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