The Image Segmentation for Plant Disease project focuses on developing a deep learning-based solution to segment leaf regions in images for identifying plant diseases. Utilizing the Leaf Disease Segmentation Dataset from Kaggle (https://www.kaggle.com/datasets/fakhrealam9537/leaf-disease-segmentation-dataset), the project implements three segmentation models: Segformer,  UNet++, DeepLabV3, both with ResNet34 encoders pretrained on ImageNet, and a custom Swin Transformer-based model with a convolutional decoder. The models are trained to generate binary masks, identifying leaf pixels labeled as 38 in grayscale masks, using data preprocessing with Albumentations for resizing, normalization, and augmentation. Training involves 10 epochs with Adam optimizer and Dice Loss (UNet++, DeepLabV3) or BCEWithLogitsLoss (Swin Transformer). Performance is evaluated using Dice Score and IoU metrics, with UNet++ achieving the highest scores (Dice: 0.8401, IoU: 0.7423), followed by DeepLabV3 (Dice: 0.7693, IoU: 0.6603), and Swin Transformer (Dice: 0.2461, IoU: 0.1554). The UNet++ model is deployed in a Flask web application, enabling user registration, login, and image upload for segmentation predictions, supported by a MySQL database for user management. This project demonstrates an effective approach to plant disease segmentation with practical deployment for real-world use.