REMOTE SENSING / CHANGE DETECTION
SCA-Net
A research architecture for small-change detection in satellite imagery, focused on buildings, roads, and spatial context
Small-object lift
Reported for small change detection
Metrics
Benchmarks on reference datasets
Core
SegFormer-B1 with multi-scale processing
ARCHITECTURE / SCA-NET
Small changes need serious architecture
This case page behaves like a satellite map: architecture, results, and decision layers sit together to show why road and building change detection is more than comparing two images
SCA-Net focuses on the place where many models fail: small, thin, contextual changes. Built as an improved direction over Change-Agent, it combines a Difference Pyramid Block, multi-scale processing, and spatial-contextual aggregation to separate real change from texture and noise. The project notes report gains on LEVIR-CD and LEVIR-MCI, especially for small objects
Remote SensingSegFormer-B1Siamese NetworkTransformerAttentionPyTorch
MULTI-SCALE / LAYERS
Change map from above
Difference Pyramid Block
Multi-scale difference analysis to separate true change from texture, shadow, and sensor noise
Adaptive Multi-scale Processing
Spatial-contextual aggregation using attention and transformer mechanisms
Controlled compute cost
A Siamese SegFormer-B1 direction balancing accuracy and processing load
Next mission · 04
Generative Face Aging
↗