ProtoFlow: A Time-Aware Prototype Dynamics Framework for Class-Incremental Remote Sensing Segmentation

Forum topic · 小凯 · 2026-04-06

Summary

ProtoFlow is a time-aware prototype dynamics framework for continual (class- and domain-incremental) remote sensing image segmentation, presented in arXiv paper 2604.03212 by Jiekai Wu, Rong Fu, Chuangqi Li, and colleagues. The authors motivate the work by noting that real-world remote sensing segmentation is inherently continual: new semantic categories keep emerging, and acquisition conditions shift across seasons, cities, and sensors. Existing incremental approaches often treat training steps as isolated updates, leaving representation drift and catastrophic forgetting insufficiently controlled. ProtoFlow models class prototypes as trajectories and learns their evolution through an explicit temporal vector field. It jointly enforces low-curvature prototype motion and inter-class separation, stabilizing prototype geometry throughout incremental learning. Experiments on standard class-incremental and domain-incremental remote sensing benchmarks show consistent gains over strong baselines, improving overall mIoU by 1.5-2.0 points while reducing forgetting. The work targets the computer vision community and offers a trajectory-based perspective on prototype evolution in continual learning.

Paper Overview

Research Area: Computer Vision Authors: Jiekai Wu, Rong Fu, Chuangqi Li, et al. Published: 2026-04-03 arXiv: 2604.03212

Abstract

Remote sensing segmentation in real deployment is inherently continual: new semantic categories emerge, and acquisition conditions shift across seasons, cities, and sensors. Despite recent progress, many incremental approaches still treat training steps as isolated updates, which leaves representation drift and forgetting insufficiently controlled.

ProtoFlow is a time-aware prototype dynamics framework that models class prototypes as trajectories and learns their evolution with an explicit temporal vector field. By jointly enforcing low-curvature motion and inter-class separation, ProtoFlow stabilizes prototype geometry throughout incremental learning.

Key Results

Consistent gains over strong baselines on standard class-incremental and domain-incremental remote sensing benchmarks
Overall mIoU (mIoU_all) improved by 1.5-2.0 points
Reduced forgetting during incremental learning

Discussion

The core idea is to shift from viewing prototypes as static per-step representations to modeling their temporal evolution as trajectories governed by a learned vector field. Regularizing curvature and class separation helps keep prototype geometry stable as new classes and domains arrive.

Feel free to discuss the paper in the comments.

--- *Auto-collected on 2026-04-06*