CODEWITHSUNDEEP
object-detectionlidarkitti
Mai
Mai

Reputation: 31

How to label my own point cloud and image data to have the 3D training labels (.txt) files like KITTI 3D object detection dataset?

I have point cloud files (pcd + bin) and image (png) files. I would like to label them by my own and extract text files like KITTI format (KITTI = 2D from image + 3D from point cloud).

KITTI format

Can someone please let me know which annotator tools should I go for and how to make the text file? Does that mean I need to label image and pcd separately and write the text file by combining image json information and point cloud json information?

I tried Supervisely and I got json but not enough parameters like KITTI format.

my json

Upvotes: 3

Views: 1075

Answers (1)

Christofel Ganteng
Christofel Ganteng

Reputation: 1

Because KITTI Dataset format using 2D Bounding Box and 3D Bounding Box Coordinates you need to label 2D Bounding Box Coordinates and 3D Bounding Box Coordinates... Then you can make KITTI Dataset format by making label 2D Bounding Box Coordinates and 3D Bounding Box Coordinates..

Usually I using LATTE Annotation Tools for Bounding Box Labelling Object Detection... I use LATTE Annotation Tools for Labelling 3D Bounding Box because LATTE Annotation Tools easy to use and easy to install...

You can install LATTE Annotation Tools in Github here :

https://github.com/bernwang/latte

By the way KITTI Dataset Label uses 3D Bounding Box Camera Coordinates to Label 3D Bounding Box... So you have to change 3D Bounding Box LiDAR Coordinate to 3D Bounding Box Camera coordinates...

Hope this answer can help you make 3D Bounding Box KITTI Dataset.. If there is more advice or problem in making Bounding Box Coordinate kindly chat me through Stack Overflow or my email at [email protected] ...

Upvotes: 0

Related Questions