/face-detection-benchmarks

Benchmarking of all publicly accessible implementations of face detection models

Primary LanguagePythonMIT LicenseMIT

face-detection-benchmarks

Benchmarking of all publicly accessible implementations of face detection models. A summary is provided in the section below.

System Information:

{
    "platform": "Ubuntu 22.04",
    "architecture": "x86_64",
    "processor": "Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz",
    "python_version": "3.9.12 (64 bit)",
}
Model Img W Img H Min (ms) Max (ms) Mean (ms) FPS
Dlib HOG 1920 1080 638.162 923.955 773.509 1.29281
OpenCV Haar 1920 1080 56.5779 213.733 98.8368 10.1177
MTCNN TensorFlow 1920 1080 552.752 1203.43 686.743 1.45615
MTCNN OpenCV 1920 1080 149.676 572.335 222.918 4.48596
InsightFace SCRFD 500MF 1920 1080 9.71508 14.7526 11.7123 85.3803
InsightFace SCRFD 10GF 1920 1080 61.5664 110.729 85.6653 11.6733
Ultra-Light-320 1920 1080 4.13775 10.1066 5.42747 184.248
Ultra-Light-640 1920 1080 11.4336 18.7521 13.8899 71.9949
Mediapipe BlazeFace 1920 1080 2.54941 4.98104 3.06858 325.883
Dlib HOG 1280 720 281.621 425.411 337.644 2.9617
OpenCV Haar 1280 720 29.5238 102.435 49.3472 20.2646
MTCNN TensorFlow 1280 720 409.094 764.485 480.591 2.08077
MTCNN OpenCV 1280 720 73.1196 331.995 109.978 9.09273
InsightFace SCRFD 500MF 1280 720 10.4578 13.1874 11.1333 89.8208
InsightFace SCRFD 10GF 1280 720 58.6989 122.684 84.7475 11.7998
Ultra-Light-320 1280 720 3.75271 6.52814 4.387 227.946
Ultra-Light-640 1280 720 10.7315 17.1158 13.6277 73.3802
Mediapipe BlazeFace 1280 720 2.25854 3.58415 2.82665 353.775
Dlib HOG 640 360 73.1757 103.079 88.1228 11.3478
OpenCV Haar 640 360 11.5402 26.2287 16.7382 59.7436
MTCNN TensorFlow 640 360 285.993 569.742 318.214 3.14254
MTCNN OpenCV 640 360 22.3444 71.3766 33.3338 29.9996
InsightFace SCRFD 500MF 640 360 8.35896 13.5674 9.99501 100.05
InsightFace SCRFD 10GF 640 360 51.2414 98.8905 75.815 13.19
Ultra-Light-320 640 360 3.27849 8.59141 4.12667 242.326
Ultra-Light-640 640 360 10.2892 22.0606 12.8298 77.9437
Mediapipe BlazeFace 640 360 2.17795 5.17941 2.87227 348.156
{
    "platform": "MacOS Ventura 13.0",
    "architecture": "arm64",
    "processor": "Apple M1",
    "python_version": "3.9.6 (64 bit)",
}
Model Img W Img H Min (ms) Max (ms) Mean (ms) FPS
Dlib HOG 1920 1080 515.025 714.769 608.831 1.64249
OpenCV Haar 1920 1080 32.9342 166.14 61.1988 16.3402
MTCNN TensorFlow 1920 1080 407.807 769.818 494.972 2.02032
MTCNN OpenCV 1920 1080 163.354 394.903 207.8 4.81231
InsightFace SCRFD 500MF 1920 1080 21.0996 23.5791 22.1802 45.0852
InsightFace SCRFD 10GF 1920 1080 76.1549 78.449 77.2323 12.9479
Ultra-Light-320 1920 1080 7.03168 7.98583 7.34379 136.169
Ultra-Light-640 1920 1080 21.0688 22.6569 21.6872 46.1102
Mediapipe BlazeFace 1920 1080 2.11191 4.82035 2.7504 363.584
Dlib HOG 1280 720 228.091 316.919 269.445 3.71134
OpenCV Haar 1280 720 17.9412 44.1539 26.3793 37.9085
MTCNN TensorFlow 1280 720 296.698 531.125 343.125 2.91439
MTCNN OpenCV 1280 720 79.2742 223.47 105.74 9.45712
InsightFace SCRFD 500MF 1280 720 21.3211 24.5569 22.9314 43.6083
InsightFace SCRFD 10GF 1280 720 76.108 78.788 77.0228 12.9832
Ultra-Light-320 1280 720 6.54078 7.96604 7.05832 141.677
Ultra-Light-640 1280 720 20.2301 23.2451 21.1584 47.2626
Mediapipe BlazeFace 1280 720 1.73187 2.77972 2.14672 465.827
Dlib HOG 640 360 56.5281 80.2209 67.5753 14.7983
OpenCV Haar 640 360 6.60491 13.519 9.09099 109.999
MTCNN TensorFlow 640 360 197.786 342.821 231.7 4.31592
MTCNN OpenCV 640 360 29.8293 61.682 36.552 27.3583
InsightFace SCRFD 500MF 640 360 20.4778 24.5581 22.1396 45.1679
InsightFace SCRFD 10GF 640 360 76.525 79.9201 77.5459 12.8956
Ultra-Light-320 640 360 6.32882 7.77984 7.06168 141.609
Ultra-Light-640 640 360 19.6509 22.2452 20.6835 48.3476
Mediapipe BlazeFace 640 360 1.52087 2.39682 1.85917 537.875