DBKNet v1.0.0

DBKNet v1.0.0

This release provides the complete trained checkpoints and training records used in the DBKNet experiments.

Contents

• Checkpoints for the formal DBKNet model
• Checkpoints for comparison models
• Checkpoints for ablation experiments
• Checkpoints for spatial-validation experiments
• Checkpoints for loss-function experiments
• Original training and evaluation logs
• Experiment configurations saved in the original work directories
• Checkpoint manifest with file sizes and SHA-256 values

Main Settings

• Input size: 256 × 256
• Input channels: 6
• Number of classes: 3
• Ignore index: 255
• Formal training iterations: 10,000
• Random seed: 0

The source code and experiment configurations are available in the main repository. Raw Landsat imagery is not included because of its large file size.