Project 🚀
Collection
Some self-explaration works on transformer: audio, images, text, ...
•
4 items
•
Updated
•
1
For this version, the model has been trained for 50 epochs on the SRResNet pre-training phase and 100 epochs on the GAN training phase. Since purple tint is observed when trained for 15 epochs, this version is trained for longer to check if the issue persists.
SRGAN Model for Super-Resolution
This repository contains a Super-Resolution Generative Adversarial Network (SRGAN) model trained on the DIV2K dataset. The model is designed to enhance the resolution of low-resolution images using deep convolutional neural networks.
Experiment Implementation
The results as follows:
- PSNR: 18.99 dB
- SSIM: 0.6599