Object detection tooling
DataSet Pro
Make bounding box annotation understandable and local, from opening an image folder to exporting labels in the coordinate format a training pipeline expects.
- 01INPUTA local folder of images
- 02SYSTEMA bounding box and label workspace
- 03OUTCOMEPortable YOLO or Pascal VOC labels
Repository grounded case study. No staged product demo.
WHY IT EXISTS
The product starts with the problem.
Dataset preparation often hides simple geometry behind large cloud platforms. DataSet Pro keeps the images on the user’s machine and makes boxes, classes, shortcuts, and coordinate conversion part of one focused desktop workflow.
WHAT IT IS MEANT TO DO
Three jobs, one clear promise.
These are the product outcomes the repository is organized to deliver.
Annotate without uploading
Users choose a local folder, move through its images, and create or resize boxes directly on the source dimensions.
Keep classes reusable
Labels carry names, colors, and keyboard shortcuts so repeated annotation work stays consistent.
Translate coordinate systems
The same project can save normalized YOLO coordinates or pixel based Pascal VOC bounds.
PRODUCT FLOW
How the idea becomes useful.
A short path from the user’s starting point to a useful result.
- 01
Open an image folder
The desktop process reads supported images and remembers the last working folder locally.
- 02
Define the vocabulary
Create label classes with colors and shortcuts before or during annotation.
- 03
Draw and refine boxes
Boxes can be created, selected, moved, resized, and adjusted with zoom, pan, and keyboard controls.
- 04
Save the required format
Per image labels and project metadata are written beside the dataset for later training or conversion.
ENGINEERING IDEA
The choices that define the product.
Architecture matters here because each constraint changes what a user can safely believe or do.
A narrow desktop bridge
Electron’s context bridge exposes specific folder, file, label, setting, and conversion operations to the renderer.
Geometry remains visible
YOLO normalization and Pascal VOC pixel bounds are derived from the actual image size instead of hidden behind an upload pipeline.
Folder based ownership
Labels and project metadata live with the user’s files, so the dataset is portable without a vendor account.
REPOSITORY PROOF
Evidence, not feature theatre.
Each point below comes from the current repository structure, code, or documented reference data.
Export formats
YOLO and Pascal VOC cover normalized and pixel based bounding boxes.
Image handling
Selected source images remain on the machine running the desktop app.
Application shell
A preload bridge separates renderer interactions from file system work.
HONEST BOUNDARIES
What this product is not.
- DataSet Pro focuses on bounding boxes for object detection, not segmentation or model training.
- Windows is the primary packaging target documented by the repository.
- The project works with local files and does not provide cloud collaboration or hosted dataset storage.