This project focuses on satellite image segmentation for the Dubai region. Key aspects include resizing images and generating masked segmentation images. The data was prepared for deep learning using Patchify, and a U-Net model was built with encoder-decoder architecture. Custom loss functions and metrics were used, and the training process was tracked with Weights and Biases. The activation outputs and gradients for each TensorFlow layer were visualized using Keract. Finally, the model was deployed on Hugging Face for further use and experimentation.