UNet Corm & Damage Semantic Segmentation Model

This repository contains the weights, performance logs, and hyperparameters for experiment unet_run_2026-06-23_13-16.

Model Hyperparameters

• Architecture Type: UNet (SMP wrapper)
• Backbone Encoder: UNet
• Pretrained Weights: imagenet
• Input Channels: 3
• Number of Classes: 3 (0: background, 1: damage, 2: root)
• Training Resolution: 256x256

Training Configurations

• Batch Size: 8
• Optimizers / Lr: AdamW / 0.0001
• Maximum Epochs: 200
• Seed Configuration: 42

Metrics Curves

Below are the training performance plots generated for this run: