CogVis: Cognitive Memory-Guided Framework for Open-Vocabulary Change Detection
A recent study available on arXiv (2608.06150) introduces CogVis, a cognitive memory-based framework designed for Open-Vocabulary Change Detection (OVCD) in monitoring Earth's surface. This approach redefines OVCD through a perception-memory-verification model, tackling challenges found in current techniques that mix temporal perception, semantic differentiation, and region validation, which can result in inconsistent outcomes and unnecessary computations. CogVis utilizes a Scene Change Perceptron (SCP) to derive a reusable, category-independent change prior from stable bi-temporal features, separating temporal data from semantic category choices. Additionally, a Semantic Memory Calibrator (SMC) adjusts for category-related score variations by estimating a specific decision threshold for each image query. An Adaptive Region Filter (ARF) refines connected candidates using this learned threshold. This research, relevant to computer vision and remote sensing, has implications for environmental monitoring and urban planning.
Key facts
- CogVis is a cognitive memory-guided framework for Open-Vocabulary Change Detection (OVCD).
- It reformulates OVCD as a perception-memory-verification paradigm.
- The framework uses a Scene Change Perceptron (SCP) to extract a category-agnostic change prior.
- A Semantic Memory Calibrator (SMC) estimates an image-query-specific decision threshold.
- An Adaptive Region Filter (ARF) filters connected candidates.
- The paper is available on arXiv with identifier 2608.06150.
- The research addresses limitations of existing OVCD methods.
- The framework is inspired by human visual change perception.
Entities
Institutions
- arXiv