Preprocessing steps for custom rgbd dataset

[msingh27]

Hi, thanks a lot for sharing the code-base.

I wanted to know how one could use the pre-trained model on custom RBG-D dataset. Regarding this I had the following questions:

1. Given 2 consecutive frame depth maps, do I also need to pre-compute a 2d flow between them for the model inference as shown here

2. What should be preprocessing steps for the rgbd data for the pre-trained model inputs. Should I normalize the range of depth map values ( I think for Kitti scene its [-1,80]) , resize the input image dimensions, remove certain point cloud (like plane ground surface).

Best, Mayank