Arthur151 / ROMP

Monocular, One-stage, Regression of Multiple 3D People and their 3D positions & trajectories in camera & global coordinates. ROMP[ICCV21], BEV[CVPR22], TRACE[CVPR2023]
https://www.yusun.work/
Apache License 2.0
1.35k stars 230 forks source link

Is the video taken smooth #419

Open zbc-l opened 1 year ago

zbc-l commented 1 year ago

Thank you for your excellent work. When trying to run the demo, I found that the real-time 3D model generated by webcam interaction produced a jitter. Still, there was no significant jitter when using video as the input source. Is this due to the added smoothing measures between frames?

zbc-l commented 1 year ago

In addition, during the experiment, I found that ROMP seems to need the overall picture of the human body, and if only part of the human body appears, it seems easy to identify errors. Is this because the length and width of the regression after selecting the centre is the position of the SMPL model as a whole?