frankiethull/kuzco/kuzco-recognition

kuzco-recognition

Recognize and locate instances of an object in an image using a local vision language model, exposed as an ellmer tool by the kuzco R package.

When to use

The user wants to find a specific kind of object in an image: faces, cars, dogs, text regions. Counting instances and describing/locating each one. For general "what is in this image?" questions, use kuzco-classification instead.

Setup

library(kuzco)
library(ellmer)
 
chat <- chat_ollama(model = "qwen2.5vl")
chat$register_tool(tool_llm_image_recognition())

Tool arguments (LLM-controlled)

• image (required): local file path to a jpeg, jpg, or png image.
• recognize_object (optional): the object to recognize, e.g. "face", "car", "dog".

Defaults to "face".

• additional_prompt (optional): extra instructions, e.g. "only count adults".
• language (optional): output language, e.g. "English".

Result

Compact JSON text with object_recognized, object_count, object_description, and object_location. Report the count first, then describe/locate instances. If the count is zero, say so plainly rather than hedging.

Example

User: "How many faces are in img/group.jpg and where?"

1. Call kuzco_image_recognition with image = "img/group.jpg", recognize_object = "face".

2. Report the count and describe each location.