Super Resolution using ESRGAN

This project aims on achieving super resolution through pretrained ESRGAN model on Keras.

Samples:

1. Leftmost image - Original (Low resolution image)
2. Middle image - Bicubic (x4) image
3. Rightmost image - Super-resoluted image (ESRGAN Super resolution O/P)

Metrics:

1. PSNR value achieved for the above super-resoluted sample - 26.097332

Takeaways:

1. ESRGANs perform very well on the real-time samples regardless of the colors.
2. In the provided notebook, Super resolution is achieved not only for images, but for videos too :)

References:

1. http://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Wang_ESRGAN_Enhanced_Super-Resolution_Generative_Adversarial_Networks_ECCVW_2018_paper.pdf
2. https://www.semanticscholar.org/paper/ESRGAN%3A-Enhanced-Super-Resolution-Generative-Wang-Yu/1bdd30a8acc75c58a1bdd4daa4545d5f3971a826