search

mohamedhassanmus / prox

Resolving 3D Human Pose Ambiguities with 3D Scene Constraints https://prox.is.tue.mpg.de
Other
220 stars 19 forks source link

Human mask in SMPL-X parameter fitting #13

Closed cuiyuexin closed 4 years ago

cuiyuexin commented 4 years ago

Hello, nice work!

I have a question now. I found two masks in the PROX-Qualitative dataset, i.e. depth mask and RGB mask. I wonder which mask do you use in the (optional) E_D loss term in human fitting process. Are the masks calculated by yourself or provided by Kinect-One? When I visualize raw data by viz_raw_data.py, is the human point cloud segmentation the result of RGB mask or depth mask?

Thank you very much!

mohamedhassanmus commented 4 years ago

I use the RGB mask which is computed from Mask-RCNN. I found this to be more stable.

cuiyuexin commented 4 years ago

OK, I get it, thank you for your kindly help!