Generalized Referring Expression Segmentation

Generalized Referring Expression Segmentation (GRES) is a computer vision task introduced by Liu et al. at CVPR 2023. The goal of this task is to handle the correspondence between natural language expressions and multiple target objects in images, predicting the masks of the target objects given an image and a referring expression. The application value of GRES lies in enhancing the naturalness and accuracy of human-computer interaction, especially in multi-object recognition and segmentation in complex scenes.