Completely weakly supervised class-incremental learning for semantic segmentation

doi:10.1016/j.patrec.2025.05.004

Completely weakly supervised class-incremental learning for semantic segmentation

2025-05-28

0

PRE

AI

DOI：10.1016/j.patrec.2025.05.004

Original

Original request for help

Save

Abstract

En 中文

• We present the first weakly supervised class-incremental segmentation method. • It uses only image-level labels to train a network for base and novel classes. • We generate pseudo-labels using a localizer and foundation models. • We present an exemplar-guided data augmentation method.

Completely weakly supervised class-incremental learning for semantic segmentation

Abstract

Journal

Organization