WoodScape

WoodScape包括四个环视摄像机和九个任务，包括分割，深度估计，3D边界盒检测和新型污染检测。为超过10,000个图像提供了实例级别的40个类的语义注释。借助WoodScape，我们希望鼓励社区为鱼眼相机调整计算机视觉模型，而不是使用天真的矫正。

WoodScape is a dataset equipped with four surround-view cameras and covers nine tasks, including semantic segmentation, depth estimation, 3D bounding box detection, and a novel contamination detection task. It provides instance-level semantic annotations for 40 distinct classes across over 10,000 images. With this dataset, we aim to encourage the computer vision community to adapt computer vision models specifically for fisheye cameras, rather than relying on naive correction.

搜集汇总

背景与挑战

背景概述

WoodScape是一个专注于自动驾驶的多任务、多摄像头鱼眼数据集，包含四个环视摄像机和九个计算机视觉任务，如语义分割、深度估计和3D边界盒检测。该数据集提供了超过10,000个图像的实例级别语义注释，涵盖40个类，总大小约39.5GB，旨在推动社区针对鱼眼相机优化模型，而非依赖简单的图像矫正方法。