search

dontLoveBugs / DORN_pytorch

PyTorch implementation of Deep Ordinal Regression Network for Monocular Depth Estimation
299 stars 67 forks source link

i want to ask you a question about alpha and beta #6

Closed czy341181 closed 4 years ago

czy341181 commented 5 years ago

in you code,your alpha and beta is 0.02 and 10 , do you calculate all the images in NYU? I only download the 1449 NYU image and make use of your code cal_alpha_beta.py,and I got the alpha and beta,18 and 254. I want to know why, Is there a problem with my unit? Or my dataset? I would appreciate it if you could answer me.

dontLoveBugs commented 5 years ago

The alpha and beta is min and max depth value of the pictures which you use for training. 18 and 254 is too big and obviously is wrong, because NYU Depth is indoor dataset. Maybe your depth data is wrong.

czy341181 commented 5 years ago

Which dataset do you use? Is it downloaded from NYU official website? How many training datasets ?

dontLoveBugs commented 5 years ago

The raw dataset has almost 500G data and this paper(DORN) use 120k pictures. You can download 500G raw data and use the matlab toolbox provided in official site to get much more pictures for training.

czy341181 commented 5 years ago

thank you. 500G is too big. Ok,fine, I will find other ways. Thank you very much!