nlabel

WebApp for creating dataset of labeled images

Usage:

Start the webapp
$ npm start

1. Access localhost:3000 from your browser
2. Enter absolute path of directory of images for which labels are to be made
3. Set image dimensions
4. Click "+class" to create a labeling category of object
5. Click "+label" below a class to add a new label name-value set for that class
6. Click "Next" to save configuration in public/data/config.json
7. You will be redirected to /gallery where you can set labels
8. Select a class from dropdown on sidebar. Label set will change according to selected class after page reloads
9. Draw bounding box over a region of interest. Set label value from drop down or enter a custom value
10. Click "Save current box". Data from bottom texarea will be written to public/data/outputs/image_name.json
11. Browse through images by pressing Next/Previous
12. If you perform any undo/redo/discard operation, you must click on "Save All" to commit changes to existing file

Changelog:

18/04/2018

• First release v0.0.1
• Fully functional but with least error checks

13/05/2018

• Complete revamp
• Define classes and labels with their default values
• Create bounding box and label each box instead of entire image
• Save annotations of every image in a separate file
• Symlink user directory instead of individual files
• Dataset stored as .json files stored in public/data/outputs
• Store aboslute coordinates of bounding box with "location" as key and relative coordinates (wrt scaled image) with "scaled" json key

Todos:

• Choose images directory by browsing local filesystem
• Set filename of dataset before downloading
• Manage user sessions to resume dataset creation
• Customize input type of each label
• Automatically set a default input type on basis of label name