Each image moves through a dedicated stage. Specialists pick up where the previous role left off — no context-switching, no email chains.
Draw, drag, and resize bounding boxes over regions of interest. Keyboard shortcuts (V / D / E) keep your hands on the canvas.
DINOv3 suggests a top-K taxonomy in real time. Accept the AI label or pick from the taxonomy panel — batch-classify a whole image in one click.
Drill down to species level. Add identification notes, confidence flags, and finalise observations one box at a time.
Finalised images export to clean JSON — ready to feed downstream models, papers, or shared with collaborators.
Top-K AI suggestions per box, batch-classify entire images, toggle auto-suggest off when you want pure manual control.
Draw, drag, resize from any handle. Per-box opacity slider keeps the underlying imagery legible while you work.
Annotators, classifiers, researchers, viewers, and admins each see only the work that needs them — with per-role queue counts.
Drop images directly onto the gallery page to upload. Files go straight to cloud storage — no extra steps, no page reload.
Slice the gallery by cluster, taxonomy category, or pipeline stage. Find the next image to work on in seconds.
Researchers and admins export finalised annotations as clean JSON — directly consumable by downstream ML pipelines.
V to select, D to draw, E to erase, Esc to deselect, Delete to remove. The canvas stays under your cursor.
A live per-image breakdown of category counts and finalisation status, so researchers always know what is left.
Permissions are enforced end-to-end. Each user sees only the controls that belong to their role, keeping the pipeline focused and the data clean.
Five top-level categories, color-coded across every panel and overlay, extensible from the admin dashboard.
Sign in with your lab account to start uploading. Your team's pipeline, queues, and taxonomy are pre-configured.