Real3D-Aug: Point Cloud Augmentation by Placing Real Objects with Occlusion Handling for 3D Detection and Segmentation

Real3D-Aug is a open source project for 3D object detection and semantic segmentation.

Official paper is published on arxiv.

Real3D-Aug was proposed as lidar augmentation framework, which reuses real data and automatically finds suitable placements in the scene to be augmented, and handles occlusions explicitly. Due to the usage of the real data, the scan points of newly inserted objects in augmentation sustain the physical characteristics of the lidar, such as intensity and raydrop.

Overview

• Introduction
• Proposed pipeline
• Content
• Getting Started
• Licence
• Publication / Citation
• Acknowledgement
• Contribution

Introduction

Object detection and semantic segmentation with the 3D lidar point cloud data require expensive annotation. We propose a data augmentation method that takes advantage of already annotated data multiple times. We propose an augmentation framework Real3D-Aug.

Proposed pipeline

The pipeline proves competitive in training top-performing models for 3D object detection and semantic segmentation. The new augmentation provides a significant performance gain in rare and essential classes, notably 6.65% average precision gain for “Hard” pedestrian class in KITTI object detection or 2.14 mean IoU gain in the SemanticKITTI segmentation challenge over the state of the art.

As it is shown on image below, the process of augmentation is divided into 4 steps.

1. Preprocessing - In the first step we need to create Rich map if it is not provided. For semantic segmentation dataset we also provide method how to create bounding boxes, which are necessary in further stages.
2. Placing In this stage the possible placements are found.
3. Occlusion handling in spherical coordinates - To ensure the reality of scanning the occlusion is handled.
4. Output - Augmented scene is created and saved.

Content

Project is divided into two parts, semantic segmentation and 3D object detection. Each section contains README.md, where is clearly described how to use the framework.

``` objectdetection README.md config/ cutobject/ Real3DAug/ richmap/ pseudolabels/

semanticsegmentation/ README.md config/ cutobject/ Real3DAug/ richmap/ boundingboxes/

```

Getting Started

Continue to semantic segmentation or 3D object detection part of the project depending on the task you are working on.

Licence

Real3D-Aug is released under the MIT License

Publication

This repository is connected to publication: Real3D-Aug: Point Cloud Augmentation by Placing Real Objects with Occlusion Handling for 3D Detection and Segmentation arxiv. @misc{https://doi.org/10.48550/arxiv.2206.07634, doi = {10.48550/ARXIV.2206.07634}, url = {https://arxiv.org/abs/2206.07634}, author = {Šebek, Petr and Pokorný, Šimon and Vacek, Patrik and Svoboda, Tomáš}, keywords = {Computer Vision and Pattern Recognition (cs.CV), Artificial Intelligence (cs.AI), FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {Real3D-Aug: Point Cloud Augmentation by Placing Real Objects with Occlusion Handling for 3D Detection and Segmentation}, publisher = {arXiv}, year = {2022}, copyright = {Creative Commons Attribution 4.0 International} }

Acknowledgement

This work was supported in part by OP VVV MEYS funded project CZ.02.1.01/0.0/0.0/16 019/0000765 “Research Center for Informatics”, and by Grant Agency of the CTU Prague under Project SGS22/111/OHK3/2T/13. Authors want to thank colleagues from Valeo R&D for discussions and Valeo company for a support.

Contribution

Feel free to contact us for any potential contributions.