Skip to content

Latest commit

 

History

History
28 lines (18 loc) · 804 Bytes

README.md

File metadata and controls

28 lines (18 loc) · 804 Bytes

References:

https://github.com/ChHanXiao/license-plate-recoginition

step 1

git clone https://github.com/ChHanXiao/license-plate-recoginition.git

step 2

install polygraphy:

pip install colored polygraphy --extra-index-url https://pypi.ngc.nvidia.com

step 3

run export_onnx.py

step 4

convert onnx to trt engine

fp16:

trtexec --onnx=workspace/lpr_20231102_all_types.onnx --saveEngine=workspace/lpr_20231102_all_types_fp16.engine --fp16 --workspace=10240

int8:

  • modify dataloader_for_calibration.py, use your dataset
  • polygraphy convert ./workspace/lpr_20231102_all_types.onnx --int8 --data-loader-script ./dataloader_for_calibration.py --calibration-cache ./workspace/lpr_calib.cache -o ./workspace/lpr_20231102_all_types_int8.engine

step 5

run detect_lpr_trt.py