Evaluating Text-Image Alignment using Gecko2K

The Text-to-image(T2I) models are transforming the way images are generated, enabling seamless creation of visuals from text prompts. A critical aspect of advancing these models lies in evaluating how well the generated images align with their corresponding textual descriptions. Gecko2K provides a structured benchmark for systematically assessing this alignment. In this paper we discuss the evaluation of Instruct Pix2Pix model’s ability to alter the images based on its prompt using the Gecko2K framework. Highlighting the areas of improvement that could enhance the image altering capability of the text-to-image models based on its prompts.