GSO

交互式3D模拟已经在机器人和计算机视觉方面取得了突破，但是模拟深度学习所需的广泛环境需要照片般逼真的3D对象模型。因为这些模型是扫描的而不是手工建模的，所以它们是现实的，而不是理想化的，降低了将学习从模拟转移到现实的难度。这些模型经过预处理，可用于点火凉亭和子弹模拟平台，但很容易适应其他模拟器。

Interactive 3D simulations have achieved breakthroughs in robotics and computer vision. However, the diverse environments required for simulated deep learning demand photorealistic 3D object models. Since these models are scanned rather than handcrafted, they are realistic rather than idealized, reducing the difficulty of transferring learned knowledge from simulation to the real world. These models have been preprocessed for use with the Ignition Gazebo and Bullet simulation platforms, but can be easily adapted to other simulators.

搜集汇总

背景与挑战

背景概述

GSO是一个用于机器人和计算机视觉的交互式3D模拟数据集，包含照片般逼真的3D对象模型，适用于多种模拟平台，由Google等机构联合发布。