zxm97 / license-plate-recoginition_trt

convert license-plate-recoginition PyTorch model to TensorRT engine

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