Databox

Databox is a localhost-first computer vision dataset labeling tool built with React, TypeScript, and Vite. This repository is a fork and reboot of the Make Sense codebase, reshaping the existing annotation foundation into a folder- native dataset workflow with editable JSON annotations and planned YOLO export.

Databox is forked from and inspired by SkalskiP/make-sense. The upstream project is licensed under GPL-3.0, and Databox preserves that attribution.

What this repo is today

• React + TypeScript + Vite frontend application.
• Local browser app for image annotation.
• Editable JSON annotations are the intended source of truth.
• The cleanup phase is focused on rebrand, docs, and repo hygiene.
• Folder-native dataset persistence and export workflows are planned next.

What this repo is not today

• Not a finished dataset manager.
• Not a SaaS product or cloud service.
• Not a marketing landing page.
• Not a Next.js app.

Current status

Databox is in Phase 1: Cleanup and repo documentation. The current app continues to behave like the inherited annotation foundation while the project identity, docs, and package metadata are updated.

Local setup

git clone https://github.com/Danncode10/databox.git
cd databox
npm install
npm run dev

Open:

http://localhost:3000

Available commands

npm run dev
npm start
npm run build
npm run lint
npm run typecheck
npm test

TypeScript typecheck support is now available via npm run typecheck.

Target dataset structure

datasets/
  sample-dataset/
    raw/
      image_001.jpg
      image_002.jpg
      image_003.jpg
    annotations/
      image_001.json
      image_002.json
      image_003.json
    images/
      train/
      val/
      test/
    labels/
      train/
      val/
      test/
    data.yaml
``` m

Notes:

- `raw/` stores original imported images and extracted keyframes.
- `annotations/` stores editable JSON annotation files.
- `images/`, `labels/`, and `data.yaml` are generated export outputs.
- YOLO `.txt` files should not be treated as the editable source of truth.
- `datasets/sample-dataset/` can be committed as a demo dataset.
- Other local datasets should remain outside of source control.

## Planned workflow

```text
open or create dataset
-> import raw images
-> label each image
-> save annotations/<image-name>.json
-> review completed images
-> export YOLO training data

Roadmap

The current detailed plan is in MASTERPLAN.md.

Phase 1 cleanup tasks:

• [P1.1] Inspect baseline and fix repo hygiene
• [P1.2] Rebrand package, app identity, and upstream references
• [P1.3] Rewrite README, instructions, and planning docs
• [P1.4] Clean GitHub metadata and dataset ignore policy
• [P1.5] Modernize package baseline and verify cleanup

Verification notes for Phase 1 cleanup are available in docs/tests/.

Future phases:

• Dataset chooser UI
• Local dataset storage layer
• Folder-native editor workflow
• YOLO export pipeline
• Video keyframe workflow

Thesis context

Databox is designed to support a computer vision thesis workflow focused on local dataset labeling, annotation quality, and export-ready training data.

Example pipeline:

drone or camera captures images or video
-> keyframes/images are stored in raw/
-> visible objects are labeled with bounding boxes
-> annotations are saved as JSON
-> YOLO-ready dataset is exported
-> model training uses the generated files

Attribution

Databox is forked from and inspired by Make Sense:

@MISC{make-sense,
  author = {Piotr Skalski},
  title = {{Make Sense}},
  howpublished = "\url{https://github.com/SkalskiP/make-sense/}",
  year = {2019},
}

License

This project is licensed under the GPL-3.0 License. See LICENSE for details.