Annotation tools are the first step in processing raw data. Whether it is detection tasks, segmentation tasks, 3D perception, point clouds, etc., it is necessary to make real values to supervise network learning. Enterprise-level labeling solutions are generally completed through internal self-developed tools or professional labeling teams. For individuals or small teams, an open source and easy-to-use labeling tool is very important. This article summarizes commonly used tools in the field. Annotation tool, involving 2D detection segmentation/3D detection segmentation and multi-sensor calibration synchronization.
For computer vision researchers and staff, it is very practical. It is recommended to collect and pay attention so as not to get lost.
Detection Segmentation Calibration
1. Label me
Project address : https://github.com/wkentaro/labelme
Main support:
-
Polygon segmentation, semantic segmentation, 2D frame, line annotation, point annotation (can be used for target detection, image segmentation and other tasks)
-
video annotation
Mark the saved file as JSON, VOC and COCO formats, etc.;
2.LabelImg
Project address : https://github.com/heartexlabs/labelImg
There are three options for saving annotation files: PASCAL VOC, YOLO, and CreateML, which only support data annotation for target detection tasks.
3. BLOOM
Project address : https://github.com/openvinotoolkit/cvat
Free online interactive video and image segmentation and annotation tools;
4. WITNESS
Project address : https://github.com/microsoft/VoTT
A JavaScript-based annotation tool for image object detection released by Microsoft, developed using React+Redux, supports Windows and Linux platforms. The software also provides a method of automatic labeling and manual correction based on the faster-rcnn model trained by CNTK, which can greatly reduce the workload required for labeling.
5. EISeg
EISeg (Efficient Interactive Segmentation) is an efficient and intelligent interactive segmentation and labeling software developed based on Paddle. It covers high-quality interactive segmentation models in general, portrait, remote sensing, medical, video and other directions. In addition, by applying the annotations obtained by EISeg to other segmentation models provided by PaddleSeg for training, a high-precision model of a customized scene can be obtained, and the whole process of segmentation tasks from data annotation to model training and prediction can be opened up.
Old version :
https://github.com/PaddleCV-SIG/EISeg
New version : https://github.com/PaddlePaddle/PaddleSeg/tree/release/2.6/EISeg
main feature:
-
An efficient semi-automatic labeling tool, which has been launched on multiple Top labeling platforms
-
Covering remote sensing, medical, video, 3D medical and many other vertical scenarios
-
Compatible with multiple platforms, easy to use, and supports multi-category tag management
img
6. RHYTHM
Samsung open source interactive calibration tool RITM : https://github.com/saic-vul/ritm_interactive_segmentation
For interactive click-based segmentation, the model uses the segmentation mask output by the pre-trained inference model, which not only segments a completely new object, but also starts from an external mask and corrects it;
Multi-Sensor Calibration
1.OpenCalib
Project address : https://github.com/PJLab-ADG/SensorsCalibration
SenseTime's open source toolkit supports calibration between camera, lidar, imu, and radar:
The calibration tool supports checkerboard, circle, square, etc.!
3D detection and calibration
1.point-cloud-annotation-tool
Project address : https://github.com/springzfx/point-cloud-annotation-tool
It is mainly used to calibrate the box information of the target in the 3D point cloud, and supports exporting KITTI format and Apollo 3D format!
2.annotate
Project address : https://github.com/Earthwings/annotate
Based on ROS frame calibration, generate 3D detection frame, xyz and whl information;
3.3D BAT
Project address : https://github.com/walzimmer/3d-bat
It can be used to calibrate 3D targets such as cars, trucks, motorcycles, self-driving cars, pedestrians, etc., and supports any target labeling greater than 10 points!
4.SUSTechPOINTS
Project address : https://github.com/naurril/SUSTechPOINTS
Main support:
-
9 degrees of freedom box editing
-
Object Type/ID/Attribute Editing
-
Interactive/Automatic Cuboid Fitting
-
Batch mode editing
-
Perspective/Projection ViewEdit
-
Multiple camera images, can automatically switch cameras
-
Binary/pcd file of point cloud data
-
Jpg/png image file
-
Object/Box/Point Shading
-
Focus mode, hide the background, easy to see details
-
stream play/stop
-
Object ID generation
3D point cloud segmentation calibration
1.semantic-segmentation-editor
Project address : https://github.com/Hitachi-Automotive-And-Industry-Lab/semantic-segmentation-editor
A web-based tool that supports image and point cloud formats for segmentation and annotation of 2D and 3D data!
