Scannet++v2 Processed Folder + Coverage

[jinja12]

Hello @btsmart @brandonsmartdev !
Thank you for this work! I was hoping to get an insight on how is the processed folder (https://github.com/btsmart/splatt3r/blob/main/data/scannetpp/scannetpp.py#L57) for scannet++v2 data setup and what does that exactly contain in your work's context ? You extract the cameras metadata (transforms_undistorted.json) from this folder while the transforms.json filde from the raw folder. It would be great if you could kindly explain why the above is done ?

Another query which I had is about the construction of the coverage directory (https://github.com/btsmart/splatt3r/blob/main/data/scannetpp/scannetpp.py#L161), how exactly is this built from the raw data provided by the data authors ? It would be great if you could give some examples for the same.

Thanks again :)

[brandonsmartdev]

Hello @jinja12,

We follow SplaTAM and process the data using the ScanNet++ Toolkit, which is how we get the undistorted images and depth (and how we get transforms_undistorted.json). In order to keep the original ScanNet++ data clean, we keep the raw data in the 'raw' folder, and modify the ScanNet++ toolkit to save everything it generates to a new 'processed' folder. To follow what we did, you can modify scannetpp_toolkit/common/configs/render.yml and scannetpp_toolkit/dslr/configs/undistort.yml and scannetpp_toolkit/dslr/configs/undistort.yml to save files to a new processed folder, or you can modify our dataloader to read these files from the default save locations used by the toolkit.

Regarding the coverage files, please see the information in this comment. The code here outlines how the coverage files can be reproduced. If you have any questions about this, please let me know, as the code in the link above is a little old and rough.