Comparison of various image captioning models
On September 25, 2023, OpenAI released ChatGPT Vision. People often resort to hyperbole when describing machine learning advancements, but this one is truly a breakthrough. ChatGPT Vision understands images.
Just three months ago, in July, I spent a good deal of time testing image captioning models. Cutting-edge models for the time, like Microsoft Azure Computer Vision, were able to produce fairly reasonable captions most of the time. But ChatGPT Vision blows them all out of the water.
The image displays an abstract artwork or design. It consists of a textured, blue, furry or spiky surface with carved-out patterns resembling flower shapes, revealing a beige or light tan color underneath. The design contrasts the soft or spiky texture of the blue areas with the smooth areas of the carved-out patterns. The entire piece is set against a plain white background.
"Abstract artwork with a textured blue, spiky surface and carved-out flower-like patterns revealing a beige color on a white background."
