we0091234/chinese_plate_tensorrt

chinese_LPR tensorrt

车牌识别TensorRT

环境：tensorrt 8.2

1. 修改CMakeLists.txt 换成你的cuda tensorrt 以及opencv 路径

   #cuda 
   include_directories(/mnt/Gu/softWare/cuda-11.0/targets/x86_64-linux/include)
   link_directories(/mnt/Gu/softWare/cuda-11.0/targets/x86_64-linux/lib)
   
   #tensorrt 
   include_directories(/mnt/Gpan/tensorRT/TensorRT-8.2.0.6/include/)
   link_directories(/mnt/Gpan/tensorRT/TensorRT-8.2.0.6/lib/)
   

2. build

   1. mkdir build
   2. cmake ..
   3. make
   
   

3. onnx 转成tensorrt模型 onnx模型在 onnx_model文件夹 或者自己训练模型看这里车牌检测 车牌识别

   当前在build目录
   #1 生成检测模型
   ./onnx2trt/onnx2trt  ../onnx_model/plate_detect.onnx ./plate_detect.trt  1
   #2 生成识别模型
   ./onnx2trt/onnx2trt  ../onnx_model/plate_rec.onnx ./plate_rec.trt  1
   

4. 推理

   ./plate_detect_rec ./plate_detect.trt  ./plate_rec.trt ../test_imgs
   

   结果显示在控制台

chinese_plate_tensorrt's Issues

在windows下fp16支持吗？推理完检测框不准确，好多重叠框

Segment fault when convert plate_detect.onnx

When I convert plate_detect.onnx, there is segment fault, the log is as follows.

./onnx2trt/onnx2trt  ../onnx_model/plate_detect.onnx ./plate_detect.trt  1
[02/15/2023-10:48:50] [W] [TRT] onnx2trt_utils.cpp:366: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[02/15/2023-10:48:51] [W] [TRT] TensorRT was linked against cuBLAS/cuBLASLt 11.6.5 but loaded cuBLAS/cuBLASLt 11.5.4
[02/15/2023-10:48:51] [W] [TRT] TensorRT was linked against cuDNN 8.2.1 but loaded cuDNN 8.0.5
段错误 (核心已转储)