DeshiA: Open-Source Image Annotation Workstation
DeshiA is an open-source image annotation workstation designed for structured computer vision dataset creation. Inspired in part by CVAT, it provides a professional annotation workflow through both a web interface and a Tauri desktop application, with local filesystem storage and SQLite-based persistence.
DeshiA was developed to provide a dedicated, annotation-focused environment for creating structured computer vision datasets without requiring cloud-based data processing. The initial workflow focuses on Bangladeshi Rickshaw and E-Rickshaw imagery. Instead of limiting annotation to image-level classes, DeshiA supports class-specific views and component-level bounding boxes for important structural, mechanical, and electrical components. The application supports persistent annotation progress, validation, recovery, semantic bounding-box colors, structured dataset organization, and Pascal VOC export. The project is fully local, with the same application available through a browser-based interface and as a Tauri desktop application.
Method
DeshiA follows a schema-driven annotation workflow: Configure source and output directories. Scan and register the image dataset. Persist dataset state using SQLite. Select the vehicle class and view. Annotate predefined semantic components using bounding boxes. Assign deterministic semantic colors to different components. Validate required annotations and visibility states. Persist annotation progress and support recovery. Export structured annotations and organize dataset outputs. Continue automatically to the next pending image. The initial ontology includes Rickshaw and E-Rickshaw, with Front, Side, and Back views and view-specific semantic components. The development process was also AI-assisted, with AI coding tools used for implementation support, debugging, architecture exploration, documentation, and iterative development. AI is a development aid for the project, not a component of the DeshiA annotation system itself.
Deshi Annotation, an extensible image annotation workstation for structured computer vision dataset creation, hierarchical labeling, bounding-box annotation, and reproducible dataset generation.