image_captioner
A Flutter/Dart Android app written by a noob for producing Text files (Captions/Annotations) against a folder of Pictures, with a view to support producing a valid Stable Diffusion Dataset for training
Notes
- The application will prompt you to create a directory that is has permissions to read/write from on first boot; since Android is dumb and won’t let it access anywhere, even with the “Full Files Permission” toggle enabled.
- To make your folder of images available to the app:
- When prompted to select a directory by the app, it will also prompt you to create a new directory. Create that directory
- With the new (empty) directory created, come out of the app, and copy your dataset images into the new directory with something like Google Files or Total Commander etc
- Re-open the app to load your directory of images.
- You will then be able to sift through the images, and create captions/annotations as you go.
- There is no multi-select/batch operations support, or anything fancy really, but this was a decent first slice to learning Flutter/Dart.
- Text files are added next to the selected images, as
<image_name>.txt - Though there is an iOS part to this project, it is unbuilt, and untested
- Currently using Android Studio “Giraffe” for development
- App is only in Dark Mode theme, because i’m too lazy to implement a toggle or anything fancy
Wishlist for Future Features
- Fix the App Name being displayed as
image_captionerin the Apps list. Should be nicer. - Fix the App logo being the default Flutter Icon
- Display the images in a Grid (similar to a regular photo gallery app), and allow batch operations when selecting multiple (eg. Append a Prefix/Suffix tag // add/remove specific tag across all selected images etc)
- Still allow for viewing single images from the grid by selecting one, and then support next/previous as usual
- Make a note of common tags used during an annotation job, and provide auto-complete options and/or a selectable list of tags under the text input to save time
- Be able to read/write where we want, without stupid permissions getting in the way
- Support generating flipped versions of all images in a dataset, and handle the tagging of the flipped duplicates as part of a single process (ie. tags changed on one image should be reflected on the flipped image too, without showing both images to the end user)
- iOS app testing/build
- Github Actions for automated APK builds
- Publish to F-Droid
- Add some tests?
- Validate on multiple Android versions (currently only manually tested against Android 11 + 13)
- Review features of other applications (like https://github.com/toshiaki1729/stable-diffusion-webui-dataset-tag-editor) to see if other features would be good to port to mobile?
- See if there’s any demand to build this out to a Desktop app for Windows/Mac/Browser from Flutter too?
- Add support for deleting images (with confirmation for safety)
- Add basic edit support, specifically crop!
