garrying/GSD-S
收藏资源简介:
GSD-S是一个用于玻璃表面检测的数据集,并增加了每像素的语义标签。该数据集在NeurIPS 2022的论文《Exploiting Semantic Relations for Glass Surface Detection》中提出。每个样本包含一张RGB照片、一个二进制玻璃掩码和一个43类语义分割图,支持联合玻璃检测和场景语义推理。数据集分为训练集(3,911个样本)和测试集(608个样本),总计4,519个样本。图像分辨率为640×480像素(JPEG格式),所有标注图为PNG格式。数据集包含的列有:image_id(原始文件名)、image(RGB照片)、mask(二进制玻璃掩码)、seg(语义分割图)和seg_colored(用于可视化的假彩色分割图)。语义类别共有43类,包括未知、墙、玻璃、地板、天花板等。
GSD-S is a glass surface detection dataset augmented with per-pixel semantic labels, introduced in the NeurIPS 2022 paper "Exploiting Semantic Relations for Glass Surface Detection". Each sample pairs an RGB photograph with a binary glass mask and a 43-class semantic segmentation map, enabling joint glass detection and scene-semantic reasoning. The dataset is divided into a training set (3,911 samples) and a test set (608 samples), totaling 4,519 samples. Images are 640 × 480 pixels (JPEG format), and all annotation maps are PNG format. The dataset includes the following columns: image_id (original filename stem), image (RGB photograph), mask (binary glass mask), seg (semantic segmentation map), and seg_colored (false-color rendering of seg for visualization). There are 43 semantic classes in total, including unknown, wall, glass, floor, ceiling, etc.




