Moondream Pointer

Tools

• moondream_pointer_detect Open-vocabulary object localization and visual question answering with the Moondream vision-language model. The shape of the returned JSON depends on the "mode" argument. In "point" mode it returns a JSON list with one entry per located instance of "object_name"; each entry is an object {"x": float, "y": float} giving that instance's center as coordinates NORMALIZED to 0.0-1.0 of the input image (x is fraction of width from the left edge, y is fraction of height from the top edge). In "object_detection" mode it returns a JSON list with one entry per located instance; each entry is a bounding box {"x_min": float, "y_min": float, "x_max": float, "y_max": float} where all four values are NORMALIZED to 0.0-1.0 of the ORIGINAL input image ("x_min"/"y_min" is the top-left corner, "x_max"/"y_max" is the bottom-right corner) -- this is the albumentations "albumentations" bounding-box format (normalized [x_min, y_min, x_max, y_max]), documented at https://albumentations.ai/docs/3-basic-usage/bo

No tool matches.

As the server's own listing gives them. We never call a tool.