Falling Things (FAT)
收藏资源简介:
Falling Things (FAT) 数据集由英伟达创建,专注于提升机器人领域中物体检测和3D姿态估计的技术水平。该数据集包含61,500张注释照片,涉及21种家用物品,每张图片均提供3D姿态、逐像素类别分割及2D/3D边界框坐标。数据集通过合成高质量的物体模型和背景,生成具有精确3D姿态注释的逼真图像。创建过程中,利用Unreal Engine 4的定制插件,在多种虚拟环境中随机放置和旋转物体,模拟自然下落过程,以收集数据。FAT数据集适用于训练和评估机器人场景理解算法,特别是在处理复杂光照和多传感器模式下的物体检测和姿态估计问题。
Falling Things (FAT) dataset was created by NVIDIA, focusing on advancing the state-of-the-art in object detection and 3D pose estimation for robotic applications. This dataset contains 61,500 annotated photographs covering 21 household items, with each image providing 3D pose annotations, pixel-wise category segmentation masks, and 2D/3D bounding box coordinates. The dataset is constructed by synthesizing high-quality object models and backgrounds to generate photorealistic images with accurate 3D pose annotations. During the development process, a custom plugin for Unreal Engine 4 was employed to randomly place and rotate objects across various virtual environments, simulating natural falling trajectories for data acquisition. The FAT dataset is well-suited for training and evaluating robotic scene understanding algorithms, especially for object detection and pose estimation tasks under complex lighting conditions and across multi-sensor modalities.

- Falling Things (FAT)数据集首次发表,旨在为机器人抓取任务提供一个标准化的评估平台。
- FAT数据集首次应用于机器人抓取算法的研究,促进了相关领域的技术进步。
- FAT数据集被广泛用于多个国际会议和竞赛中,成为评估机器人抓取性能的重要基准。
- 1Falling Things: A Synthetic Dataset for 3D Object Detection and Pose EstimationNVIDIA · 2018年
- 2A Benchmark for 3D Object Detection and Pose Estimation in Cluttered ScenesUniversity of California, Berkeley · 2020年
- 3Real-Time 3D Object Detection and Pose Estimation for Autonomous DrivingStanford University · 2021年
- 43D Object Detection and Pose Estimation Using Deep Learning: A SurveyMassachusetts Institute of Technology · 2022年
- 5Towards Robust 3D Object Detection and Pose Estimation in Cluttered EnvironmentsCarnegie Mellon University · 2023年



