A Realistic Synthetic Mushroom Scenes Dataset.

In this work, we present the Realistic Synthetic Mushroom Scenes Dataset, which encompasses images depicting mushrooms in various settings in relatively cluttered scenes. The dataset is composed of 15,000 high-quality, realistic images with various useful annotations. The dataset can be leveraged to address problems associated with mushroom detection, instance segmentation, and 3D pose estimation. These tasks are of paramount importance in automating the mushroom harvesting process in mushroom farms, which is a challenging and costly procedure. Also, we proffer a three-step pipeline that can generate annotated and realistic synthetic images, commencing with a singular 3D model that can be easily applied to a range of crops beyond mushrooms (https://github.com/dafniana/SyntheticMushroom-Dataset).