Hi! I'd like to ask why after training an action classifier model using CreateML, I consistently fail to detect my wrist joint while using the live preview feature, while other joints are detected correctly. What could be causing this result? Thank you to anyone for providing answers.
Training Data Sources:
YouTube videos, searching keyword: "20min workout at home" All videos display the full body.
Training Data Quantity:
Model 1: Target action: 1 video / Irrelevant action: 1 video
Model 2: Target action: 6 videos / Irrelevant action: 40 videos
Both Model 1 and Model 2 fail to detect the wrist joint during live preview.
