search

TUI-NICR / ESANet

ESANet: Efficient RGB-D Semantic Segmentation for Indoor Scene Analysis
Other
236 stars 51 forks source link

A question about the unit of depth value(m or mm) #44

Closed amokame closed 2 years ago

amokame commented 2 years ago

Thanks a lot for your work!

I have a question about the unit of depth value, should it be meters or milimeters? In nyuv2 files, it seems you change the unit to milimeters, but I didnot find similar process in Cityscapes data.

danielS91 commented 2 years ago

It depends on the dataset: For all indoor datasets except of SUNRGB-D, we use mm as our data during application is in mm as well. For Cityscapes, we use m as the outdoor distances are much larger - mm would not fit an uint16 png. However, during preprocessing, we apply a standardization with mean and std computed on the current input dataset anyway.

You have a look in our follow up work EMSANet and the related dataset package - here we use hardcoded depth stats for standardization.

amokame commented 2 years ago

Thanks for your replying. So the choice of the mm or m is dependent on the distance range of the particular dataset, got it.