Single image inference example?

[antithing]

Hi, and thank you for making this code available. What do i need to do to run inference (using the pre-trained model) on my own random image as cv::Mat?

Do I also need to provide a bounding box?

Thank you.

[Daniil-Osokin]

Hi! It needs general steps for this group of methods: to find the pose, person need to be centered inside the image, and only the pose of this one person will be found. There is OpenVINO demo that automates step: runs person detector and for each found person runs pose estimation, however it requires to use models in internal format. If you want to do it by hands, I suggest to do the following:

• Make a crop of person from an image, so person will be centered inside this crop.
• Normalize the crop and make a tensor to feed the network.
• Use routines from val.py to extract keypoints and visualize them.

[Daniil-Osokin]

Hope, it is clear now.