0.Introduction
Automatic annotation software is a very labor-saving operation, and with the development of deep Xi, these automatic calibration software are also increasing. This article will focus on the more classic auto-annotation software.
1. autolabelimg
In addition to the initial functions of LabelIMG, AutoLabelIMG contains more than 10 auxiliary annotation functions, which are under the two new menu bars of annoatate-tools and video-tools, as follows:
Automatic annotation: The model based on yolov5 is automatically annotated and converts the detection results of yolov5 into . XML annotation file.
Tracking annotation: Automatic annotation based on the tracking module of OpenCV, the first frame of annotation, and the subsequent segment of tracking is used.
Magnifier: The magnified display of the area near the mouse is convenient for marking some small targets, and the magnifying glass function can be turned on or off.
Data Enhancement: Randomly enhance with panning, flipping, zooming, brightness, gama, blur, and more**.
Query system: more than 10 new functions, I don't know what it do?It doesn't matter, just search for it, fuzzy search is supported.
Other batch processing tools such as: category filtering, renaming statistics, annotation file attribute correction, extraction and synthesis, renaming, etc., welcome to get a detailed introduction through the query system in the software, or directly get started.
2. x-anylabeling
X-AnyLabeling is a new interactive automatic annotation tool, which is built and redeveloped based on AnyLabeling, on this basis, it expands and supports many models and functions, and provides powerful AI support with the help of mainstream models such as Segment Anything and Yolo.
Image annotation of polygons, rectangles, circles, lines, and points is supported.
Supports text detection, recognition, and KIE (Key Information Extraction) annotation.
Detection-classification cascade models are supported for fine-grained classification.
Support one-click face and key point detection function.
Mainstream deep learning Xi frameworks such as PaddlePaddle, OpenMMLab, and Pytorch-TIMM are supported.
It can be converted to standard COCO-JSON, VOC-XML, and YOLOV5-TXT file formats.
Advanced detectors are available, including YOLOV, YOLOX6, YOLOv7, YOLOv8, YOLOX, and DETR series models.
3. labeltrack
LabelTrack is an automatic annotation tool written for multi-object tracking MOT, which can complete fast pedestrian annotation by importing ** streams and other operations.
Import MP4 files or frame folders, manually annotate and modify the labeling boxes, including size, labels, IDs and other information, and use the SOTA object tracking model to pre-track the frames, export and import the Visdrone format dataset
4. plabel
This tool is independently developed by Pengcheng Laboratory, integrating algorithms such as frame extraction, target detection, tracking, REID classification, face detection, etc., to realize the automatic annotation of images, and can manually annotate the results of automatic algorithms, and finally obtain annotation results.
At the same time, it can also manually annotate the data related to ** and medical (including DICOM files and pathological images), and the annotation results support COCO and VOC formats.
Multi-person collaborative annotation is supported. At present, a new GPU-based segment Anything segment is added to automatically label the image, which can be split arbitrarily.
5. auto-mos
The project can automatically generate training data for lidar-based moving object segmentation. The tool achieves this by processing data offline in batches.
It first utilizes occupancy-based dynamic object removal to roughly detect possible dynamic objects. Second, it extracts segments from the proposal and tracks them using a Kalman filter.
Based on the tracked trajectory, it marks objects that are actually moving, such as driving cars and pedestrians, as moving.
Conversely, non-moving objects (e.g., parked cars, lights, roads, or buildings) are marked as static.
Reference Links: