Depth-Breadth Fusion

Integrates detailed visual information by combining features from low-level (pixel) to high-level (global semantic) extracted from various layers (depth) of the vision encoder.