CLIDD: Cross-Layer Independent Deformable Description for Efficient and Discriminative Local Feature Representation
CLIDD introduces Cross-Layer Independent Deformable Description to create robust local feature representations for spatial intelligence tasks like navigation. It achieves superior discriminability by sampling directly from independent feature hierarchies using learnable offsets, avoiding the need for dense feature maps. The method is trained with a composite loss function and optimized with kernel fusion, resulting in highly accurate matching at high efficiency.