mirage: Extract structured data from a single image

Mirage, released by Cactus Compute, is a vision model designed for image understanding and structured data extraction. By inputting a single photo file, it converts the image's content into text and returns it. Distributed under the Apache 2.0 license, it can be freely used in commercial projects.

Beyond simple image description, it performs various visual tasks such as object detection, image segmentation, image classification, visual question answering (VQA), and document question answering (DQA). In particular, it offers high flexibility by recognizing new objects or classes without training data through a zero-shot approach.

Installing the Python package `cactus-compute` automatically downloads and caches the engine and weights. Inference is available immediately by simply calling the `describe` method, allowing for rapid prototyping without the need to configure complex preprocessing pipelines.

It supports 7 languages and is actively used in the community, with 12,844 downloads in the past month. Its inclusion in the Automation Foundation Models For Tiny Devices collection suggests its potential for use in edge device environments.