Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Dynamic Category Queries Transformer for Generalized Few-shot Semantic Segmentation

Huang, Kunze ; Yang, Jieyuan ; Jakobsson, Andreas LU orcid ; Tang, Luyao ; Tu, Xiaotong ; Ding, Xinghao and Huang, Yue (2025) 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025 In ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
Abstract

Few-shot segmentation (FSS) tackles data scarcity using multiple priors, but its simplicity limits handling base and novel classes with limited data access. Generalized few-shot semantic segmentation (GFSS) enhances model performance for base classes with abundant data, while novel classes have limited data access, improving generalization with scarce data. Building on the design of query-based segmentation models, which decouple the mask and classification tasks for individual optimization, we here present the Dynamic Category Queries Transformer (DCQ-Former) which forms a novel approach to the GFSS. The proposed DCQ-Former first uses category suggested dynamic queries to perform mask segmentation and category classification tasks on a... (More)

Few-shot segmentation (FSS) tackles data scarcity using multiple priors, but its simplicity limits handling base and novel classes with limited data access. Generalized few-shot semantic segmentation (GFSS) enhances model performance for base classes with abundant data, while novel classes have limited data access, improving generalization with scarce data. Building on the design of query-based segmentation models, which decouple the mask and classification tasks for individual optimization, we here present the Dynamic Category Queries Transformer (DCQ-Former) which forms a novel approach to the GFSS. The proposed DCQ-Former first uses category suggested dynamic queries to perform mask segmentation and category classification tasks on a large amount of base class data. Considering the case when the novel classes only have access to a limited amount of training data, the queries for the novel classes are instead dynamically composed from the base classes in order to prevent the category suggested module from providing limited suggestion queries given the representativeness of the few-shot samples. Extensive experiments on COCO-20i and Pascal-5i datasets show that DCQ-Former achieves superior accuracy and generalization than current state-of-the-art methods. Our code are available at https://github.com/fallpavilion/DCQ-Former.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
few-shot learning, query-based transformer, semantic segmentation
in
ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
conference name
2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025
conference location
Hyderabad, India
conference dates
2025-04-06 - 2025-04-11
external identifiers
  • scopus:105009592845
ISSN
1520-6149
DOI
10.1109/ICASSP49660.2025.10890147
language
English
LU publication?
yes
additional info
Publisher Copyright: © 2025 IEEE.
id
b37d8558-ecc1-4c5a-bc03-08495ae2f24a
date added to LUP
2026-01-14 16:01:25
date last changed
2026-01-15 12:33:45
@article{b37d8558-ecc1-4c5a-bc03-08495ae2f24a,
  abstract     = {{<p>Few-shot segmentation (FSS) tackles data scarcity using multiple priors, but its simplicity limits handling base and novel classes with limited data access. Generalized few-shot semantic segmentation (GFSS) enhances model performance for base classes with abundant data, while novel classes have limited data access, improving generalization with scarce data. Building on the design of query-based segmentation models, which decouple the mask and classification tasks for individual optimization, we here present the Dynamic Category Queries Transformer (DCQ-Former) which forms a novel approach to the GFSS. The proposed DCQ-Former first uses category suggested dynamic queries to perform mask segmentation and category classification tasks on a large amount of base class data. Considering the case when the novel classes only have access to a limited amount of training data, the queries for the novel classes are instead dynamically composed from the base classes in order to prevent the category suggested module from providing limited suggestion queries given the representativeness of the few-shot samples. Extensive experiments on COCO-20<sup>i</sup> and Pascal-5<sup>i</sup> datasets show that DCQ-Former achieves superior accuracy and generalization than current state-of-the-art methods. Our code are available at https://github.com/fallpavilion/DCQ-Former.</p>}},
  author       = {{Huang, Kunze and Yang, Jieyuan and Jakobsson, Andreas and Tang, Luyao and Tu, Xiaotong and Ding, Xinghao and Huang, Yue}},
  issn         = {{1520-6149}},
  keywords     = {{few-shot learning; query-based transformer; semantic segmentation}},
  language     = {{eng}},
  series       = {{ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings}},
  title        = {{Dynamic Category Queries Transformer for Generalized Few-shot Semantic Segmentation}},
  url          = {{http://dx.doi.org/10.1109/ICASSP49660.2025.10890147}},
  doi          = {{10.1109/ICASSP49660.2025.10890147}},
  year         = {{2025}},
}