CropAndWeedAndLeaf

CropAndWeedandLeaf is a dataset to benchmark leaf-segmentation models across multiple plant species, as described in the corresponding ReLeaf paper, presented at the Agriculture-Vision Workshop at CVPR 2026. The dataset is based on images from the CropAndWeed dataset, which are cropped to individual plant instances and enhanced with instance-segmentation masks for individual leaves. Source code, models and further documentation can be found on our GitHub page. Data Structure The repository provides the following subdirectories: images: cropped images containing all annotated plant instances labels: annotations corresponding to each image in YOLOv8 instance-segmentation format Plant Species The label IDs correspond to the original labels defined in the CropAndWeed dataset. 1: Maize (Zea mays) 7: Sugar beet (Beta vulgaris s. vulgaris) 13: Pea (Pisum sativum) 14: Zucchini (Cucurbita pepo var. gir.) 15: Squash (Cucurbita) 18: Potato (Solanum tuberosum) 22: Poppy (Papaver) 24: Common sunflower (Helianthus annuus) 26: Common bean (Phaseolus vulgaris) 27: Broad bean (Vicia faba) 29: Maple-leaf goosefoot (Chenopodium hybridum) 30: Black-bindweed (Fallopia convolvulus) 32: Red-root amaranth (Amaranthus retroflexus) 33: White goosefoot (Chenopodium album) 34: Thornapple (Datura stramonium) 38: Creeping thistle (Cirsium arvense) 39: Field sowthistle (Sonchus arvensis) 66: Redshank (Persicaria maculosa) 71: Cornflower (Centaurea cyanus) 72: Common corncockle (Agrostemma githago) 77: Ribwort plantain (Plantago lanceolata) 89: Copse bindweed (Fallopia dumetorum) 94: Soybean (Glycine max) File Naming Convention File names of images and annotations extend the image names in CropAndWeed with the row number of the extracted plant in the original object-detection annotations: <subset>-<session>-<image>-<row> Example ave-0045-0011-012: subset ave, session 45, image 11, the plant was extracted from row 12 of the corresponding csv-file (object detection) Citing If you use the CropAndWeedAndLeaf benchmark for your research, please cite the original paper: Martinko, R., Steininger, D., Simon, J., Trondl, A., Blaickner, M., 2026. ReLeaf: Benchmarking Leaf Segmentation across Domains and Species. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops.CropAndWeedAndLeaf是一款用于跨多种植物物种开展叶片分割模型基准测试的基准数据集，相关研究细节见于发表于2026年计算机视觉与模式识别会议（CVPR 2026）农业视觉研讨会的ReLeaf论文。该数据集基于CropAndWeed数据集的图像资源，将其裁剪为单株植物实例，并为单个叶片添加了实例分割掩码（instance-segmentation masks）。源码、模型及更多文档可在我们的GitHub页面获取。 ## 数据结构 本仓库包含以下子目录： - images：收录所有已标注植物实例的裁剪图像 - labels：存储与每张图像对应的标注文件，格式为YOLOv8实例分割格式 ## 植物物种 标签ID与CropAndWeed数据集中定义的原始标签一一对应： 1: 玉米（Zea mays） 7: 甜菜（Beta vulgaris s. vulgaris） 13: 豌豆（Pisum sativum） 14: 西葫芦（Cucurbita pepo var. gir.） 15: 南瓜属（Cucurbita） 18: 马铃薯（Solanum tuberosum） 22: 罂粟属（Papaver） 24: 普通向日葵（Helianthus annuus） 26: 普通菜豆（Phaseolus vulgaris） 27: 蚕豆（Vicia faba） 29: 杂配藜（Chenopodium hybridum） 30: 卷茎蓼（Fallopia convolvulus） 32: 反枝苋（Amaranthus retroflexus） 33: 藜（Chenopodium album） 34: 曼陀罗（Datura stramonium） 38: 加拿大蓟（Cirsium arvense） 39: 野苦苣菜（Sonchus arvensis） 66: 酸模叶蓼（Persicaria maculosa） 71: 矢车菊（Centaurea cyanus） 72: 麦仙翁（Agrostemma githago） 77: 长叶车前（Plantago lanceolata） 89: 篱蓼（Fallopia dumetorum） 94: 大豆（Glycine max） ## 文件命名规则 图像与标注文件的命名在CropAndWeed数据集的图像名称基础上，追加了原始目标检测标注中提取目标植物所在的行号，命名格式为：<子集>-<会话>-<图像>-<行号> 示例：ave-0045-0011-012：代表子集为ave、会话编号45、图像编号11，该植物从对应csv目标检测文件的第12行提取得到。 ## 引用说明 若您在研究中使用CropAndWeedAndLeaf基准数据集，请引用以下原始文献： Martinko, R., Steininger, D., Simon, J., Trondl, A., Blaickner, M., 2026. ReLeaf: Benchmarking Leaf Segmentation across Domains and Species. 载于《IEEE/CVF计算机视觉与模式识别会议（CVPR）研讨会论文集》