Record tensorflow C ++ inference model

tensorflow an offline end side in the field of industrial vision is essential, tensorflow official from 1.10 after it only supports bazel compiled tensorflow C ++ source code, in order to use tensorflow end side to the modeling of industrial imaging, you need to apply to the tensorflow windows dll / lib.

1. First compile windows tensorflow-gpu 1.10.1 of the dll and lib, tutorials self-examination

2.keras training model, to generate the converted file into h5 tf the .ph files, can be directly used tensorflow trained document pb

3. Invoke .pb file constructed FIG visual studio 2019, before calculating the conversion into opencv session after the initialization for the tensor

4. The output results are as follows:

 

 

Guess you like

Origin www.cnblogs.com/ChrisInsistPy/p/11493161.html