Repository containing code used for training object detection models for tomatoš disease identification.
Model training notebooks:
- train_yolov1.ipynb
- tomato_disease_detection_yolov8_nano.ipynb
- tomato_disease_detection_yolov8_small.ipynb
- Roboflow_EfficientDet_v2.ipynb
- Roboflow_Custom_Detectron2.ipynb
Results aggregation notebooks:
- Metrics_for_different_models.ipynb
We used train_yolov1.ipynb notebook to train YOLOv1 model, but model training took a lot of time compared to other notebooks, so we were able to train on a small number of epochs and did not include this model in the benchmark.
We trained nano and small versions of YOLOv8 using tomato_disease_detection_yolov8_nano.ipynb and tomato_disease_detection_yolov8_small.ipynb notebooks respectively.
The notebooks use ultralitics CLI to perform the task.
We used Roboflow_EfficientDet_v2.ipynb notebook to train EfficientDet model. The notebook is a slightly changed version of the official Roboflow tutorial notebook.
Notebook named Roboflow_Custom_Detectron2.ipynb has been used to train and evaluate Faster-RCNN model. Note: GitHub repository downloaded in that notebook has been changed slightly, so you may encounter some problems with different devices.
- Source code for mobile application.
- Model weights trained with these notebooks.