Densely Segmented Supermarket (D2S) dataset
收藏资源简介:
D2S数据集是一个专为工业领域实例感知语义分割设计的新型基准,包含21000张高分辨率图像,每张图像都有像素级的所有对象实例标签。该数据集涵盖了60个类别的日常商品,如水果、蔬菜、谷物包装、意大利面和瓶子等。数据集的训练图像仅包含单一类别对象,背景均匀,而验证和测试集则更为复杂多样。D2S数据集旨在模拟自动结账、库存或仓库系统的真实世界设置,通过不同的光照、旋转和背景变化来进一步测试实例分割方法的鲁棒性。数据集的标注精确无误,允许使用单个实例的裁剪进行人工数据增强。D2S数据集覆盖了该领域中几个高度相关的挑战,如有限的训练数据量和测试及验证集的高度多样性。
The D2S dataset is a novel benchmark tailored for industrial instance-aware semantic segmentation, consisting of 21,000 high-resolution images, each annotated with pixel-level instance labels for all objects present in the image. This dataset encompasses 60 categories of daily commodities, including fruits, vegetables, cereal packages, pasta, bottles and other similar everyday items. The training images of the dataset only contain single-category objects with uniform backgrounds, whereas the validation and test sets are far more complex and diverse. The D2S dataset is intended to simulate real-world scenarios of self-checkout, inventory management or warehouse systems, and evaluates the robustness of instance segmentation methods by incorporating variations in lighting, rotation and background conditions. The annotations of the D2S dataset are meticulously precise, allowing for manual data augmentation through cropping of individual object instances. This benchmark addresses several highly relevant challenges in the field, including limited training data scale and the pronounced diversity of the validation and test sets.

- 1MVTec D2S: Densely Segmented Supermarket DatasetMVTec Software GmbH · 2018年



