Actor-Critic Driven Continual Learning with Memory Replay for Generative Crowd Counting

2026-05-06

PRE

Jianyong Wang

Xianxun Zhu

Mingliang Gao *

DOI：10.1016/j.patrec.2026.04.029

Original

Original request for help

Save

Abstract

En 中文

• An MAACNet is proposed for continual crowd counting to mitigate catastrophic forgetting. • An actor-critic replay strategy is built to select the high-value samples for knowledge retention. • A dynamic memory management is formulated to prioritize informative data over random sampling.

Keywords:

MAACNet

continual learning

crowd counting

memory replay

actor-critic strategy

Actor-Critic Driven Continual Learning with Memory Replay for Generative Crowd Counting

Abstract

Journal

Organization

Cited Papers

Citing Papers