An Implementation of the “Guess Who?” Game Using CLIP

CLIP (Contrastive Language-Image Pretraining) is an efficient method for learning computer vision tasks from natural language supervision that has powered a recent breakthrough in deep due to its zero-shot transfer capabilities. By training image-text pairs available on the internet, model transfers non-trivially most without need any data set specific training. In this work, we use implement engine of popular game "Guess who?", so player interacts with using prompts and automatically decides whether image board fulfills prompt or not. We study performance approach by benchmarking different ways prompting questions CLIP, show limitations capabilites.