Unable to Run Gaze Estimation Demo with “Open/Closed Eye” and “Gaze Estimation” Models

Gaze Estimation Demo execution fails in OpenVINO™ versions 2021.1 and 2021.2.

These models executed well: face-detection-adas-0001, head-pose-estimation-adas-000, and facial-landmarks-35-adas-0002

These models failed to execute: open-closed-eye and gaze-estimation-adas-0002 

1. Download all the required models using Model Downloader:
   •  gaze-estimation-adas-0002
   •  face-detection-retail-0004
   •  face-detection-adas-0001
   •  head-pose-estimation-adas-0001
   •  facial-landmarks-35-adas-0002
   •  open-closed-eye-0001
2. Convert open-closed-eye.onnx model to Intermediate Representation (IR) format using Model Optimizer:

   python mo.py --input_model "C:\Program Files (x86)\Intel\openvino_2021\deployment_tools\open_model_zoo\tools\downloader\public\open-closed-eye-0001\open-closed-eye.onnx" --mean_values [127.0,127.0,127.0] --scale_values [255,255,255] --output 19

   The conversion argument is available here.

3. Run the demo:

   python <path_to_demo_applications_binaries>\gaze_estimation_demo.exe -i <path_to_video>\input_video.mp4 -m <path_to_model>\gaze-estimation-adas-0002.xml -m_fd <path_to_model>\face-detection-retail-0004.xml -m_hp <path_to_model>\head-pose-estimation-adas-0001.xml -m_lm <path_to_model>\facial-landmarks-35-adas-0002.xml -m_es <path_to_model>\open-closed-eye.xml

   For target device GPU, add these arguments:

   -d GPU, -d_fd GPU, -d_hp GPU, -d_lm GPU, and -d_es GPU

Refer to Gaze Estimation Demo for more information.