Skip to content

Files

Latest commit

LyazSLyazS
Update README.md
Jan 22, 2020
aeebb0b · Jan 22, 2020

History

History
35 lines (24 loc) · 1.04 KB

README.md

File metadata and controls

35 lines (24 loc) · 1.04 KB

Convert tflite to pb

  1. Load TFLite by using native TF operators
  2. create .pb file

How to use

Convert tflite to json

  • First, using google flatbuffer -- flatc，flatc can convert the tflite file to json format

  • flatbuffer：https://github.com/google/flatbuffers

  • Install it：

    1. download the git
    2. cmake -G "Unix Makefiles" //create the MakeFile
    3. make //create the flatc
    4. make install //install flatc

  • Convert to json:

    1. copy the structure file 'schema.fbs' from tensorflow to the root of flatbuffer
    2. #./flatc -t schema.fbs -- xxxxx.tflite
    3. and you get the json
    4. using func tflite2json() change json to dictionary format json

Load into TF native ops

  • check your path (test_json) in Code4TFv15/TFLite2TFv15.py

  • run Change2NewJson.py and get the new json

  • run TFLite2TFv15.py and get a .pb file

  • then your can test the speed using .pb/.tflite/OpenCV.DNN

Need to know

  1. I just test the Project Mediapipe tflite,for your own using, check the ops definition in Operators.py