问题描述

[11/16 16:54:10 trt_export]: Loading ONNX file from path ../../outputs/onnx_model/baseline_R18.onnx...
[11/16 16:54:10 trt_export]: Beginning ONNX file parsing
[11/16 16:54:10 trt_export]: Completed parsing of ONNX file.
D:\fast-reid-master\fast-reid-master\tools\deploy\trt_export.py:132: DeprecationWarning: Use set_memory_pool_limit instead.
  config.max_workspace_size = max_workspace_size * (1 << 25)
[11/16 16:54:10 trt_export]: Building an engine from file ../../outputs/onnx_model/baseline_R18.onnx; this may take a while...
Traceback (most recent call last):
  File "D:\fast-reid-master\fast-reid-master\tools\deploy\trt_export.py", line 166, in <module>
    onnx2trt(onnx_file_path, engineFile, args.mode, int8_calibrator=int8_calib)
  File "D:\fast-reid-master\fast-reid-master\tools\deploy\trt_export.py", line 145, in onnx2trt
    engine = builder.build_cuda_engine(network)
AttributeError: 'tensorrt.tensorrt.Builder' object has no attribute 'build_cuda_engine'

问题原因

tensorrt版本大于等于7时，不再使用build_cuda_engine

解决办法

将

    engine = builder.build_cuda_engine(network)

改为：

    profile = builder.create_optimization_profile()
    config = builder.create_builder_config()
    config.add_optimization_profile(profile)
    engine = builder.build_engine(network, config)

参考文章:
https://blog.csdn.net/qq_38333041/article/details/122481404