PeP: a Point enhanced Painting method for unified point cloud tasks
This work addresses the need for better point cloud feature encoding in computer vision, offering a plug-and-play solution that is model-agnostic, though it appears incremental as it builds on existing point painting and encoder methods.
The paper tackles the problem of point cloud recognition by proposing PeP, a novel module that combines refined point painting with a language model-based encoder to enhance feature encoding for downstream tasks. Experiments on nuScenes and KITTI datasets show strong performance in semantic segmentation and object detection across lidar and multi-modal settings.
Point encoder is of vital importance for point cloud recognition. As the very beginning step of whole model pipeline, adding features from diverse sources and providing stronger feature encoding mechanism would provide better input for downstream modules. In our work, we proposed a novel PeP module to tackle above issue. PeP contains two main parts, a refined point painting method and a LM-based point encoder. Experiments results on the nuScenes and KITTI datasets validate the superior performance of our PeP. The advantages leads to strong performance on both semantic segmentation and object detection, in both lidar and multi-modal settings. Notably, our PeP module is model agnostic and plug-and-play. Our code will be publicly available soon.