Skip to content

jpeg: use libjpeg-turbo built-in color conversions#24875

Merged
asmorkalov merged 1 commit intoopencv:4.xfrom
tailsu:sd/jpeg-turbo-color-extensions
Jan 23, 2024
Merged

jpeg: use libjpeg-turbo built-in color conversions#24875
asmorkalov merged 1 commit intoopencv:4.xfrom
tailsu:sd/jpeg-turbo-color-extensions

Conversation

@tailsu
Copy link
Copy Markdown
Contributor

@tailsu tailsu commented Jan 17, 2024 •
edited
Loading

libjpeg-turbo has optimized built-in support for encoding/decoding from/to BGR and BGRA [1].

This PR integrates this optimized support when available. As a result both encoding and decoding from/to BGR are faster by ~2% and the same for BGRA is faster by ~3%, measured on a Mac M1 (ARM64) with a 512x512 image. Grayscale performance is unchanged.

Test code

TEST(Imgcodecs_Jpeg, encode_benchmark)
{
    cvtest::TS& ts = *cvtest::TS::ptr();
    string input = string(ts.get_data_path()) + "../cv/shared/lena.png";
    cv::Mat img = cv::imread(input);
    ASSERT_FALSE(img.empty());

    // uncomment to test BGRA branch
    // cv::cvtColor(img, img, cv::COLOR_BGR2BGRA);

    int bestTimeUs = -1;
    for (int i = 0; i < 1000; ++i) {
        std::vector<uchar> output;
        std::chrono::high_resolution_clock::time_point start = std::chrono::high_resolution_clock::now();
        cv::imencode(".jpg", img, output);
        cv::Mat roundtrip = cv::imdecode(output, cv::IMREAD_COLOR);
        std::chrono::high_resolution_clock::time_point end = std::chrono::high_resolution_clock::now();
        int timeUs = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
        if (bestTimeUs == -1 || timeUs < bestTimeUs) {
            bestTimeUs = timeUs;
        }

        if (i == 0) {
            cv::imwrite("roundtrip.jpg", roundtrip);
        }
    }
    std::cout << "Best time: " << bestTimeUs << " us" << std::endl;
}

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

  • I agree to contribute to the project under Apache 2 License.
  • To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
  • The PR is proposed to the proper branch
  • There is a reference to the original bug report and related work
  • There is accuracy test, performance test and test data in opencv_extra repository, if applicable
    Patch to opencv_extra has the same branch name.
  • The feature is well documented and sample code can be built with the project CMake

@asmorkalov asmorkalov self-requested a review January 17, 2024 10:35
@asmorkalov asmorkalov added this to the 4.10.0 milestone Jan 17, 2024
@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Jan 23, 2024 •
edited
Loading

Performance results for Intel i5-1135G7 (mobile), system instance of libjpeg (Ubuntu 20.04):

Geometric mean (ms)

Name of Test 4.x-1  patched-1 patched-1 
                                  vs    
                                4.x-1   
                              (x-factor)
Decode::JPEG 44.903  34.509      1.30   
Encode::JPEG 39.511  31.819      1.24   
decode::PNG  29.752  29.849      1.00   
encode::PNG  55.645  55.357      1.01

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Jan 23, 2024 •
edited
Loading

Jetson TK1 (ARM v7), CPU_BASELINE="", no NEON. libjpeg - system instance of libjpeg (Ubuntu 16.04).

ubuntu@jetson1:~/Projects/perf-jpeg-turbo$ python3 ../opencv/modules/ts/misc/summary.py ./4.x-1.xml ./patched-1.xml 

Geometric mean (ms)

Name of Test  4.x-1  patched-1 patched-1 
                                   vs    
                                 4.x-1   
                               (x-factor)
Decode::JPEG 153.790  130.682     1.18   
Encode::JPEG 194.503  168.693     1.15   
decode::PNG  82.813   82.154      1.01   
encode::PNG  135.009  135.075     1.00

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Jan 23, 2024

The same Intel system with self-built libjpeg from 3rdparty with NASM installed:

Geometric mean (ms)

Name of Test 4.x-1  patched-1 patched-1 
                                  vs    
                                4.x-1   
                              (x-factor)
Decode::JPEG 43.528  34.220      1.27   
Encode::JPEG 34.849  28.169      1.24   
decode::PNG  29.651  29.787      1.00   
encode::PNG  55.511  55.781      1.00

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Jan 23, 2024

The same Intel system with self-built libjpeg from 3rdparty without NASM:

Geometric mean (ms)

Name of Test  4.x-1  patched-1 patched-1 
                                   vs    
                                 4.x-1   
                               (x-factor)
Decode::JPEG 79.568   71.328      1.12   
Encode::JPEG 132.320  124.864     1.06   
decode::PNG  29.715   29.695      1.00   
encode::PNG  55.693   55.663      1.00

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Jan 23, 2024

Jetson-tk1 result with self-built libjpeg with NASM:

Geometric mean (ms)

Name of Test  4.x-1  patched-1 patched-1 
                                   vs    
                                 4.x-1   
                               (x-factor)
Decode::JPEG 134.620  130.283     1.03   
Encode::JPEG 136.431  116.768     1.17   
decode::PNG  83.308   83.628      1.00   
encode::PNG  137.066  136.243     1.01

asmorkalov
asmorkalov approved these changes Jan 23, 2024
View reviewed changes
Copy link
Copy Markdown
Contributor

@asmorkalov asmorkalov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👍 Thanks for the great job! The patch makes sense, even if NASM assembler is not found.

@asmorkalov asmorkalov merged commit 48ba45f into opencv:4.x Jan 23, 2024
@asmorkalov asmorkalov self-assigned this Jan 23, 2024
@tailsu tailsu deleted the sd/jpeg-turbo-color-extensions branch January 23, 2024 14:20
@asmorkalov asmorkalov mentioned this pull request Jan 23, 2024
Merged
@Kumataro Kumataro mentioned this pull request Mar 27, 2024
Closed
4 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@asmorkalov asmorkalov asmorkalov approved these changes

Assignees

@asmorkalov asmorkalov

Labels

category: imgcodecs category: 3rdparty optimization

Projects

None yet

Milestone

4.10.0

Development

Successfully merging this pull request may close these issues.

2 participants

@tailsu @asmorkalov