YOLOv8 grapevine leaf blade and vein pixel segmentation mask

This repository contains a fully reproducible pipeline for grapevine leaf instance segmentation using YOLOv8. The code is designed to be a lightweight and easily distributable solution for researchers, viticulturists, and ampelographers to prepare their own data, train a state-of-the-art model, and perform inference.

Key Components:

0_prepare_yolo_dataset.py: A preprocessing script that takes raw image and annotation data and converts it into the necessary YOLOv8 format, including creating segmentation masks and labels for training. It handles multiple directory structures and automatically splits the data into training, validation, and testing sets.

1_train_yolo.py: This script trains a YOLOv8n-seg model on the prepared dataset. It utilizes a lightweight architecture to achieve high performance with minimal computational resources, making it suitable for a wide range of hardware.

2_inference.py: A final script to perform inference on new images, saving the predictions in an organized folder structure for easy visualization and analysis.

How to Reproduce the Analysis

To reproduce the analysis and train a custom model, place your raw image and annotation data in the ./data/ folder and then run the scripts in numerical order. Please note: all data is provided to reproduce buliding and running inference using the model. However, you can add your own data as a new folder to ./data/ and then update the 0_prepare_yolo_dataset.py script to accomodate the new data.