Spatial Grounding

Enable Grounding

Enable per-entity bounding box annotations for spatial grounding. Requires .bbox sidecar files alongside images.

Auto-Detect Bounding Boxes

Enable Auto-Detect

Automatically generate .bbox sidecar files using Florence-2. Images that already have .bbox files will be skipped.

Auto-detect requires a local backend.

Model

HuggingFace Florence-2 model ID.

Labels (optional)

Comma-separated labels for guided detection. Leave empty for automatic captioning and grounding.

Batch Size

Number of images per inference batch.