video-intelligence / detection.py

Commit History

5 days ago

  1. Format prompts per model: bare categories for Grounding DINO and OmDet-Turbo, 'a photo of' queries for OWLv2, articles on typed names

    Zmu
    Verified signature
    6d66381
  2. Grounding DINO scored per class from token logits; label presets (default, coco, objects365, openimages) with upload and typed overrides; Advanced nested in Detection settings and Tracking

    Zmu
    Verified signature
    6fd109f

6 days ago

  1. Split backend into hub.py, detection.py, captioning.py; zero-shot labels from a JSON vocabulary (labels/default.json) with typed override

    Zmu
    Verified signature
    b5a226e