Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.

Regression results with openvino

Thib
Beginner
571 Views

Hi,

I'm trying to run a face detection model on inference engine. I got two outputs: one of them is classification output and the other one is a regression output. 

From the regression output the shape array is (1,896,16). From this I would like to use an equivalent function of get_tensor function with openvino to create landmarks but have no idea on how to do it.

If someone has an idea please.

0 Kudos
5 Replies
Munesh_Intel
Moderator
558 Views

Hi Thibault,


I would suggest you start by reading the following article, 'Facial landmarks with dlib, OpenCV, and Python’ , available at the following link:


https://www.pyimagesearch.com/2017/04/03/facial-landmarks-dlib-opencv-python/


Regards,

Munesh


Thib
Beginner
548 Views

Thanks for the reply and your help

They're using coordinates to create landmarks but I'm not sure to have this coordinates directly in output of my model:

Thib_0-1594886539910.png

I have 896 arrays of 16 values for each image capture from the blazeface model in my stream video. I guess this is just tensors.

Munesh_Intel
Moderator
521 Views

Hi Thibault,


I would suggest you refer to the following pre-trained object recognition models for more insights.

·    landmarks-regression-retail-0009

·    facial-landmarks-35-adas-0002


They can be downloaded via Model Downloader (<OPENVINO_INSTALL_DIR>/deployment_tools/open_model_zoo/tools/downloader).


Regards,

Munesh


Thib
Beginner
502 Views

Ok thanks, I already check the code from face_detection demo but they're also working with coordinates in output of the model if I'm not mistaken. I think the solution from this demo with landmarks model is not appropriate to my problem. I'm going to check again thanks.

Munesh_Intel
Moderator
459 Views

Hi Thibault,


It’s been a while since we heard from you. If you need any additional information from Intel, please submit a new question as Intel is no longer monitoring this thread.


Regards,

Munesh


Reply