As the perception range of LiDAR expands, LiDAR-based 3D object detection contributes ever-increasingly to long-range in autonomous driving. Mainstream detectors often build dense feature maps, where cost is quadratic range, making them hardly scale up settings. To enable efficient detection, we first propose a fully sparse detector termed FSD. FSD built upon general voxel encoder and novel instance recognition (SIR) module. SIR groups points into instances applies highly-efficient...

Animated video separates foreground and background elements into layers, with distinct processes for sketching, refining, coloring, in-betweening. Existing generation methods typically treat animation as a monolithic data domain, lacking fine-grained control over individual layers. In this paper, we introduce LayerAnimate, novel architectural approach that enhances layers within diffusion model, allowing users to independently manipulate in To address the challenge of limited layer-specific...

This paper aims for high-performance offline LiDAR-based 3D object detection. We first observe that experienced human annotators annotate objects from a track-centric perspective. They label in track with clear shapes, and then leverage the temporal coherence to infer annotations of obscure objects. Drawing inspiration this, we propose detector perspective instead conventional object-centric Our method features bidirectional tracking module learning module. Such design allows our refine...

As an essential nonverbal cue, the human gaze reveals intentions and plays a crucial role in daily activities. Therefore, automatic detection of person's target has drawn interests computer vision community. This is useful not only for identifying whether children are attentive class but also locating items interest to humans retail settings. Existing gaze-following methods have explored exploited scenes context head cues. Considering significance human-object interaction understanding...

Gaze-following is a kind of research that requires locating where the person in scene looking automatically under topic gaze estimation. It an important clue for understanding human intention, such as identifying objects or regions interest to humans. However, survey datasets used gaze-following tasks reveals defects way they collect point labels. Manual labeling may introduce subjective bias and labor-intensive, while automatic with eye-tracking device would alter person's appearance. In...

This paper aims for high-performance offline LiDAR-based 3D object detection. We first observe that experienced human annotators annotate objects from a track-centric perspective. They label the with clear shapes in track, and then leverage temporal coherence to infer annotations of obscure objects. Drawing inspiration this, we propose detector perspective instead conventional object-centric Our method features bidirectional tracking module learning module. Such design allows our refine...

We report the photoelectric properties and structural changes of ZnO nanoparticles (∼63 nm) under pressures up to 22.9(5) GPa using in situ Raman spectroscopy, photocurrent measurements, theoretical calculations. The show enhanced (2.8 mA) responsivity (4.78 × 106 mA W–1) 365 nm irradiation wurtzite (B4) phase. On contrary, rock-salt (B1) phase, emerging post exhibits a smaller band gap decreased absorption coefficient, leading reduced photocurrents responsivity. These findings highlight...

Ports play an important role in transportation networks, but the ecological efficiency of many ports cannot reach effective level.To improve ports, we need to analyze current and propose policy recommendations based on results.In this paper, first employed super-efficiency SBM model calculate four major Yangtze River Delta port group from 2011 2021.The results showed that average comprehensive was 0.647, which did not level.The overall curve U-shaped.Then used Tobit key factors affecting...

Free-floating bike-sharing is characterized by environmental-friendly and a sustainable form of urban transportation, which becomes increasingly popular globally.To achieve development, we establish an analytical framework to explore the optimal business models free-floating bikesharing platforms (FBSPs) including ordinary, premium, hybrid that are three often employed FBSPs.This paper aims address platform's decision, characterize conditions under model can Pareto improvement for platform...

Label-efficient LiDAR-based 3D object detection is currently dominated by weakly/semi-supervised methods. Instead of exclusively following one them, we propose MixSup, a more practical paradigm simultaneously utilizing massive cheap coarse labels and limited number accurate for Mixed-grained Supervision. We start observing that point clouds are usually textureless, making it hard to learn semantics. However, geometrically rich scale-invariant the distances from sensors, relatively easy...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Semiconductor-based photocatalysts innovation and heterojunction design incorporating novel materials have shed light on the enhancement of photoelectrochemical (PEC) performances concerning energy fields, among which incorporation plasmonic metal/semiconductor nanostructures has raised extensive research attentions. In this work, an innovatively sandwiched structure composed Ta3N5 nanorods (NRs)/ultrathin TiO2/Au Nanoparticles (NPs) was designed, exhibiting successively improved PEC...

In this paper, we introduce Trim 3D Gaussian Splatting (TrimGS) to reconstruct accurate geometry from images. Previous arts for reconstruction Gaussians mainly focus on exploring strong regularization. Instead, a fresh perspective, propose obtain of scene by trimming, which selectively removes the inaccurate while preserving structures. To achieve this, analyze contributions individual and contribution-based trimming strategy remove redundant or Gaussians. Furthermore, our experimental...

In this research, the author develops databases based on deep learning and cloud computing technology. On basis of designing overall architecture database with distributed C/S mode as core, use J2EE (Java 2 Platform, Enterprise Edition) development tool, apply Oracle server database, extract data features in-depth technology, allocate processing tasks so to finally complete fusion compression. Finally, is completed by backup scheme external encryption. The test results show that developed...

As the perception range of LiDAR expands, LiDAR-based 3D object detection contributes ever-increasingly to long-range in autonomous driving. Mainstream detectors often build dense feature maps, where cost is quadratic range, making them hardly scale up settings. To enable efficient detection, we first propose a fully sparse detector termed FSD. FSD built upon general voxel encoder and novel instance recognition (SIR) module. SIR groups points into instances applies highly-efficient...

Polyacrylamide-Poly(methyl methacrylate) (PAM-PMMA) copolymer emulsion was synthesized by pre-polymerization of MMA, and then copolymerization with AM. The final yield PAM-PMMA can be up to 93.69%, its molecular weight 11.2×104Da, as well the characteristic peaks ester group (-COO) MMA amide (-NH2 ) AM appeared on structure. component content from effectively controlled adjusting conversion MMA. exhibited spherical morphology, particle size 50-100 nm. Moreover, correlation between...