Places in the Wild
收藏资源简介:
Places in the Wild是由哥伦比亚大学巴纳德学院创建的一个大规模、高分辨率RAW格式真实世界场景照片数据集,旨在为视觉神经科学和计算机视觉研究提供生态效度更高的数据资源。该数据集包含来自810个物理位置的67,574张图像,覆盖260个基础场景类别,总计像素量超过3万亿,并同时提供14位RAW原始格式和JPEG压缩格式文件,数据来源于在纽约市及周边地区实地采集的室内、城市和自然环境。数据采集过程采用标准化协议,使用4500万像素全画幅相机在固定点位以5度为间隔进行全景拍摄,确保了视角的密集采样和空间一致性。本数据集主要应用于研究人类和模型的视角依赖识别、在真实条件下训练和评估场景理解系统、表征自然场景统计特性,以及需要近全视野视觉显示的科学实验,以解决现有网络数据集分辨率低、拍摄条件未知且空间上下文有限的核心局限。
Places in the Wild is a large-scale, high-resolution RAW-format real-world scene photograph dataset created by Barnard College of Columbia University, which aims to provide data resources with higher ecological validity for visual neuroscience and computer vision research. This dataset contains 67,574 images from 810 physical locations, covering 260 basic scene categories, with a total pixel count exceeding 3 trillion. It provides both 14-bit RAW original format and JPEG compressed format files, and the data is collected from indoor, urban, and natural environments in New York City and its surrounding areas via field surveys. The data collection process adopts a standardized protocol, using 45-megapixel full-frame cameras to conduct panoramic shooting at fixed sites at 5-degree intervals, ensuring dense sampling of viewpoints and spatial consistency. This dataset is mainly applied to research on viewpoint-dependent recognition of humans and models, training and evaluating scene understanding systems under real-world conditions, characterizing the statistical properties of natural scenes, and scientific experiments requiring near-full-field visual displays, so as to address the core limitations of existing web-based datasets, including low resolution, unknown shooting conditions, and limited spatial context.
数据集概述:Places in the Wild: Ecologically-sampled RAW photographs
- 数据集名称:Places in the Wild: Ecologically-sampled RAW photographs
- 作者:Michelle R. Greene
- 类型:数据集
- 标签:scenes、photographs、raw photographs
- 摘要:
- 包含超过 67,000 张 RAW 格式 图像。
- 每张图像使用 45 兆像素 Canon EOS R5 全画幅无反相机 拍摄。
- 以 5 度间隔 捕获,提供 360 度覆盖。
- 覆盖 800 多个独特位置。
- 涵盖 260 个基础场景类别(包括室内和室外环境,如卧室、火车站、森林、停车场等)。
- 文件信息:
- 文件名:places_in_the_wild.tar
- 文件大小:2.62 TB
- 许可证:CC BY-NC-SA 4.0
- 相关 URL:https://zenodo.org/records/20479095

- 1Places in the Wild: A Large, High-Resolution RAW Photograph Dataset for Ecologically Valid Vision Research哥伦比亚大学·巴纳德学院 · 2026年



