Foundation models for weakly supervised segmentation

Paper Title:

Foundation Model Assisted Weakly Supervised Semantic Segmentation

Published on:

6 December 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Xiaobo Yang,

Xiaojin Gong

Bullets

Key Details

•

Proposes method to use CLIP and SAM for weakly supervised segmentation

•

Designs coarse-to-fine framework with learned prompts

•

Applies SAM-based seeding module to generate seeds

•

Achieves state-of-the-art on PASCAL VOC 2012

Explore the topics in this paper

clip

computer vision

deep learning

foundation models

image segmentation

AI generated summary

Foundation models for weakly supervised segmentation

This paper proposes a method to leverage foundation models like CLIP and SAM to generate high-quality segmentation masks using only image-level labels. A coarse-to-fine framework with learned prompts is designed to produce segmentation seeds. Experiments show state-of-the-art performance on PASCAL VOC 2012.