From my current understanding, Hamer/VITPose do not predict occlusion or provide a segmentation map of the detected hand. Is there any recommended way to render the hand with occlusion using the existing models? I understand the detectron model can detect persons and provide masks, but this often does not detect hands which are ultimately detected by VITPose. Any advice would be greatly appreciated. Thanks.
I also encountered this problem. Even though my finger was occluded, its valid value was still greater than 0.5.
How to identify the occluded and non-occluded points?
@geopavlakos
From my current understanding, Hamer/VITPose do not predict occlusion or provide a segmentation map of the detected hand. Is there any recommended way to render the hand with occlusion using the existing models? I understand the detectron model can detect persons and provide masks, but this often does not detect hands which are ultimately detected by VITPose. Any advice would be greatly appreciated. Thanks.