The toolbox for 3RScan dataset

This is a simple data processing toolbox of 3RScan. It converts the 3RSCan dataset into rosbag format.

Input data format

In ${ROOT_DIR}/${SCAN_ID}/sequnece/,

1. frame-xxxxxx.bb.txt
2. frame-xxxxxx.pose.txt 
3. frame-xxxxxx.color.jpg
4. frame-xxxxxx.depth.pgm (The raw depth, in low resolution)
5. frame-xxxxxx.rendered.depth.png (The reproduced depth, the same resolution as RGB)
6. frame-xxxxxx.rendered.instances.png
7. frame-xxxxxx.rendered.labels.png (Pixel level semantic segmentation)

Notice that the images that *.render* are pose-processed images by the RIO render provided by the author. Currently, we only used 1~5.

In ${ROOT_DIR}/${SCAN_ID}/, global mesh and global semantic labeled instances are provided for evaluation.

For more details, please refer to the documents of 3RScan.

Output RosBag format

Our toolbox creates a rosbag from the 3RScan data. The output data in the format,

1. bbox: geometry_msgs/PoseArray (2D bounding box)
    poses[]
        position.x: topleft.u
        position.y: toleft.v
        position.z: global object ID (Ground-truth provided by 3RScan)
        orientation.x: bottomright.u 
        orientation.y: bottomright.v
        orientation.z: semanticTypesEnum
2. Camera_pose
3. rgb
4. raw_depth
5. depth 
6. path (for visualization)

A sample dataset is provided for reference link.