Hannah Schieber∗(1), Fabian Duerr∗(2), Torsten Schoen(3), and Jurgen Beyerer(2,4)

• (1) Human-Centered Computing and Extended Reality, Friedrich-Alexander University (FAU) Erlangen-Nurnberg, Erlangen, Germany
• (2) Vision and Fusion Laboratory, Karlsruhe Institute of Technology, Karlsruhe, Germany
• (3) Research Institute AImotion Bavaria, Technische Hochschule Ingolstadt, Ingolstadt, Germany
• (4) Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB), Fraunhofer Center of Machine Learning, Karlsruhe

(*) equal contribution, contact e-mail: hannah.schieber[at]fau.de

Abstract

Robust environment perception for autonomous vehicles is a tremendous challenge, which makes a diverse sensor set with e.g. camera, lidar and radar crucial. In the process of understanding the recorded sensor data, 3D semantic segmentation plays an important role. Therefore, this work presents a pyramid-based deep fusion architecture for lidar and camera to improve 3D semantic segmentation of traffic scenes. Individual sensor backbones extract feature maps of camera images and lidar point clouds. A novel Pyramid Fusion Backbone fuses these feature maps at different scales and combines the multimodal features in a feature pyramid to compute valuable multimodal, multi-scale features. The Pyramid Fusion Head aggregates these pyramid features and further refines them in a late fusion step, incorporating the final features of the sensor backbones. The approach is evaluated on two challenging outdoor datasets and different fusion strategies and setups are investigated. It outperforms recent range view based lidar approaches as well as all so far proposed fusion strategies and architectures.

Quantitative Results

Results on SemanticKitti

Results on PandaSet

Code

Please finde the Code here: GitHub

pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

If you build upon our code please cite our work:

@inproceedings{schieber2022deep,
  title={Deep Sensor Fusion with Pyramid Fusion Networks for 3D Semantic Segmentation},
  author={Schieber, Hannah and Duerr, Fabian and Schoen, Torsten and Beyerer, J{\"u}rgen},
  booktitle={2022 IEEE Intelligent Vehicles Symposium (IV)},
  pages={375--381},
  year={2022},
  organization={IEEE}
}

Part of the Code is borrowed from these repositories EfficientPS and PSPNet. Furthermore, it builds upon the fusion idea of Duerr et. al:

@INPROCEEDINGS{9287974,  author={Duerr, Fabian and Weigel, Hendrik and Maehlisch, Mirko and Beyerer, Jürgen}, 
	 booktitle={2020 Fourth IEEE International Conference on Robotic Computing (IRC)},  
	 title={Iterative Deep Fusion for 3D Semantic Segmentation},   
	year={2020},  
	volume={},  number={},  pages={391-397},  doi={10.1109/IRC.2020.00067}
}

if you cite our work please also consider the previous approach Iterative Deep Fusion for 3D Semantic Segmentation.

Talk from IEEE IV 2022