Open-vocabulary learning has revolutionized image segmentation, enabling the delineation of arbitrary categories from textual descriptions. While current methods often employ specialized architectures, OVAMTSeg presents a unified framework for Open-Vocabulary and Multitask Image Segmentation. Leveraging adaptive prompt learning, OVAMTSeg excels in capturing category-sensitive concepts, ensuring robustness across diverse multi-task scenarios. Text prompts effectively capture semantic and contextual features, while cross-attention and cross-modal interactions facilitate seamless fusion of image and text features. The framework incorporates a transformer-based decoder for dense prediction. Experimental results demonstrate OVAMTSeg's effectiveness, achieving a 47.5 mIoU in referring expression segmentation, 51.6 mIoU on Pascal-VOC with four unseen classes, 46.6 mIoU on Pascal-Context in zero-shot segmentation, 65.9 mIoU on Pascal-5i, and 35.7 mIoU on COCO-20i datasets for one-shot segmentation.
Pan et al. (Mon,) studied this question.