Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

TFLite Support for RVC2 #23

Closed
wants to merge 4 commits into from
Closed

TFLite Support for RVC2 #23

wants to merge 4 commits into from

Conversation

kozlov721
Copy link
Collaborator

Added support for conversion from tflite files for RVC2 platform.

@kozlov721 kozlov721 added enhancement New feature or request RVC2 Changes affecting RVC2 export labels Aug 5, 2024
@kozlov721 kozlov721 self-assigned this Aug 5, 2024
Copy link
Collaborator

@klemen1999 klemen1999 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are models exported with OpenVino>=2024.2.0 runnable on DepthAI since there the default version is lower (2022.1 based on online blobconverter)

Otherwise LGTM

@kozlov721
Copy link
Collaborator Author

Are models exported with OpenVino>=2024.2.0 runnable on DepthAI

I tested the exported models with modelconverter benchmark and they do run fine, so I hope it won't cause any issues. I'll tag the latest RVC2 image before this change so we can quickly test with older version in case there will be an issue with RVC2 models not running on DepthAI.

@tersekmatija
Copy link

Are models exported with OpenVino>=2024.2.0 runnable on DepthAI

IMO this is highly dangerous. We'd rather test this on the whole ZOO as the MYRIAD was deprecated after 2022.3

@tersekmatija
Copy link

Possibly the right path instead would be TFLITE -> ONNX -> IR?

I am mainly worried about "compile_tool" failing if IR includes any new opsets for specific operations. I'd imagine at least some were introduced and some updated? These are the models we should be testing.

@kozlov721
Copy link
Collaborator Author

kozlov721 commented Aug 6, 2024

Another option might be to use the old versions when converting from ONNX and only switch to newer version of mo when converting from TFLITE. And print some warnings that this is not guaranteed to work. Additionally, we might add a switch that would keep the old version even for TFLITE and try TFLITE -> ONNX conversion first.

@kozlov721 kozlov721 closed this Aug 7, 2024
@kozlov721 kozlov721 deleted the feature/rvc2-tflite branch August 7, 2024 23:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request RVC2 Changes affecting RVC2 export
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants