in roboflow’s auto distill library, when we define caption ontology, can we have a many to one mapping, something of this sorts
[“striped shirt”,“black and white stripes”, “person not wearing team jersey”,“wearing striped uniform”]: “referee”
- Project Type: Personal
- Operating System & Browser: Personal device
- Project Universe Link or Workspace/Project ID: NA
- Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No): NA
Hi @Sharanya_Anil
Yes, many-to-one mapping is fully supported in Autodistill’s CaptionOntology.
While you might expect a list-to-string mapping, the actual syntax is a flat dictionary (Dict[str, str]). You simply map multiple unique keys (prompts) to the same value (class label).
Each prompt is sent independently to the base foundation model (e.g., Grounding DINO). Any detection triggered by any of those prompts is automatically labeled as "referee" in the final output dataset.