3D 장면에서의 멀티모달 상황 인식: 대규모 데이터셋 및 벤치마크 소개
본 논문에서는 대규모 멀티모달 상황 인식 데이터셋인 MSQA를 제시하고, 이를 활용하여 3D 장면에서의 상황 인식 및 행동 예측 능력을 평가하는 벤치마크를 소개합니다.
ALOcc: Balancing Speed and Accuracy in 3D Semantic Occupancy and Flow Prediction for Autonomous Driving
This paper introduces ALOcc, a novel convolutional architecture that achieves state-of-the-art speed and accuracy in predicting 3D semantic occupancy and flow from surround-view camera data, addressing key challenges in 2D-to-3D view transformation and multi-task feature encoding for autonomous driving applications.
단일 이미지에서 신경망 필드와 기반 모델 특징을 추출하여 3D 장면을 인식하는 DistillNeRF
DistillNeRF는 제한된 2D 이미지 입력에서 풍부한 3D 장면 표현을 예측하기 위해 오프라인 NeRF 및 시각적 기반 모델에서 지식을 추출하는 자기 지도 학습 프레임워크입니다.
SCRREAM: A Framework and Benchmark for Annotating Dense 3D Indoor Scenes
This paper introduces SCRREAM, a novel framework for creating highly accurate and dense 3D annotations of indoor scenes, addressing limitations in existing datasets that prioritize scale over detailed geometry.
Learning-free Uplifting of 2D Visual Features to 3D Gaussian Splatting Scenes for Efficient Semantic Segmentation
This paper introduces a novel, learning-free method for uplifting 2D visual features to 3D Gaussian Splatting scenes, enabling efficient and effective semantic segmentation in 3D scenes without requiring computationally expensive optimization procedures.
SAM-Guided Masked Token Prediction for Enhanced 3D Scene Understanding via Knowledge Distillation from 2D Foundation Models
This paper introduces a novel method for improving 3D scene understanding by using a two-stage masked token prediction framework guided by SAM (Segment Anything Model) to transfer knowledge from pre-trained 2D foundation models to 3D models.
A New Method for 3D Scene Understanding Using Panoptic Segmentation within Neural Radiance Fields and Guided by Perceptual Priors
This paper proposes a novel method for 3D scene understanding that leverages 2D panoptic segmentation information within a neural radiance field framework, guided by perceptual priors, to achieve accurate and consistent 3D panoptic segmentation.
オープンボキャブラリーを用いた3Dシーングラフによるオブジェクトグラウンディング
提案手法BBQは、3Dシーングラフ表現と大規模言語モデルを組み合わせることで、複雑な自然言語クエリに基づいて3Dシーン内のオブジェクトを正確に特定することができる。
OpenSUN3D Workshop Challenge on Open-Vocabulary 3D Scene Understanding
Advancing open-vocabulary 3D scene understanding through innovative methods and challenges.
