Missing pre-trained GoogleNet based SSD* files

[stevens__chris]

Hi there,

I was hoping to get the openvino interactive face demo running from here: https://software.intel.com/en-us/articles/OpenVINO-IE-Samples#interactive-face-detection. But it seems the pretrained model files are missing from here: https://software.intel.com/file/609199/download.

Any help.

Thanks

[Monique_J_Intel]

Hi Chris,

Thank you for letting us know about the broken link. We will make sure that it is fixed.

In the meantime you can use the model_downloader.py script in the package to download the googlenet topology by doing the following:

python3 downloader.py --print_all (shows you all pre-trained topologies you can download to your local machine)

python3 downloader.py --name googlenet-V4

Please disregard, what I have stated above. This topology that get's downloaded for all of the googlenets via model downloader are classification models. We are working on fixing the link.

Kind Regards,

Monique Jones

[stevens__chris]

Hi Monique,

Thanks for getting back to me. Yeah I tried to do this using the model you suggested but hit an error: "Face Detection network output layer(prob) should be DetectionOutput intel, but was SoftMax". So I'm guessing this model is not quite right? Do you have a rough timeline on when the correct model will be re-instated?

Thanks,

Chris Stevens

[Sergei_N_Intel]

Hey, Chris!

Actually, you can look in the package for the set of pretrained models. For example, deployment_tools/intel_models/face-detection-retail-0004

You will find age-gender, head-pose and emotion recognition models there, as well.

[stevens__chris]

Thanks Sergei, that did it.

[stevens__chris]

Hi Sergei, Monique, Do you document where these models come from? The source code for building and the data used to train? Is this knowledge public? I've seen it hinted at but can't find any actual data.

Thanks, Chris Stevens

[Monique_J_Intel]

Hi Chris,

The documentation on these models are in the in-package documentation here (file:///C:/Intel/computer_vision_sdk_2018.3.343/deployment_tools/documentation/docs/Pre_Trained_Models.html) for Windows

(//opt/Intel/computer_vision_sdk_2018.3.343/deployment_tools/documentation/docs/Pre_Trained_Models.html) for Linux

Of course the install directory maybe different depending on what version of OpenVINO you are using and where you installed the package.

The source code isn't yet exposed only the IR files(.xml and .bin) and there is  on-going discussion on this so stay tuned on this topic.

Kind Regards,

Monique Jones

[stevens__chris]

Many thanks for the quick reply Monique. Nice to hear, if I am allowed my 2cents as a developer working with OpenVino then the source code for these would be very helpful in truly being able to buy into the OV Intel ecosystem for ML development. Is there somewhere I should keep a lookout for this?

[Monique_J_Intel]

Hi Chris,

I totally agree with you. Once we release the source code of some of the models in the package I will create an announcement at the top of the forum. Stay tuned and thanks for your patience and feedback.

Kind Regards,

Monique Jones