LiDAR-based or RGB-D-based object detection is used in numerous applications, ranging from autonomous driving to robot vision. Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point cloud LiDAR data. However, problems remain, including a slow inference speed and low orientation estimation performance. We therefore investigate an improved sparse convolution method for such networks, which significantly increases the speed of both training and inference. We also introduce a new form of angle loss regression to improve the orientation estimation performance and a new data augmentation approach that can enhance the convergence speed and performance. The proposed network produces state-of-the-art results on the KITTI 3D object detection benchmarks while maintaining a fast inference speed.
Download full-text PDF |
Source |
|---|---|
| http://www.ncbi.nlm.nih.gov/pmc/articles/PMC6210968 | PMC |
| http://dx.doi.org/10.3390/s18103337 | DOI Listing |
Publication Analysis
Top Keywords
Similar Publications
EFGs: A Complete and Accurate Implementation of Ertl's Functional Group Detection Algorithm in RDKit.
J Chem Inf Model
January 2025
Biostatistics and Bioinformatics Unit, IMDEA Food, E28049 Madrid, Spain.
Functional groups are widely used in organic chemistry, because they provide a rationale to analyze physicochemical and reactivity properties. In medicinal chemistry, they are the basis for analyzing ligand-biomacromolecule interactions. Ertl's algorithm is an approach to extract functional groups in arbitrary organic molecules that does not depend on predefined libraries of functional groups.
View Article and Find Full Text PDFSpatial differentiation is the key element for edge detection and holds unquestionable significance in the current information era. All-optical computation based on metasurfaces has emerged as a powerful platform for spatial differentiation due to its advantage of high integration and parallel processing. However, while most current works focus on one- or two-dimensional (2D) spatial differentiation, three-dimensional (3D) all-optical computation for compact spatial differentiator remains elusive.
View Article and Find Full Text PDFIn the field of image processing, optical neural networks offer advantages such as high speed, high throughput, and low energy consumption. However, most existing coherent optical neural networks (CONN) rely on coherent light sources to establish transmission models. The use of laser inputs and electro-optic modulation devices at the front end of these neural networks diminishes their computational capability and energy efficiency, thereby limiting their practical applications in object detection tasks.
View Article and Find Full Text PDFIn this paper, we theoretically analyze the optimization of a Fabry-Pérot cavity for the purpose of detecting partially absorbing objects placed inside without photon exchange. Utilizing the input-output formalism, we quantitatively relate the probability of correctly inferring the presence or absence of the object to the probability of avoiding absorption. We show that, if the cavity decay rate due to absorption by the object is comparable to that of the empty cavity and to the object-induced detuning, the product of the two probabilities is maximized by an undercoupled cavity, in which case detection in transmission is favorable to that in reflection.
View Article and Find Full Text PDFPhase-shifting Fringe projection profilometry (FPP) excels in 3D measurements for many macro-scale applications, but as features-of-interest shrink to the microscopic scale, depth-of-field limitations slow measurements and necessitate mechanical adjustments. To address this, we introduce digital holography (DH) for fringe image capture, enabling numerical refocusing of defocused object regions. Our experiments validate this approach and compare depth measurement noise with other DH and FPP methods.
View Article and Find Full Text PDFWant AI Summaries of new PubMed Abstracts delivered to your In-box?
Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!