Object Detection Studio (Pro)

Tabs

• Capture — pick a target window and hotkey, collect screenshots into a dataset. Left-side list shows each capture; stats show total / labelled / unlabelled / train-val split / class count.
• Annotations — define classes on the right, draw boxes on the canvas and pick a class per box. Pan / zoom the image. Verify saved overlays the saved boxes to confirm. Click-assist: click an object once and a helper model draws the box (SAM 2). Auto-detect: scans all unlabelled captures and suggests boxes.
• Pre-Training — model size (small / medium / large), image size, epochs, batch size (or Auto). Export dataset packs images, boxes, class list, settings and a training notebook into a single ZIP. Fast transfer via Google Drive for Desktop is optional.
• Training — step-by-step for Colab (upload notebook, switch to GPU, Run all, download best.onnx, put in Models). "Open Colab in browser" opens Colab in your normal browser, reusing your Google session.
• Models — activate a trained model and live-test it on a chosen window. Controls: CPU/GPU inference, confidence slider, class filter, size filter (two independent lists: include and ignore), Start/Stop the live test.
• Settings — dataset location, capture-flow hotkeys, dataset folder rescan, and a Clear All Annotations action with a confirm.