Colorectal cancer (CRC) is one of the leading causes of cancer-related deaths, largely due to its development from precancerous polyps that can be detected during colonoscopy. However, the limited resolution of colonoscopy and Narrow Band Imaging (NBI) often hinders the accurate identification of these critical polyp structures. In this work, we propose a state-of-the-art super-resolution framework specifically designed to enhance the clarity and detail of colonoscopy polyp images. We extend the conventional Super-Resolution Generative Adversarial Network (SRGAN) by integrating custom content losses—including pixel-wise Mean Squared Error (MSE), VGG-based perceptual, and Structural Similarity (SSIM) losses—to improve the visual quality of polyp images. Additionally, we employ the transformer-based SwinIR model with transfer learning and compare the impact of different loss functions (MSE versus Mean Absolute Error (MAE)) on image reconstruction. Trained on the Piccolo dataset, which includes a diverse set of polyp images, our approach demonstrates significant improvements in image clarity and detail. These advancements can facilitate earlier and more accurate detection of polyps, ultimately contributing to improved CRC diagnosis and treatment outcomes.