Point Transformer-Based Salient Object Detection Network for 3-D Measurement Point Clouds

While salient object detection (SOD) on 2-D images has been extensively studied, there is very little SOD work on 3-D measurement surfaces. We propose an effective point transformer-based SOD network for 3-D measurement point clouds, termed PSOD-Net. PSOD-Net is an encoder-decoder network that takes full advantage of transformers to model the contextual information in both multiscale point- and scenewise manners. In the encoder, we develop a point context transformer (PCT) module to capture region contextual features at the point level; PCT contains two different transformers to excavate the relationship among points. In the decoder, we develop a scene context transformer (SCT) module to learn context representations at the scene level; SCT contains both upsampling-and-transformer (UT) blocks and multicontext aggregation (MCA) units to integrate the global semantic and multilevel features from the encoder into the global scene context. Experiments show clear improvements of PSOD-Net over its competitors and validate that PSOD-Net is more robust to challenging cases such as small objects, multiple objects, and objects with complex structures.