prior_mean

[ZZZENG]

How is the prior_mean calculated?
Thanks~

[ZZZENG]

prior_mean

kitti is 2.54 and nyu is 1.54

[cnexah]

It's the mean of the log of depth values at all valid pixels at all training images.

[ZZZENG]

It's the mean of the log of depth values at all valid pixels at all training images.

Does this mean that before each training session, it is necessary to calculate the depth mean of all current training images?

[cnexah]

If you train on KITTI or NYU, just use the fixed number 2.54 or 1.54 respectively. It's not necessary to compute it again.
If you train on your own data set, I would recommend to compute the depth mean.

[ZZZENG]

If you train on KITTI or NYU, just use the fixed number 2.54 or 1.54 respectively. It's not necessary to compute it again. If you train on your own data set, I would recommend to compute the depth mean.

Got it, thank you very much.