SHREC 2020: 3D point cloud semantic segmentation for street scenes. (December 2020)

• Record Type: Journal Article
• Authors: Ku, Tao; Veltkamp, Remco C.; Boom, Bas; Duque-Arias, David; Velasco-Forero, Santiago; Deschaud, Jean-Emmanuel; Goulette, Francois; Marcotegui, Beatriz; Ortega, Sebastián; Trujillo, Agustín; Suárez, José Pablo; Santana, José Miguel; Ramírez, Cristian; Akadas, Kiran; Gangisetty, Shankar
• Is Part Of: Computers & graphics. Volume 93(2020)
• Page Start: 13
• Page End: 24
• Publication Date: 2020-12
• Subjects: SHREC 2020 -- 3D point cloud -- Semantic segmentation -- Benchmark; Computer graphics -- Periodicals; 006.6
• DOI: 10.1016/j.cag.2020.09.006
• Languages: English
• ISSNs: 0097-8493

Abstract

Highlights: Provide a large-scale 3D street-scene point cloud dataset for 3D semantic segmentation. Evaluate different algorithms on the dataset and help finding solutions for large-scale 3D point cloud processing. We have five algorithms under evaluation with one based on handcrafted detectors, one based on 3D-to-2D projection learning, and the other three being end-to-end learning-based methods. The results show that point-set based end-to-end learning methods can learn representative features directly from 3D points and performs better than handcrafted methods. Graphical abstract: Abstract: Scene understanding of large-scale 3D point clouds of an outer space is still a challenging task. Compared with simulated 3D point clouds, the raw data from LiDAR scanners consist of tremendous points returned from all possible reflective objects and they are usually non-uniformly distributed. Therefore, its cost-effective to develop a solution for learning from raw large-scale 3D point clouds. In this track, we provide large-scale 3D point clouds of street scenes for the semantic segmentation task. The data set consists of 80 samples with 60 for training and 20 for testing. Each sample with over 2 million points represents a street scene and includes a couple of objects. There are five meaningful classes: building, car, ground, pole and vegetation. We aim at localizing and segmenting semantic objects from these large-scale 3D point clouds. Four groups contributed their results with different methods. The results show that learning-based methods are the trend and one of them achieves the best performance on both Overall Accuracy and mean Intersection over Union. Next to the learning-based methods, the combination of hand-crafted detectors are also reliable and rank second among comparison algorithms.