I was trying to run Human pose estimation demo (cpp sample) in Openvino R3 and with the model downloaded from https://download.01.org/openvinotoolkit/2018_R4/open_model_zoo/human-pose-estimation-0001/FP32/ . I got the rendered video as expected.
While using -r flag while running the demo I was getting these values
-1,-1 1000,367 -1,-1 -1,-1 -1,-1 -1,-1 -1,-1 -1,-1 992,451 992,510 -1,-1 983,443 -1,-1 -1,-1 -1,-1 -1,-1 -1,-1 -1,-1 3
for every frame, the code outputs these values. In this output there are 18 keypoint pairs which corresponding to each body point. And from the code, the final value its printing is a score which ranges from 3 to 400. I couldn't find any explanation for where this score is used and what could be its threshold?
Can anyone provide me any explanation on how the score is calculated and how its used to threshold the keypoint pairs?
Hi K, Kannan,
Please refer Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields to get a better understanding of Human pose estimation model as mentioned in model specification.