What did the Founder of KITTI Dataset (Chief Scientist of Surfing Tech) Shared with us?

Surfing Tech

Andreas Geiger, the chief scientist of Surfing Tech, stayed in Beijing for a short time. He gave an academic report in Tsinghua and then went to Surfing. He made a black technology for all Surfers to automatically convert the annotation of 3D bounding boxes in KITTI dataset and 3D space into the result of semantic segmentation and annotation in 2D images. We should know that the annotation of Lidar 3D bounding boxes in lidar radar is in its infancy. Is it surprising and unexpected that the result of semantics segmentation can be automatically converted into 2D images, which is the exclusive original of Dr. Geiger's laboratory?


Programmers all know that the semantic segmentation of vehicle-borne video images is very important. Common data sets such 


as city scape, but the cost of annotation is very high, which is to annotate 2D images. (Cityscape provides multiple stereo video sequences recorded from 50 different urban street scenes)


Lidar is very important in the field of unmanned driving. For the use of 3D street view data collected by lidar, the annotation of enclosure box in 3D space is very important. Dr. Geiger's KITTI dataset includes this direction and has achieved remarkable results.


Andreas Geiger Laboratory has a deep research on the automatic conversion of 3D bounding box labeling results into semantic segmentation structures in 2D images. See the article on CVPR16 for details. This technology can greatly improve the efficiency of 2D semantic annotation.


However, there are still two problems in this black technology:

(1) How to label 3D bounding boxes in 3D space requires manpower and perfect labeling process and specifications to control quality;

(2) Although the results of automatic conversion are of high accuracy, there are still some errors, which need manual modification to be used as ground Truth, which still needs manpower to do.


Surfing Tech and Dr. Geiger's laboratory are working together in these two directions.

Contact us if you need us.

Request a Demo