Fig. 5: The overall framework diagram depicts our approach as follows: Initially, an embedding network is employed to extract both support and query features. | npj Heritage Science

Fig. 5: The overall framework diagram depicts our approach as follows: Initially, an embedding network is employed to extract both support and query features.

From: Learning discriminative universal background knowledge for few-shot point cloud semantic segmentation of architectural cultural heritage

Fig. 5: The overall framework diagram depicts our approach as follows: Initially, an embedding network is employed to extract both support and query features.

Subsequently, the Prototype Generator (PG) extracts the support features to generate semantic prototype representations. Next, the Background Modeling Module (BMM) initializes background prototypes, learning background knowledge from the support and query features, and optimizes this process using Adaptive Background Loss (ABL). Concurrently, the Background Filtering Module (BFM) eliminates background noise. Finally, Prototype Contrastive Learning (PCL) differentiates between the target and background regions.

Back to article page