search

gengshan-y / expansion

Upgrading Optical Flow to 3D Scene Flow through Optical Expansion, CVPR 2020 (Oral).
https://gengshan-y.github.io/expansion/
MIT License
172 stars 27 forks source link

motion in depth normalized? #10

Open tonytu16 opened 3 years ago

tonytu16 commented 3 years ago

When I plug in the tcz for our camera into the equation Screen Shot 2020-12-30 at 12 01 37 AM to calculate the absolute depth, do I directly plug in the values from the MID pfm files? It looks like the values are very close to 0. So 1 / (1- tau) is essentially a one matrix. Which values should I plug in to the above equation? Have the motion in depth values been normalized?

gengshan-y commented 3 years ago

The .pfm files store log motion-in-depth values. The following would get tau, 

from dataloader.depthloader import disparity_loader
tau = np.exp(disparity_loader('path-to-mid-file'))
tonytu16 commented 3 years ago

Thank you for your reply. I just want to confirm the base of the log. Is it log_10 or natural log? Thank you

gengshan-y commented 3 years ago

It's the natural logarithm.