Title: Places in the Wild: A High-Resolution RAW Image Dataset for Ecologically Valid Vision Studies

The advancement of computer vision and cognitive neuroscience has been significantly driven by the availability of large-scale image datasets. Nevertheless, the majority of these resources consist of low-resolution JPEGs sourced from the internet, often lacking transparency regarding capture conditions and offering restricted spatial context. To address these limitations, the "Places in the Wild" dataset introduces a collection of 67,574 high-resolution photographs. These images were gathered in situ across 810 distinct physical locations, encompassing 260 basic-level scene categories that include natural, urban, and indoor environments.

Data acquisition employed a 45-megapixel Canon EOS R5 camera mounted on a panoramic tripod. At every site, the setup captured 72 images at 5-degree horizontal intervals, supplemented by 12 images taken at varying elevations. This methodology facilitated a dense sampling of 360-degree viewpoints. All recordings were made simultaneously in both 14-bit RAW (CR3) format and compressed JPEG. The retention of RAW files ensures the preservation of sensor-level detail, enabling precise analysis of image statistics such as luminance, contrast, and color.

The dataset is supported by comprehensive EXIF metadata and a range of image-quality metrics. It is designed to facilitate various research applications, including the study of viewpoint-dependent recognition in both humans and artificial models, the training and evaluation of scene-understanding systems under realistic conditions, the characterization of natural scene statistics, and experiments that necessitate near-full-field visual displays.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC