This repository contains the implementation of a Synthetic Aperture Radar (SAR) image colorization model using a UNet2DModel. The model transforms grayscale SAR images into their optical (colorized) counterparts, enhancing the interpretability of SAR data for various applications such as remote sensing and environmental monitoring.
- Encoder-Decoder Architecture: Built using the flexible UNet2DModel.
- Attention Mechanisms: Incorporated to improve the quality of colorization.
- Custom Dataset Support: Compatible with paired SAR and optical image datasets.
- High Performance: Optimized for efficient GPU usage and minimal memory consumption.
- Training Logs: Includes visualizations for loss, PSNR, and SSIM metrics.