This repository includes scripts for batch downloading and processing of Scannet raw data.
For downloading Scannet data, you have to request the data following the instructions on the scannet website (we cannot provide it). Once you have the download script, place it in this directory. Then you can run the download_scannet_batch.sh on a file such as the scannetv2_val.txt.
The downloaded Scannet data will include the poses, color, and depth images in a .sens file. To extract the data in a friendlier format, Scannet official repository offers a Python data exporter. However, this code is developed and tested on Python 2.7 only. This repository contains an updated version of the processing script. The usage instructions are the same as in the original repository.
IMPORTANT NOTE: The semantic labels of the raw Scannet data are the id in the scannetv2-labels.combined.tsv. If you downloaded the reduced version of the dataset, termed tasks/scannet_frames_25k, the labels are transformed to the nyu40id labels. It is highly recommended to use nyu40id labels for tasks as they are more standard. This will require you to map the value of label images when you use the data in your own project.