zoom-in

name	zoom-in
description	refine ambiguous recognition by zooming into a selected region for closer analysis
characteristics	["very fast local visual inspection tool","produces cropped reference image for reasoning only","helps resolve tiny/occluded target ambiguity"]
when_to_use	["target is small, far, blurry, or partially occluded","model needs closer inspection before choosing object part"]
when_not_to_use	["object and part are already clearly visible","do not pass zoomed image as detection input"]

Inputs:

• image_path (string, required): absolute path to the RGB scene image.
• object_name (string, required): The name of the object to be identified.

Behavior (for the model to follow, no scripts needed):

• When the current recognition result is uncertain, you can select a local area, crop and zoom in on it, and reanalyze only that zoomed-in area.
• Use the zoomed-in view to refine which object/part it is and update your reasoning before calling other skills (e.g. rex_omni, sam2).

Outputs (conceptual, no actual file I/O required):

• refined_hypothesis (string): updated, more confident description of the object or part in the zoomed region.