CLIP score

A reference-free measure of image–text compatibility computed as the cosine similaritycosine similarity, n.: A score from −1 to 1 for how closely two lists of numbers point the same way, ignoring their size. It is the standard test of whether two texts mean similar things. between the image embedding and the text embedding produced by a CLIPCLIP, prop. n.: An OpenAI model that learned to match pictures with their captions from hundreds of millions of web examples, so it can judge how well any piece of text describes any image. model, usually rescaled (in the original CLIPScore, multiplied by 2.5 with negative values clipped to zero). First proposed for judging image captionsimage captioning, n.: Having a computer write a short description of a picture, such as 'a dog catching a frisbee in a park', to help blind people, search engines and other AI systems. without reference captions, it became a standard automatic measure of prompt adherence for text-to-imagetext-to-image model, n.: An AI system that creates a picture from a written description, such as 'a fox reading a newspaper, in watercolor', usually within seconds. and videotext-to-video model, n.: An AI that creates a short video clip — increasingly with matching sound — from a written description, such as 'a paper boat sailing down a rainy street'. generation, typically averaged over a benchmark set of prompts and reported alongside FIDFréchet inception distance, n.: A score for image generators that compares the overall statistics of their pictures with those of real photos, as seen by an image-recognition network; lower is better. to show the fidelity–alignment trade-off of classifier-free guidanceclassifier-free guidance, n.: A setting that makes an image generator follow the prompt more strictly by comparing what it would draw with and without the prompt, then exaggerating the difference.. It inherits CLIP's weaknesses — poor sensitivity to counting, spatial relations, negation and attribute binding, and susceptibility to images that simply contain the prompt's words as text — so modern evaluations add question-answering-based metrics, learned human-preference models and vision-language-model judgesLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers..