Question about scaling?

[ryancll]

Hi @zachteed @heilaw @anewell @jiadeng

I notice that you scale both the depth map and the translation in pose matrix with scaling ratio 0.1 when training the KITTI dataset. However, in the data streaming script for NYU and SCANNET, I only find the scaling for depth map with scaling ratio 1/5000 and 1/1000. Could you please explain why we don't need to scale the translation for NYU and SCANNET?

Thank you so much!

[zachteed]

The 1/5000 and 1/1000 come from the fact that depth maps for scannet and nyu are saved as 16bit png images and not floating points. You need to multiply by these values to convert from 16bit int to meters.

The KITTI depth map was scaled by .1 meters to make the range of depth maps consistent with the other datasets (.5 - 10m) during training. It is unscaled during testing to convert back to meters.