benchmark_blur.md

Blur SDK Benchmark

The results are obtained using benchmark_snapshot.py.

• The images used are assets/car-4k.jpg and assets/cars-4k.jpg resized to target resolution.

• API is called 50 times for each configuration.

• API calls are parallelized. The client code makes 4 calls concurrently.

• Default config with blur enabled by setting PLATES=10 and FACES=10

git clone https://github.com/parkpow/deep-license-plate-recognition.git
cd deep-license-plate-recognition
python3 -m benchmark.benchmark_snapshot --blur --sdk-url http://localhost:8001

Notes

• Speed is duration / number_of_calls.

• All numbers are in milliseconds.

Current Version Benchmarks

PlateRecognizer Blur: v1.0.8 PlateRecognizer Snapshot: v1.54.0

Local Machine Benchmark - 12th Gen Intel Core i3-1215U (6 Cores, 8 Threads, 16 GiB RAM)

Processor: Intel Core i3-1215U @ 1.20GHz Manufacturer: GenuineIntel NumCores: 6 NumLogicals: 8 CPUFrequency: 4400 Passmark: 10432

WORKERS = 1

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	185.2	57.04, 7.51	529.71, 508.80
1280x720	180.4	45.23, 8.20	426.93, 510.40
1920x1080	241.5	46.40, 9.40	430.35, 520.90
2560x1440	353.3	45.86, 8.45	489.37, 529.97
3840x2160	598.4	60.96, 8.65	510.27, 580.17

WORKERS = 1

Image: assets/cars-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	388.0	45.84, 7.52	289.65, 338.73
1280x720	377.6	40.62, 9.11	346.34, 340.29
1920x1080	400.1	41.80, 9.82	295.51, 350.61
2560x1440	447.4	42.29, 10.08	303.67, 370.47
3840x2160	631.3	54.01, 8.31	375.21, 410.12

WORKERS = 2

Image: assets/cars-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	744.0	45.21, 8.79	396.69, 338.70
1280x720	742.7	40.71, 9.52	359.91, 338.91
1920x1080	779.9	40.22, 9.63	323.12, 350.52
2560x1440	878.3	43.36, 10.31	372.47, 370.47
3840x2160	1237.2	52.9, 8.71	375.41, 410.03

Previous Version Benchmarks

AWS Instance Type - t2-large 2-vCPU(2 core) 8 GiB Memory

Processor: Intel Xeon CPU E5-2686 v4 @ 2.30GHz NumSockets: 1 Manufacturer: GenuineIntel NumCores: 2 NumLogicals: 2 CPUFrequency: 2300 Passmark: 2420.9990

WORKERS = 1

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	406.9	144.84, 28.51	705.61, 235.30
1280x720	426.6	125.23, 48.10	681.43, 235.30
1920x1080	528.0	115.30, 46.70	692.35, 236.80
2560x1440	975.7	112.86, 46.75	746.27, 235.77

WORKERS = 2

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	350.0	177.74, 35.43	851.23, 378.21
1280x720	350.7	155.78, 66.82	765.73, 382.18
1920x1080	435.9	149.61, 61.38	781.14, 390.07
2560x1440	786.3	194.81, 83.06	868.37, 441.07

AWS Instance Type - t3-xlarge 4-vCPU(2 core) 16 GiB Memory

Processor: Intel Xeon Platinum 8259CL @ 2.50GHz NumSockets: 1 Manufacturer: GenuineIntel NumCores: 2 NumLogicals: 4 CPUFrequency: 2499 Passmark: 4039.57

WORKERS = 1

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	375.0	265.66, 29.75	499.87, 222.80
1280x720	384.7	225.70, 41.67	519.84, 226.04
1920x1080	519.1	212.69,57.34	497.72, 227.89
2560x1440	964.2	209.77, 47.42	568.27, 238.00

WORKERS = 2

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	306.2	356.95, 39.76	1319.49, 387.57
1280x720	303.5	333.44, 73.56	1273.29, 392.47
1920x1080	387.6	303.41, 83.80	1297.73, 404.14
2560x1440	717.5	338.79, 104.97	1362.45, 448.89

WORKERS = 4

Image: assets/car-4k.jpg

Resolution	Speed	Peak CPU% (Blur, Snapshot)	Peak MEM in MiB (Blur, Snapshot)
800x600	295.1	378.88, 41.19	1665.84, 661.40
1280x720	291.8	396.42, 68.414	1630.47, 690.65
1920x1080	360.1	371.16, 93.14	1711.66, 715.22
2560x1440	614.7	402.30, 164.91	1800.31, 790.79