We study the problem of efficient semantic segmentation for 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 In this paper, we introduce RandLA-Net, an lightweight neural architecture directly infer per-point semantics The key our approach is use random instead more complex selection approaches. Although remarkably computation memory...

Extracting robust and general 3D local features is key to downstream tasks such as point cloud registration reconstruction. Existing learning-based descriptors are either sensitive rotation transformations, or rely on classical handcrafted which neither nor representative. In this paper, we introduce a new, yet conceptually simple, neural architecture, termed SpinNet, extract rotationally invariant whilst sufficiently informative enable accurate registration. A Spatial Point Transformer...

In this paper, we propose a novel 3D-RecGAN approach, which reconstructs the complete 3D structure of given object from single arbitrary depth view using generative adversarial networks. Unlike existing work typically requires multiple views same or class labels to recover full geometry, proposed only takes voxel grid representation as input, and is able generate occupancy by filling in occluded/missing regions. The key idea combine capabilities autoencoders conditional Generative...

We propose a novel, conceptually simple and general framework for instance segmentation on 3D point clouds. Our method, called 3D-BoNet, follows the design philosophy of per-point multilayer perceptrons (MLPs). The directly regresses bounding boxes all instances in cloud, while simultaneously predicting point-level mask each instance. It consists backbone network followed by two parallel branches 1) box regression 2) prediction. 3D-BoNet is single-stage, anchor-free end-to-end trainable....

We present a simple yet powerful neural network that implicitly represents and renders 3D objects scenes only from 2D observations. The models geometries as general radiance field, which takes set of images with camera poses intrinsics input, constructs an internal representation for each point the space, then corresponding appearance geometry viewed arbitrary position. key to our approach is learn local features pixel in project these points, thus yielding rich representations. additionally...

An essential prerequisite for unleashing the potential of supervised deep learning algorithms in area 3D scene understanding is availability large-scale and richly annotated datasets. However, publicly available datasets are either relative small spatial scales or have limited semantic annotations due to expensive cost data acquisition annotation, which severely limits development fine-grained context point clouds. In this paper, we present an urban-scale photogrammetric cloud dataset with...

We study the problem of efficient semantic segmentation 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 In this paper, we introduce RandLA-Net, an lightweight neural architecture directly infer per-point semantics for The key our approach is use random instead more complex selection approaches. Although remarkably computation memory...

We report results on the direct observation of microscopic origins backswitching in ferroelectric thin films. The piezoelectric response generated film by a biased atomic force microscope tip was used to obtain static and dynamic images individual grains polycrystalline material. demonstrate that polarization reversal occurs under no external field (i.e., loss remanent polarization) via dispersive continuous-time random walk process, identified stretched exponential decay polarization.

In a warm and humid climate, increasing the temperature set point offers considerable energy benefits with low first costs. Elevated air movement generated by personally controlled fan can compensate for negative effects caused an increased point. Fifty-six tropically acclimatized persons in common Singaporean office attire (0.7 clo) were exposed 90 minutes to each of five conditions: 23, 26, 29°C latter two cases without occupant-controlled movement. Relative humidity was maintained at 60%....

In this paper, we propose a novel approach, 3D-RecGAN++, which reconstructs the complete 3D structure of given object from single arbitrary depth view using generative adversarial networks. Unlike existing work typically requires multiple views same or class labels to recover full geometry, proposed 3D-RecGAN++ only takes voxel grid representation as input, and is able generate occupancy with high resolution 2563 by recovering occluded/missing regions. The key idea combine capabilities...

We study the problem of recovering an underlying 3D shape from a set images. Existing learning based approaches usually resort to recurrent neural nets, e.g., GRU, or intuitive pooling operations, max/mean poolings, fuse multiple deep features encoded input However, GRU are unable consistently estimate shapes given different permutations same images as unit is permutation variant. It also unlikely refine more due long-term memory loss GRU. Commonly used limited capturing partial information,...

According to the relevant test standards and characteristics of special electric vehicles, first-level evaluation indexes are set up as power, trafficability, mass geometric parameters drive parameters, second-level determined based on typical driving conditions. The usability trafficability index is analyzed, weights primary secondary by AHP CRITIC respectively, comprehensive combination qualitative analysis quantitative analysis. correlation degree vehicle calculated grey theory, result...

Odometry is of key importance for localization in the absence a map. There considerable work area visual odometry (VO), and recent advances deep learning have brought novel approaches to VO, which directly learn salient features from raw images. These learning-based led more accurate robust VO systems. However, they not been well applied point cloud data yet. In this work, we investigate how exploit estimate (PCO), may serve as critical component cloud-based downstream tasks or Specifically,...

The benefits of thermal comfort and indoor air quality with personalized ventilation (PV) systems have been demonstrated in recent studies. One the barriers for wide spread acceptance by architects HVAC designers has attributed to challenges constraints faced integration PV work station. A newly developed ceiling-mounted system addresses these provides a practical solution while retaining much apparent systems. Assessments environment, movement, were performed tropically acclimatized...

Many of the world reserves fossil fuels are located at various water depths in fine-grained sediment under seabed. The contains relatively large biogas bubbles, which has been posing challenges to stability offshore foundations supporting oil and gas platforms. Although gassy soil was found exhibit different undrained shear strengths (c u ) by altering initial pore pressure, i (relevant depth), systematic studies concerning effect on behaviours still lacking. This study reports a series...

Ternary nanocomposite systems of PVDF/Fe₃O₄/CNT and PVDF/Fe₃O₄/GN, prepared with twin screw compounding method, exhibit enhanced microwave absorption properties.