3D2cut Single Guyot Dataset
Description
The use of Deep Learning for precision agriculture is growing. This dataset consists of 1511 grapevine images, fully and precisely annotated with the vine plant structure information. Generally, there is a single grapevine plant centered on each images. A blue or white background sheet is used as a backdrop to isolate the main vine plant from the rest of the vineyard. The images were collected in 3 different vineyards, from 3 different French regions. The images were captured using smartphones or digital cameras, with a typical resolution of 4032x3024.
The dataset is split into a 'Training Set' of 1254 images and a test set of 257 images in the 'Independent Test Set'.
All branches of the main vines are fully annotated with their nodes and terminations as explained below. The annotations are contained in JSON files, so that it is easy to understand and parse. For each image, there is an annotation file called `<image file name>_annotation.json` that contains the annotation as well as metadata about the vine image.
This dataset was created and published by 3D2cut SA.
Reference
If you use this dataset, please cite the following publication:
Towards Smart Pruning: ViNet, a Deep-Learning Approach for Grapevine Structure Estimation, Computers and Electronics in Agriculture, accepted, February 2023.