Learning-Deep-Learning

MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving

April 2019

tl;dr: Combines object detection, scene classification and road segmentation into one combined net.

Overall impression

The paper comes form Raquel Urtasun’s group, and is one of the best paper with codes for kitti’s road detection. Encoder + 3 task specific decoders for multitask joint training and inference. The best contribution of this paper may be the light weight detector network YOLOv1 as a light weight RPN.

Btw the writing is surprisingly bad with numerous typos.

Key ideas

Technical details

Notes