Published on:
15 August 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Chen Shuai,
Meng Fanman,
Zhang Runtong,
Qiu Heqian,
Li Hongliang,
Wu Qingbo,
Xu Linfeng
Proposes PGMA-Net, a class-agnostic model for few-shot segmentation
Assembles masks from visual and textual priors via affinities
Avoids bias towards base classes, excels on novel classes
Achieves SOTA on PASCAL-5i and COCO-20i 1-shot tasks
Can do zero-shot segmentation, co-segmentation without retraining
Mask assembly for few-shot segmentation
This paper proposes a new approach called Prior Guided Mask Assemble Network (PGMA-Net) for few-shot segmentation. It uses a class-agnostic process to assemble masks from visual and textual priors via affinities. This avoids bias towards base classes and allows the model to perform well on novel classes. The model achieves state-of-the-art results on PASCAL-5i and COCO-20i datasets in 1-shot scenario. It can also do related tasks like zero-shot segmentation and co-segmentation without retraining.
Foundation models for weakly supervised segmentation
Aligning model objectives for zero-shot segmentation
Principal Mask Proposals for Unsupervised Semantic Segmentation
Simple open-vocabulary image segmentation
Lightweight clustering for semantic segmentation
Point-supervised panoptic segmentation with optimal transport
No comments yet, be the first to start the conversation...
Sign up to comment on this paper