search

Healthcare-Robotics / bodies-at-rest

Code + Data for CVPR 2020 oral paper "Bodies at Rest: 3D Human Pose and Shape Estimation from a Pressure Image using Synthetic Data."
Other
62 stars 8 forks source link

Synchronization #4

Closed leopqrz closed 3 years ago

leopqrz commented 3 years ago

Hi,

I'd like to know about the spatial and time synchronization of pressure mat + RGB video:

1 - What pressure mattress pixels correspond to what RGB image pixels in the camera? 2 - How were different sampling rates accounted for between the pressure mat and RGB camera? 3 - Were both devices connected to the same computer?

Thank you

henryclever commented 3 years ago

Hi,

  1. I'll have to look into this and get back to you.
  2. The data is not dynamic. It is static. So there isn't video data in the dataset. During data collection, participants did not move when data was captured. So even if the sample from one modality was say a quarter second behind another, the result is inconsequential. It was synchronized at a higher rate than that though, so there is virtually no issue from this.
  3. Yes, I collected everything on the same computer.

-Henry

henryclever commented 3 years ago

Hi, regarding question #1. You can observe this with the viz_real_cvpr_release.py script in the PressurePose dataset visualization section in the readme file. The code in that script visualizes the calibrated images side by side as shown. There are four black dots at the corners of the bed on the RGB image, which are set in the python script. These correspond with the corner taxels on the pressure mat. The pressure mat is of course a lower resolution, so the mapping isn't 1-to-1, so there are a number of RGB pixels that coincide with each taxel on the mat.

-Henry