0. Introduction
For 3d bounding box, with the popularity of autonomous driving in recent years, there are more and more annotation tools. This article does not talk about specific algorithms. Here we mainly focus on these open source 3d bounding box annotation tools and their how it is used. Here I borrowed I want to be quiet, the blogger's blog as the basis, and then combined with my own use and understanding to complete the expansion.
1. 3d-bat
In this paper, we focus on obtaining 2D and 3D labels, as well as trajectory IDs of objects on roads, with the help of the novel 3D Bounding Box Annotation Toolbox (3D BAT). Our open source web-based 3D BAT includes several smart features to improve usability and efficiency. For example, this annotation toolbox supports semi-automatic labeling of trajectories using interpolation, which is crucial for downstream tasks such as tracking, motion planning, and motion prediction.
Furthermore, annotations for all camera images are automatically obtained by projecting annotations from 3D space into the image domain. In addition to raw image and point cloud feeds, a main view consisting of a top view (bird's eye view), side view, and front view is provided, which can be used to view objects of interest from different angles. A comparison of our method with other publicly available annotation tools shows that 3D annotations can be obtained faster and more efficiently using our toolbox. Supports point cloud rendering, 2D top view, and cube annotation. The following article describes the specific usage and installation methods
2. point-cloud-annotation-tool
In order to annotate point cloud data, I finally chose this point-cloud-annotation-tool, which can be compiled and installed under Windows, based on QT, vtk and PCL. It mainly supports the following annotation formats
- Support point cloud data loading, saving and visualization
- Support point cloud data selection
- Support 3D BOX frame generation
- Support KITTI-bin format data
The following article talks about the specific usage and installation method under Windows, and for Ubuntu users, you can refer to this article
3. SUSTechPoints
After obtaining the image data, lidar data, and calibration data, it is necessary to mark the 3D target frame. The labeling tool used in this paper is: SUSTechPOINTS. This tool is included in IEEE 2020, and it is a relatively good open source project, which can enable the joint labeling of lidar and image data. It can mainly complete point cloud rendering, 3-view, 2D camera map, smooth interaction, semi-automatic frame annotation (additional package required), automatic object tracking ID generation, specific installation and usage can be found in this article
4. annotate
Evaluation of perception algorithms requires labeled ground truth data. Supervised machine learning algorithms require labeled training data. "annotate" is a tool for creating 3D labeled bounding boxes in ROS/RViz. Its labeled data can be used both as training data for machine learning algorithms and as base data for evaluation. For specific usage, please refer to this article
5. Label family
5.1 Labelhub
Labelhub is a labeling software that supports labeling on the web. It supports all types of image annotation, supports 3-view mapping of 2D images (.jpg and .png files). It also supports the labeling of targets in the 3D point cloud (.pcd file) generated by LIDAR, and supports the generation of 3D BOX frames. The best thing to use is that it can perform automatic target recognition, welting, etc. through the preset training model. The free trial address is as follows http://labelhub.cn/ . If you need to use this software specifically, you can refer to this blog
5.2 labelImg-kitty(bev)
Using this software can directly provide the labeling of the bev perspective and get the kitti format. This is an expansion of the traditional IabelImg, adding the labeling of the depth point cloud under the bev perspective. For specific usage, please refer to this article .
