Table of Contents
- LabelMe General Introduction
- Data Annotation Tool Comparison
- LabelMe Analysis
- User Interface
- Output Format
- How to Use LabelMe
LabelMe – General Introduction
1. Description: A web-based open graphical image annotation tool (Github Location: https://github.com/wkentaro/labelme)
2. Price: Free
- upports image annotation for polygon, rectangle, circle, line and point, and also image flag annotation for classification and cleaning.
- The format is JSON
4. Project management:
- It has virtually no project management properties but it does allow an easy way to import and visualize annotations and correct them if necessary.
- The simple offline interface makes the annotation process pretty fast, even though it does not support many hotkey shortcuts.
- Stable and easy to use, you can access the tool from anywhere and people can help you to annotate your images without them having to install or copy a large dataset onto their computers
- You could extract segmentation masks
- Doesn’t support team coordination
- Doesn’t support real-time annotation performance monitoring and quality check
- Need to distribute and collect statistics manually, and it increases operational cost
Comparison with Other Annotation Tools
LabelMe – User Interface
LabelMe – Workflow
LabelMe – Output Format
Step 1: Dataset Preparation
Split your data
Split your dataset into 3 Folders, namely “Training”, “Validation” and “Test”
Step 2: Class Name Preparation
Type all the Class Names (Labels) to be annotated in the “Labels.txt” file
- The “Labels.txt” file comes with the installation of LabelMe
- Keep “__ignore__” and “background” classes unchanged as the first and second
- When naming the Classes, avoid using “-” as the “-” mark will be later used to distinct instances.
Fire up with User Interface using the following command
- LabelMe [–labels labels.txt] [directory | file]
Step 3: Do Annotation
Press “Create Polygons” button then start drawing
Step 4: Name the Polygon
Pick Class Name from your predefined Class Name list
To create instance segmentation, you could manually add an instance ID after the Class Name
Step 5: Edit Polygon
To edit the shapes you created, you could click “Edit” Button.
Step 6: Save
When you have finished annotating all objects listed in “Label List” in the image, click “Save” to save .json file.