anglixjtu/msg_chn_wacv20

About normalize the input

Opened this issue · 1 comments

mzy97 commented

First, thank you for sharing your work.
In the dataloader, RGB images are normalized to [0,1], but sparse depth image just divide 256 and convert it to meter, not in the range of [0,1]. Why not normalize it and how the normalization of data will affect the result?

Hi @mzy97, in my experiments with the current model, it does not affect the result much with or without the depth data being normalized to [0,1].