Avg inference time in sec (CPU) | Avg PSNR | Avg SSIM |
ESPCN | 0.004311 | 26.6870 | 0.7891 |
EDSR | 1.607570 | 28.1552 | 0.8317 |
FSRCNN | 0.005302 | 26.6088 | 0.7863 |
LapSRN | 0.121229 | 26.7383 | 0.7896 |
Bicubic | 0.000311 | 26.0635 | 0.8754 |
Nearest neighbor | 0.000148 | 23.5628 | 0.8174 |
Lanczos | 0.001012 | 25.9115 | 0.8706 |
Images
2x scaling factor
Set5: butterfly.png | size: 256x256 | |
| | | |
PSRN / SSIM / Speed (CPU) | 26.6645 / 0.9048 / 0.000201 | 23.6854 / 0.8698 / 0.000075 | 26.9476 / 0.9075 / 0.001039 |
| | | |
29.0341 / 0.9354 / 0.004157 | 29.0077 / 0.9345 / 0.006325 | 27.8212 / 0.9230 / 0.037937 | 30.0347 / 0.9453 / 2.077280 |
3x scaling factor
Urban100: img_001.png | size: 1024x644 | |
| | | |
PSRN / SSIM / Speed (CPU) | 27.0474 / 0.8484 / 0.000391 | 26.0842 / 0.8353 / 0.000236 | 27.0704 / 0.8483 / 0.002234 |
| | LapSRN is not trained for 3x
because of its architecture | |
28.0118 / 0.8588 / 0.030748 | 28.0184 / 0.8597 / 0.094173 | | 30.5671 / 0.9019 / 9.517580 |
4x scaling factor
Set14: comic.png | size: 250x361 | |
| | | |
PSRN / SSIM / Speed (CPU) | 19.6766 / 0.6413 / 0.000262 | 18.5106 / 0.5879 / 0.000085 | 19.4948 / 0.6317 / 0.001098 |
| | | |
20.0417 / 0.6302 / 0.001894 | 20.0885 / 0.6384 / 0.002103 | 20.0676 / 0.6339 / 0.061640 | 20.5233 / 0.6901 / 0.665876 |
8x scaling factor
Div2K: 0006.png | size: 1356x2040 | |
| | |
PSRN / SSIM / Speed (CPU) | 26.3139 / 0.8033 / 0.001107 | 23.8291 / 0.7340 / 0.000611 |
| | |
26.1565 / 0.7962 / 0.004782 | 26.7046 / 0.7987 / 2.274290 | |
int rows
the number of rows and columns or (-1, -1) when the matrix has more than 2 dimensions
Definition: mat.hpp:2086
InputOutputArray noArray()
CV_EXPORTS_W Mat imread(const String &filename, int flags=IMREAD_COLOR)
Loads an image from a file.
Rect2i Rect
Definition: types.hpp:462
int waitKey(int delay=0)
Waits for a pressed key.
Template class for specifying the size of an image or rectangle.
Definition: types.hpp:316
void namedWindow(const String &winname, int flags=WINDOW_AUTOSIZE)
Creates a window.
void scale(cv::Mat &mat, const cv::Mat &range, const T min, const T max)
Definition: quality_utils.hpp:90
double PSNR(InputArray src1, InputArray src2, double R=255.)
Computes the Peak Signal-to-Noise Ratio (PSNR) image quality metric.
Size2i Size
Definition: types.hpp:347
_Tp width
the width
Definition: types.hpp:339
bool empty() const
Returns true if the array has no elements.
int cols
Definition: mat.hpp:2086
cv::Scalar compute(InputArray cmp) CV_OVERRIDE
Computes SSIM.
Vec< double, 2 > Vec2d
Definition: matx.hpp:423
Vec< double, 3 > Vec3d
Definition: matx.hpp:424
@ FONT_HERSHEY_COMPLEX_SMALL
smaller version of FONT_HERSHEY_COMPLEX
Definition: imgproc.hpp:819
Template class for 2D rectangles.
Definition: types.hpp:421
void imshow(const String &winname, InputArray mat)
Displays an image in the specified window.
Scalar_< double > Scalar
Definition: types.hpp:669
void putText(InputOutputArray img, const String &text, Point org, int fontFace, double fontScale, Scalar color, int thickness=1, int lineType=LINE_8, bool bottomLeftOrigin=false)
Draws a text string.
n-dimensional dense array class
Definition: mat.hpp:792
void copyTo(OutputArray m) const
Copies the matrix to another one.
_Tp height
the height
Definition: types.hpp:340
"black box" representation of the file storage associated with a file on disk.
Definition: affine.hpp:52
Scalar mean(InputArray src, InputArray mask=noArray())
Calculates an average (mean) of array elements.
void resize(int src_type, const uchar *src_data, size_t src_step, int src_width, int src_height, uchar *dst_data, size_t dst_step, int dst_width, int dst_height, double inv_scale_x, double inv_scale_y, int interpolation)