Facenet based Real time face recognition with NCS2

[idata]

hi,

I am using NCS2 and implementing facenet based real time face recognition. I have converted the models to IR format already. But is there any sample code on how to use this IR format files on facenet implementation with NCS2??

I see you have proved the same demo for NCS 1 based on ncsdk(ncappzoo/tensorflow/facenet demo). Similarly do you have sample demo for NCS2 or you have any guidance or instructions to use facenet code with NCS2?

Any help? much appreciated.

[idata]

Hi @jaincs

Thank you for reaching out, did you implement your facenet on inception resnet v1 or v2? There is a discussion on the Computer Vision forum around this topic, take a look.

The inception resnet v1 is currently not supported on NCS 2, however, we have opened a feature request to support it. https://software.intel.com/en-us/forums/computer-vision/topic/802451

Hope this helps!

Regards,

Jesus

[idata]

Hi Jesus,

Thanks for your reply.

Yes, I have downloaded the model from this link https://drive.google.com/open?id=1EXPBSXwTaqrSC0OhUdXNmKSh9qJUQ55- which is Inception ResNet v1 architecture. This means i cannot use facenet model with NCS2?

Or, are there any opensource model which works well with NCS2 for face recognition with only 1 photo dataset/person and > 99% accuracy?

[idata]

Hi @jaincs

Correct, the Inception resnet v1 is currently not supported on the NCS 2. However, we do have another example that contains face recognition. Take a look at the Smart Classroom C++ Demo.

https://docs.openvinotoolkit.org/latest/_inference_engine_samples_smart_classroom_demo_README.html#creating_a_gallery_for_face_recognition

Hope this helps!

Regards,

Jesus

[idata]

Hi Jesus,

May i know this intel model is based on which architecture and network? And what is the face recognition accuracy with this model? And will this model work with single image per person for Face recognition? Because, we are planning to commercialize Face recognition on one of our company product, please let me know. Thanks.

[idata]

Hi @jaincs

You can find additional information about the pre-trained models in the Open Model Zoo.

https://github.com/opencv/open_model_zoo/blob/master/intel_models/index.md

Additionally, you can find more information about training the models at the following link.

https://github.com/opencv/training_toolbox_caffe

I believe the more images you train per face will give you higher accuracy.

Regards,

Jesus