SemanticKITTI
PulseAugur coverage of SemanticKITTI — every cluster mentioning SemanticKITTI across labs, papers, and developer communities, ranked by signal.
8 day(s) with sentiment data
-
RayLift framework enhances 3D semantic scene completion using geometry priors
Researchers have introduced RayLift, a novel framework designed to improve 3D semantic scene completion for applications like autonomous driving. Unlike previous methods that rigidly use stereo depth estimates, RayLift …
-
LiDAR Scene Completion: Iterative Refinement vs. Wider Predictors Studied
A new research paper explores the effectiveness of test-time refinement strategies for LiDAR semantic scene completion. The study compares one-shot predictors, wider one-shot predictors, and iterative refinement systems…
-
GeoScene framework uses geospatial data to improve 3D scene completion
Researchers have developed GeoScene, a novel framework designed to improve 3D semantic scene completion by integrating geospatial data. This approach combines onboard imagery with structured information from OpenStreetM…
-
New UTok3D tokenizer adapts CLIP for 3D understanding tasks
Researchers have developed UTok3D, a novel parameter-efficient framework designed to adapt CLIP, a vision-language model, for 3D understanding tasks. This tokenizer addresses the challenge of applying CLIP, which is tra…
-
Proteus model offers robust LiDAR compression with 70% truncation tolerance
Researchers have developed Proteus, a novel compression model specifically designed for LiDAR point clouds. This model employs a unique strategy of separating the significant bit-planes (SIG) from the insignificant ones…
-
FPSGen framework generates 3D point cloud scenes independently of partial scans
Researchers have introduced FPSGen, a novel framework for generating 3D point cloud scenes. This method addresses limitations in existing approaches by decoupling scene generation from partial scans, thus avoiding biase…
-
New UP-Fuse framework enhances LiDAR-camera fusion for 3D segmentation
Researchers have developed UP-Fuse, a novel framework for 3D panoptic segmentation that enhances the fusion of LiDAR and camera data. This system is designed to remain robust even when camera sensors degrade or fail, a …
-
New LiDAR ground segmentation method improves autonomous navigation accuracy
Researchers have developed ACZ-GSeg, a novel two-stage method for segmenting ground points from LiDAR data. This approach utilizes an Adaptive Concentric Zone Model to dynamically adjust sector divisions, creating more …
-
New benchmark and method tackle noisy labels in 3D semantic occupancy prediction
Researchers have introduced OccNL, a new benchmark designed to evaluate 3D semantic occupancy prediction models under noisy label conditions. They found that existing 2D label noise learning strategies perform poorly in…
-
New diffusion-based attack targets LiDAR segmentation in autonomous driving
Researchers have developed a novel diffusion-based adversarial attack specifically targeting 2D range-image segmentation models used in autonomous driving. This method, detailed in a new arXiv paper, generates adversari…
-
SparseOcc++ advances 3D semantic occupancy prediction for autonomous driving · 2 sources tracked
Researchers have introduced SparseOcc++, an advanced framework for vision-based 3D semantic occupancy prediction, crucial for autonomous driving. This new method improves upon existing sparse representations by decoupli…
-
PointDiffusion advances 3D scene reconstruction for autonomous driving
Researchers have developed PointDiffusion, a novel method for reconstructing 3D scenes from sparse LiDAR data, crucial for autonomous driving. The approach utilizes a multi-token Gaussian VAE with cross-attention poolin…
-
EditSSC uses Stable Diffusion for editable 3D scene generation
Researchers have developed EditSSC, a new method for generating and editing 3D semantic scenes using 2D Bird's Eye View (BEV) representations. This approach repurposes components from Stable Diffusion, enabling training…
-
Hypergraph framework enhances point cloud segmentation for novel class discovery
Researchers have developed a novel hypergraph-based framework for point cloud segmentation that improves the discovery of unknown object classes. This method moves beyond traditional pairwise associations to model compl…
-
New LiDAR OOD Detection Method Improves Autonomous Driving Safety
Researchers have developed a new framework called Relative Energy Learning (REL) for detecting out-of-distribution (OOD) objects in 3D LiDAR point clouds, a crucial task for autonomous driving safety. Unlike previous me…
-
New diffusion models enhance 3D generation and mesh creation
Researchers are developing new methods for 3D generation using diffusion models and voxel-based approaches. SymTRELLIS enforces symmetry in 3D models by learning linear transformations on voxel latents, improving physic…
-
New U4D framework enhances 4D LiDAR scene generation using uncertainty
Researchers have developed a new framework called U4D for generating 4D LiDAR scenes, addressing the limitation of current methods that apply uniform modeling capacity across all spatial regions. U4D leverages spatial u…
-
Vanilla ViT achieves state-of-the-art in automotive point cloud segmentation
Researchers have developed VaViT, a method that effectively uses vanilla Vision Transformer (ViT) architectures for semantic segmentation of automotive lidar point clouds. This approach addresses the dominance of U-Net …
-
New AI Models Advance 3D Shape Completion and Depth Estimation
Researchers have introduced several new models for 3D shape completion and depth estimation. The Large Depth Completion Model (LDCM) uses a transformer to generate dense depth maps from sparse observations, outperformin…
-
OmniLiDAR framework unifies 3D LiDAR generation across diverse domains
Researchers have developed OmniLiDAR, a unified diffusion framework capable of generating 3D LiDAR scans across diverse domains including varied weather, sensor configurations, and acquisition platforms. This unified ap…