The segmentation mask for each pose is a 2d array of values, an image essentially, which represents the mask. This obviously should not be transferred across the native <=> js boundary on every frame, as it is a 1080p bitmap. But it supplies the information to do basic green-screening of the poser, which certainly has potential.
Somehow the web version does expose it. You can turn it on in the mediapipe web demo.
The segmentation mask for each pose is a 2d array of values, an image essentially, which represents the mask. This obviously should not be transferred across the native <=> js boundary on every frame, as it is a 1080p bitmap. But it supplies the information to do basic green-screening of the poser, which certainly has potential.
Somehow the web version does expose it. You can turn it on in the mediapipe web demo.