Local Graph Point Attention Network in Point Cloud Segmentation

Exploiting global factors and embedding them directly into local graphs in point clouds are challenging due to dense points irregular structure. To accomplish this goal, we propose a novel end-to-end trainable graph attention network that extracts features terms of graphs. Our presents general graph, which obtains the most fundamental based on order positions different neighborhoods. Central is introduced share weights with neighboring reinforce central impacts. As result, one specific can obtain from corresponding ones other neighborhoods but still focus its area by encoded weight sharing. Experimental results benchmark cloud segmentation datasets demonstrate our proposed network’s competitive performance.