CoBRA-CYL-QR: Dataset of QR Codes on Cylindrical Surfaces

Dataset summary The CoBRA-CYL-QR dataset includes 5775 images of QR Codes with various number of alignment patterns (1, 6, 13) on cylindrical surfaces with annotated ground truth. They are all generated with the CoBRA dataset synthesis system which is based on 3D rendering technology. File naming pattern There are two folders in the root directory of the dataset: images and markup. Each image in the images folder is named using the following pattern: aa_bb_cccc_ddddd.png, where aa is the number of alignment patterns in QR Code, bb is the index of used model & texture pair, cccc is the index of camera location on the predefined grid of view points, and ddddd is a unique hash of the image in the entire dataset. The corresponding ground truth for each image is stored under the same file name with the appended .json file extension in the markup folder. Annotation format Each JSON ground truth file contains one JSON object that provides information for the single QR Code located in the image: - barcode symbology (key symbology), always equal to QR_CODE;- barcode dimensions in the number of modules (keys dim1 and dim2);- barcode message (key value) in Base64 encoding;- barcode quadrangle (key quad);- barcode grid of modules (key grid);- barcode curvature value (key curvature);- contours of the three Finder Patterns of QR Code (key finder_patterns);- central modules of the three Finder Patterns of QR Code (key finder_patterns_centers);- contours of the Alignment Patterns of QR Code (key alignment_patterns);- central modules of the Alignment Patterns of QR Code (key alignment_patterns_centers); The vertices of the QR Code quadrangle (quad) are indexed starting from the upper left corner of the code spiraling clockwise.The QR Code grid is stored as a list of module corners which is ordered by rows of QR Code starting from the upper left corner of a QR Code and lasting with its lower right one.Finder Patterns and Alignment Patterns of QR Code are stored in ordered lists under the keys finder_patterns, finder_patterns_centers, alignment_patterns, alignment_patterns_centers where each element is itself an ordered list of points of a pattern. Finder Patterns are stored in the following order: upper left, upper right, lower left.Alignment Patterns are ordered like points in the grid: from the upper left pattern down to the lower right one.Their points (of contours and central modules) are indexed starting from the upper left corner of a pattern spiraling clockwise (like in quad).数据集概览 CoBRA-CYL-QR数据集包含5775张圆柱面上的二维码图像，这些二维码的对齐图案（Alignment Pattern）数量分别为1、6、13，且所有数据均配有带标注的真实真值（ground truth）。本数据集全部基于3D渲染技术的CoBRA数据集合成系统生成。 ## 文件命名规则 数据集根目录包含两个文件夹：`images`（图像文件夹）与`markup`（标注文件夹）。图像文件夹中的每张图像均遵循如下命名格式：`aa_bb_cccc_ddddd.png`，其中： - `aa` 为二维码的对齐图案数量； - `bb` 为所用模型与纹理对的索引； - `cccc` 为预定义视点网格上的相机位置索引； - `ddddd` 为该图像在全数据集中的唯一哈希值。 每张图像对应的真实标注文件与图像同名，仅追加`.json`扩展名，存储于`markup`文件夹中。 ## 标注格式 每个JSON格式的真实标注文件均包含一个JSON对象，用于描述图像中的单个二维码： 1. 条码符号类型（键`symbology`）：固定为`QR_CODE`； 2. 条码模块数维度（键`dim1`与`dim2`）； 3. 条码内容（键`value`），采用Base64编码； 4. 二维码四边形区域（键`quad`）； 5. 二维码模块网格（键`grid`）； 6. 条码曲率值（键`curvature`）； 7. 二维码三个定位图案（Finder Pattern）的轮廓（键`finder_patterns`）； 8. 二维码三个定位图案的中心模块（键`finder_patterns_centers`）； 9. 二维码对齐图案的轮廓（键`alignment_patterns`）； 10. 二维码对齐图案的中心模块（键`alignment_patterns_centers`）； 二维码四边形区域（`quad`）的顶点从该二维码的左上角起始，按顺时针螺旋方向进行索引。 二维码模块网格以模块角点列表的形式存储，按二维码的行顺序排列，起始于二维码左上角，终止于右下角。 二维码的定位图案与对齐图案分别以有序列表的形式存储于`finder_patterns`、`finder_patterns_centers`、`alignment_patterns`、`alignment_patterns_centers`键下，每个列表元素本身为对应图案的点坐标有序列表。 定位图案的存储顺序为：左上、右上、左下。对齐图案的排列顺序与模块网格一致：从左上角的图案依次至右下角的图案。 定位图案与对齐图案的轮廓点及中心模块点均从图案的左上角起始，按顺时针螺旋方向进行索引（与`quad`的索引规则一致）。