Tomato-leaf-Segmentation

An Interactive model designed for auto-segmentation for domain specific: Tomato Leaf segmentation.

Model Description

This model is a U-Net with a lightweight MobileNetV2 backbone, designed for interactive segmentation. It takes a 4-channel input (RGB image + user scribble) and outputs a binary segmentation mask. This model was trained on the LeafNet75/Annotated_Benchmarks400 dataset.

Training Performance

The model was trained for 60 epochs with early stopping (patience=10).

Final Validation Metrics:

• Loss: 0.0261
• Dice Score: 0.9475
• IoU (Jaccard): 0.9475
• Precision: 0.9711
• Recall: 0.9750

Training History

Sample Predictions