Skip to content

Conversation

@TomonobuTsujikawa
Copy link
Contributor

This PR tends to optimize the tflite model converted from nnabla, which precompute the branches with completed constant inputs.
There are 6 changes in this PR:

  • Precompute the constant branches
  • Remove Split and slice operator if the shape of input and output are same
  • Replace Split with slice if the model wants to be converted to quant model
  • Replace PReLU with a combination of operators if the model wants to be converted to quant model
  • Add the support of TRANSPOSE_CONV operator for quant model
  • Update the support of pb to pb optimization to remove redundant transpose operators (version up to tf2.x)

enable tensorflow pb optimization for v2.0
@TomonobuTsujikawa TomonobuTsujikawa added the release-note-format-converter Auto-release; Format Converter label Mar 23, 2023
@YukioOobuchi YukioOobuchi merged commit d48ced9 into master Mar 23, 2023
@YukioOobuchi YukioOobuchi deleted the feature/20230213-tflite-prune-branches branch March 23, 2023 08:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

release-note-format-converter Auto-release; Format Converter

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants