Search

We study the problem of efficient semantic segmentation of large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale point clouds. In this paper, we introduce RandLA-Net, an efficient and lightweight neural architecture to directly infer per-point semantics for large-scale point clouds. The key to our approach is to use random point sampling instead of more complex point selection approaches. Although remarkably computation and memory efficient, random sampling can discard key features by chance. To overcome this, we introduce a novel local feature aggregation module to progressively increase the receptive field for each 3D point, thereby effectively preserving geometric details. Comparative experiments show that our RandLA-Net can process 1 million points in a single pass up to 200× faster than existing approaches. Moreover, extensive experiments on five large-scale point cloud datasets, including Semantic3D, SemanticKITTI, Toronto3D, NPM3D and S3DIS, demonstrate the state-of-the-art semantic segmentation performance of our RandLA-Net. [ABSTRACT FROM AUTHOR]

Context: Synchronised acoustic recorders can be used as a non-invasive tool to detect and localise sounds of interest, including vocal wildlife and anthropogenic sounds. Due to the high cost of commercial synchronised recorders, acoustic localisation has typically been restricted to small or well funded surveys. Recently, low-cost acoustic recorders have been developed, but until now their efficacy has not been compared with higher specification recorders. Aims: The present study aimed to compare the efficacy of a newly developed low-cost recorder, the Conservation at Range through Audio Classification and Localisation (CARACAL), with an established, high-end recorder, the Wildlife Acoustics Song Meter (SM). Methods: Four recorders of each type were deployed in a paired set-up across five nights in Wisconsin, USA. The recordings allowed for manual identification of domestic dog (Canis familiaris), grey wolf (Canis lupus), coyote (Canis latrans) and barred owl (Strix varia) calls, and then the ability of each recorder type to detect and localise the vocalising animals was compared. Key results: The CARACALs were less sensitive, detecting only 47.5% of wolf, 55% of coyote, 65% of barred owl and 82.5% of dog vocalisations detected by the paired SMs. However, when the same vocalisations were detected on both recorders, localisation was comparable, with no significant difference in the precision or maximum detection ranges. Conclusions: Low-cost recording equipment can be used effectively for acoustic localisation of both wild and domestic animals. However, the lower sensitivity of the CARACALs means that a denser network of these recorders would be needed to achieve the same efficacy as the SMs. Deploying a greater number of cheaper recorders increases the labour time in the field and the quantity of data to process and store. Thus, there is a trade-off between cost and time to be considered. Implications: The ability to use low-cost recorders for acoustic localisation provides new avenues for tracking, managing and researching a wide range of wildlife species. Presently, CARACALs are more suited to monitoring species that have small home ranges and high amplitude vocalisations, and for when a large time investment for in situ equipment checks and data processing is feasible. Synchronised acoustic recorders can be used as a non-invasive tool to detect and localise sounds of interest, including vocal wildlife. This study compared the efficacy of a commercial, high-end acoustic recorder with a newly developed low-cost recorder, finding that although the commercial recorder was more sensitive, the acoustic localisation capabilities of the two were comparable. The ability to use low-cost recorders for acoustic localisation provides new avenues for tracking, managing and researching a wide range of wildlife species. Image by Bethany R. Smith. [ABSTRACT FROM AUTHOR]