Fused annotation is a powerful technique combining 2D and 3D data to provide a more comprehensive understanding of a given scene or dataset. By integrating information from different perspectives and modalities, fused annotation offers a holistic view that enhances the accuracy and richness of the annotations.

To leverage fused annotation effectively, the platform requires specific data formats for each type of data:

3D Data

For 3D data annotation, the platform accepts point cloud formats such as PTS (Point Cloud Library), LAS (LIDAR Data Exchange), or TXT (CSV) formats. These formats contain the spatial coordinates (x, y, z) of the points in the point cloud, which are essential for annotating 3D cuboids.

File structure:

3D-PC
000012_refl.ply
000011_refl.ply
000010_refl.ply

Download the example file

2D Data

For 2D data annotation, the platform accepts commonly used image formats such as PNG or JPEG and videos. These formats are suitable for annotating 2D camera frames, which are used as references to align and fuse with the 3D data

Video format

Cam-0
000012_rgb.png
000011_rgb.png
000010_rgb.png

Download the example files: video1 and video2

Video sensor calibration metadata format

Cam-0-Calib
000013_refl.csv
000012_refl.csv
000011_refl.csv

CSV structure

Download the example files: video1 sensor calibration metadata and video2 sensor calibration metadata

Sensor location metadata format

3D-PC-Fixed-World-Pos
000013_refl.csv
000012_refl.csv
000011_refl.csv

Download the example files: sensor location metadata