GrapesNet: Indian Grape Clusters RGB & RGB-D Image Datasets

There are four different RGB and RGB-D image datasets of grape clusters created in Indian vineyards are proposed in the GrapesNet:
1. Single grape cluster per image with artificial background and single grape cluster per image with real background
2. Multiple clusters per image with real background
3. RGB-D images of vineyard with natural environment
4. Single grape cluster per RGB-D image with some experimental set-ups and artificial environment.
All the datasets are taken at Indian vineyard in real environmental conditions with different light conditions, different camera angles and different occlusions to build the reliable and robust deep learning models for the real-time applications. White grapes are considered for creating the GrapesNet because most of the presently available vineyard datasets are on red grapes. Total 11000+ images are included in GrapesNet. It is strong believe that all these datasets are very helpful for training, testing and validating the grape cluster detection and segmentation using various deep learning models.

GrapesNet中提出了四种产自印度葡萄园的葡萄串RGB及RGB-D图像数据集，具体分类如下：
1. 单张图像仅含单串葡萄，分别搭配人工背景与真实背景两种场景
2. 单张图像包含多串葡萄且背景为真实自然环境
3. 自然环境下的葡萄园RGB-D图像数据集
4. 搭载特定实验设置与人工环境的单串葡萄RGB-D图像数据集

所有数据集均采集自印度葡萄园的真实环境，涵盖不同光照条件、拍摄角度与遮挡场景，旨在构建可适配实时应用的可靠且鲁棒的深度学习模型。本次数据集构建选取白葡萄作为研究对象，原因在于当前已公开的多数葡萄园数据集均以红葡萄为研究主体。GrapesNet总计包含11000余张图像。我们深信，该数据集将对基于各类深度学习模型的葡萄串检测与分割任务的训练、测试与验证工作提供有力支撑。