Open nikhilchh opened 3 years ago
My new work: https://arxiv.org/abs/2104.07300
Questions for new paper: 1- Do you provide root relative 3d pose or absolute 3d pose in the new paper ? 2- When you crop one person and analyse only that portion of the image, does the model not miss the contextual information from image-surrounding the person ? That could have been important for absolute depth-scale estimation. 3- Is the code released already ?
Aspect ratio preserving Boxes leads to huge square boxes which ends up covering multiple humans in one box. Therefore for scenario where we have humans close to each other , the model end up missing some humans and detect duplicate key points for same humans.
Is there a remedy to such situations ?