Real Sentinel-2 Super-Resolution: The Case for Misregistration-Aware Training Data

Key Points

• The central aim is to address inter-band misregistration flaws in Sentinel-2 super-resolution data.
• Proposes a progressive simulation framework for training data generation
• Utilizes a registration-aware graph neural network for improved accuracy
• Identifies inter-band misregistration as a significant flaw
• Suggests methods to mitigate domain gaps in training data