Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
parkpow
GitHub Repository: parkpow/deep-license-plate-recognition
Path: blob/master/benchmark/benchmark_snapshot.md
641 views

Snapshot SDK Benchmark

The results are obtained using benchmark_snapshot.py.

  • SDK is started with one worker per vCPU.

  • The image used is assets/car-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.

git clone https://github.com/parkpow/deep-license-plate-recognition.git cd deep-license-plate-recognition python -m benchmark.benchmark_snapshot

Notes

  • Speed is duration / number_of_calls.

  • All numbers are in milliseconds.

  • In fast mode, number of detection steps is always 1. May result in lower accuracy when using images with small vehicles.

Google Cloud Instance - Snapshot 1.3.17

n1-standard-4 (4 vCPUs, 15 GB memory), 1 x NVIDIA Tesla T4

ModeResolutionSpeed
regular800x60031.5
fast800x60021.5
regular1280x72043.7
fast1280x72023.7
regular1920x108045.6
fast1920x108027.0
regular2560x144049.7
fast2560x144029.9

n2-standard-8 (8 vCPUs, 32 GB memory)

ModeResolutionSpeed
regular800x60027.1
fast800x60016.9
regular1280x72040.8
fast1280x72021.1
regular1920x108050.2
fast1920x108028.2
regular2560x144049.9
fast2560x144027.7

c2-standard-4 (4 vCPUs, 16 GB memory)

ModeResolutionSpeed
regular800x60037.4
fast800x60023.8
regular1280x72057.7
fast1280x72030.0
regular1920x108069.3
fast1920x108040.9
regular2560x144067.5
fast2560x144038.7

n2-standard-4 (4 vCPUs, 16 GB memory)

ModeResolutionSpeed
regular800x60042.0
fast800x60026.8
regular1280x72063.9
fast1280x72033.1
regular1920x108075.7
fast1920x108045.5
regular2560x144074.7
fast2560x144042.9

n1-standard-4 (4 vCPUs, 15 GB memory)

ModeResolutionSpeed
regular800x60061.4
fast800x60037.8
regular1280x72093.8
fast1280x72047.3
regular1920x1080111.1
fast1920x108064.4
regular2560x1440110.3
fast2560x144059.4

n2-standard-2 (2 vCPUs, 8 GB memory)

ModeResolutionSpeed
regular800x60072.1
fast800x60046.9
regular1280x720107.8
fast1280x72058.2
regular1920x1080136.4
fast1920x108086.0
regular2560x1440131.3
fast2560x144082.8

e2-standard-2 (2 vCPUs, 8 GB memory)

ModeResolutionSpeed
regular800x60094.2
fast800x60061.0
regular1280x720145.0
fast1280x72076.8
regular1920x1080179.7
fast1920x1080108.6
regular2560x1440171.0
fast2560x144098.8

Bare Metal - Snapshot 1.3.17

Intel(R) Core(TM) i7-8550U CPU @ 1.80GHz

ModeResolutionSpeed
regular800x60038.3
fast800x60021.8
regular1280x72055.3
fast1280x72033.0
regular1920x108073.0
fast1920x108037.4
regular2560x144079.6
fast2560x144046.7

Raspbery Pi 4 64 bits (1 worker)

ModeResolutionSpeed
regular800x600686.7
fast800x600401.7
regular1280x7201181.0
fast1280x720554.0
regular1920x10801373.4
fast1920x1080677.8
regular2560x14401587.1
fast2560x1440816.9

Raspberry Pi 3/4 (estimates) 32 bits

ModeResolutionSpeed
regular1280x7201300
fast1280x7201000

Jetson Nano (estimates)

ModeResolutionSpeed
regular1280x720700
fast1280x720350

LattePanda Alpha (estimates)

ModeResolutionSpeed
regular1280x720170
fast1280x720119

LattePanda V1 (estimates)

ModeResolutionSpeed
regular1280x7201250
fast1280x720875

Jetson Orin Devkit - Snapshot 1.37

With 4 workers (-e WORKERS=4):

ModeResolutionSpeed
regular800x60057.8
fast800x60038.7
regular1280x72078.2
fast1280x72040.3
regular1920x108082.9
fast1920x108043.7
regular2560x144087.1
fast2560x144047.9

With 2 workers (-e WORKERS=2):

ModeResolutionSpeed
regular800x60074.1
fast800x60049.2
regular1280x720103.4
fast1280x72052.5
regular1920x1080112.1
fast1920x108059.6
regular2560x1440122.4
fast2560x144069.2

With 1 worker (-e WORKERS=1):

ModeResolutionSpeed
regular800x600119.8
fast800x60083.0
regular1280x720170.5
fast1280x72089.8
regular1920x1080190.9
fast1920x1080108.8
regular2560x1440220.1
fast2560x1440134.3

Jetson Orin Devkit - Snapshot 1.38 (INFERENCE_ENGINE=tensorrt)

With 4 workers (-e WORKERS=4):

ModeResolutionSpeed
regular800x60036.5
fast800x60029.2
regular1280x72042.1
fast1280x72029.9
regular1920x108050.7
fast1920x108038.1
regular2560x144057.5
fast2560x144043.5

With 2 workers (-e WORKERS=2):

ModeResolutionSpeed
regular800x60047.2
fast800x60036.7
regular1280x72057.2
fast1280x72041.2
regular1920x108070.3
fast1920x108053.7
regular2560x144093.8
fast2560x144064.1

With 1 worker (-e WORKERS=1):

ModeResolutionSpeed
regular800x60073.2
fast800x60059.7
regular1280x72090.2
fast1280x72069.2
regular1920x1080113.6
fast1920x108089.1
regular2560x1440153.3
fast2560x1440110.8