search

mediaeng-lfv / LFV_Disparity_Estimation

Official repository for "Depth estimation from 4D light field videos", IWAIT 2021
https://mediaeng-lfv.github.io/LFV_Disparity_Estimation/
MIT License
20 stars 3 forks source link

Is the disparity smaller than real pixel movement #1

Closed HydrogenSulfate closed 3 years ago

HydrogenSulfate commented 3 years ago

I test the disparity with a warping operation, but it is not correct when directly using the GT disparity, but is correct when using 32xdisp. Is there a scale factor between RGB image and disparity?

catdance124 commented 3 years ago

Hi @HydrogenSulfate ,

Thank you for testing our disparity.
I had overlooked whether there was a scale factor, so I looked that up.

Our data is based on MPI Sintel, which says the following:

the focal length as given by Blender, the hard-coded pixel dimensions of 32 px / mm

Unfortunately, we had left the units of focal length in [mm].
So, we need to convert the units of the focal length to [px].

disparity = (baseline[m] * focal_length[mm]) / depth[m]
  ↓
disparity = (baseline[m] * focal_length*32[px]) / depth[m]

That is to say, it is correct to use disparity x32 as you say.

Sincerely,
Takahiro Kinoshita