Can you tell me where you picked up your blazepose landmarks weights? The network seems quite different from the current mediapipe ones, whether full or half body. Your model gives 31 landmarks and Google ones 35 or 27.
I'm currently adding landmarks to roi calculation to your code to reduce detection needs. It works fine with face and hands. I will commit the code after finishing with pose.
Can you tell me where you picked up your blazepose landmarks weights? The network seems quite different from the current mediapipe ones, whether full or half body. Your model gives 31 landmarks and Google ones 35 or 27. I'm currently adding landmarks to roi calculation to your code to reduce detection needs. It works fine with face and hands. I will commit the code after finishing with pose.