Skip to content

convert tensorflow pre-trained model into OpenVX graph.

License

Notifications You must be signed in to change notification settings

reger-men/tensorflow2openvx

Repository files navigation

Convert tensorflow pre-trained models to OpenVX Graph

Build Status

This converter was created to convert my own (in Keras with tensorflow as backend trained) Model to Openvx Graph. The reason being that a direct comparison of the performance between openvx on AMD Hardware and TF_inference on Nvidia GPUs.

This implementation is still not fully implemented. However I tried to create a expandable structure to make it easy to expand the source code.

Since a caffe converter already exists, I tried to keep the structure similar for easy integration into amdovx-modules.

dependencies

To get run this project you have to download and build tensorflow. This should be resolved in the future by using NNEF

run

In the main function you have to define the model path "*.pb" and the NCHW input size.

After successful run this project extract weights and biases of the input Model simultaneously generate an GDF (Graph Descriptor File) and c++ ( with make file).

Please follow the instructions here to execute the generated OpenVX Graph.

License

MIT License (c) reger-men

About

convert tensorflow pre-trained model into OpenVX graph.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published