Dual-Selective Network for Domain-Incremental Change Detection

2026-07-02 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

Dual-Selective Incremental Network (DSINet) is a unified framework built on visual state space models designed to address challenges in domain-incremental change detection (DICD). DICD models struggle to adapt to new geographic domains while preserving prior knowledge due to a fixed label space and drastically varying domain characteristics, leading to unstable spatial change representations and knowledge degradation over long sequences. DSINet mitigates this by incorporating a Selective Spatial State Unit (S3U), which utilizes Mamba's input-dependent selective mechanism to preserve stable spatial change structures and filter domain-specific variations. Additionally, it employs a Concentration-Balanced Distillation (CBD) strategy to stabilize knowledge transfer by balancing hardness and confidence concentration effects. Experimental results indicate DSINet effectively reduces knowledge degradation across long domain sequences while maintaining the linear computational efficiency inherent to state space models.

Key takeaway

For Computer Vision Engineers developing domain-incremental change detection systems, DSINet offers a robust approach to mitigate knowledge degradation across evolving geographic domains. If your current replay-based or regularization methods struggle with scalability or computational cost over long domain sequences, consider DSINet's dual-selective mechanism and concentration-balanced distillation. This framework maintains stable spatial representations and efficient knowledge transfer, providing a more reliable solution for continuous model adaptation in dynamic environments.

Key insights

DSINet stabilizes domain-incremental change detection by selectively preserving spatial structures and balancing knowledge distillation.

Principles

• Preserve stable spatial change structures.
• Filter domain-specific variations.
• Balance distillation hardness and confidence.

Method

DSINet employs Mamba's input-dependent selective mechanism via S3U for stable spatial representations and uses CBD to balance hardness and confidence during incremental updates.

In practice

• Remote sensing change detection.
• Adapting models to new geographies.
• Mitigating knowledge degradation.

Topics

• Domain-Incremental Learning
• Change Detection
• Visual State Space Models
• Mamba Architecture
• Knowledge Distillation
• Remote Sensing

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• RAFT: Data Refinement and Adaptive Distillation for Domain Fine-Tuning with Alleviated Forgetting — RAFT improves domain fine-tuning by refining data and preserving original model behavior through adaptive distillation.
• YOTOnet: Zero-Shot Cross-Domain Fault Diagnosis via Domain-Conditioned Mixture of Experts — YOTOnet enables zero-shot cross-domain fault diagnosis in mechanical equipment using a domain-conditioned mixture of experts.
• Back to Source: Open-Set Continual Test-Time Adaptation via Domain Compensation — DOCO robustly adapts to continuous domain shifts and detects novel classes by synergistically compensating for domain and semantic shifts.
• Switch-KD: Visual-Switch Knowledge Distillation for Vision-Language Models — Switch-KD unifies VLM knowledge transfer in a shared text-probability space for efficient deployment.
• Teacher-Student Structure for Domain Adaptation in Ensemble Audio-Visual Video Deepfake Detection — A teacher-student framework enhances ensemble audio-visual deepfake detection models for robust generalization across diverse, unseen domains.
• Consistently Informative Soft-Label Temperature for Knowledge Distillation — Fixed-temperature knowledge distillation creates inconsistent soft-label entropy, hindering effective knowledge transfer.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.