CLIPSeg

CLIPSeg is a model that was designed to segment objects in images using text. It works by adding a small decoder to a frozen CLIP model in order to perform zero- and one-shot image segmentation. This model was proposed by Timo Lüddecke and Alexander Ecker in the paper titled ‘Image Segmentation Using Text and Image Prompts.’

Source: https://huggingface.co/docs/transformers/model_doc/clipseg