About

Add a batch pipeline that takes

• (a) an image corpus (folder or Parquet of binary images/URIs) and,
• (b) one or more text labels, and returns detection boxes (with scores + optional masks) for each image/label using an open-vocabulary grounding model such as OWLv2

Objective

• Support open-vocabulary text prompts

  • Single label
  • Multiple labels

• Run efficiently on GPU(s) with batch inference

• Emit results in interoperable formats with stable schema

Example

One Label Detection

- RGB Image
- Text Label: ["Fish"]

Multi-labels Detection

- RGB Image
- Text Label: ["coffee mug", "plate", "spoon"]