TJU-DHD

TJU-DHD是用于物体检测和行人检测的高分辨率数据集。数据集包含115,354高分辨率图像 (52% 图像具有1624x1200像素的分辨率，48% 图像具有至少2,560x1，440像素的分辨率) 和总共709,330个带标签的对象，它们在比例和外观上变化很大。同时，数据集在季节方差，光照方差和天气方差方面具有丰富的多样性。TJU-DHD is a high-resolution dataset tailored for object detection and pedestrian detection. It consists of 115,354 high-resolution images, with 52% of them having a resolution of 1624×1200 pixels and 48% featuring a resolution of no less than 2560×1440 pixels. The dataset also contains a total of 709,330 labeled objects that vary considerably in scale and appearance. Furthermore, it exhibits rich diversity across seasonal, illumination and weather variations.

搜集汇总

背景与挑战

背景概述

TJU-DHD是一个用于物体和行人检测的高分辨率数据集，包含约11.5万张图像和超过70.9万个标注对象，图像在分辨率、尺度、外观及季节、光照、天气等环境条件上具有丰富多样性。该数据集由Tianjin University等机构于2020年发布。