Anyone had a successful implementation of EfficientDet on Openvino

[Konakanti__Kishore]

I would like to know the success rate of EfficientDet on Openvino before going ahead to train EffDet on my custom dataset. Please let me know if any one had a a successful, both speed and accuracy wise, implementation.

[Munesh_Intel]

Hi Kishore,

EfficientDet topology is not supported by OpenVINO.

For your additional information, the following page, 'Get a Deep Learning Model Performance Boost with Intel® Platforms', contains benchmarks that demonstrate high performance gains on several public neural networks for a streamlined, quick deployment on Intel® CPU, VPU and FPGA platforms.

https://docs.openvinotoolkit.org/2020.3/_docs_performance_benchmarks.html

Regards,

Munesh

[Konakanti__Kishore]

Munesh,

Thanks for the reply. 

In my case, EfficientDet-d2 has promising results. I will look at other methods to reduce inference speeds.

Thank you,
Kishore

[Munesh_Intel]

Hi Kishore,

In that case, I would suggest you validate them on your end. Train your EfficientDet-d2 model in any of the supported framework and try optimizing using Model Optimizer.

More information is available at the following page:

https://docs.openvinotoolkit.org/2020.3/_docs_MO_DG_prepare_model_Prepare_Trained_Model.html

And do share the eventual results for the benefit of everyone in our OpenVINO community.

Regards,

Munesh

[Dmitry_K_Intel3]

OpenVINO now supports EfficientDet. Please use master branch to convert a model to IR: https://github.com/openvinotoolkit/openvino/. Then 2020.4 or later release can work with this IR.

small tutorial: https://github.com/dkurt/openvino_efficientdet