ESRGAN-X

👀 Enhanced Super-Resolution Experiments using ESRGAN

Overview

This repository contains experiments adapting Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) for high-quality image super-resolution, transforming low-resolution images into sharp, high-resolution outputs.

Key Features

• 🧠 From-Scratch Implementation — No pretrained weights or third-party frameworks used.
• 🎨 Enhanced Image Resolution — Restores textures, edges, and fine details.
• ⚙️ Optimized Architecture — Uses Residual-in-Residual Dense Blocks (RRDB) with stability refinements.

Contents

• esrgan.ipynb: Full training and evaluation pipeline
• report.pdf: Methodology and results

Installation

git clone https://github.com/anthonyhuang19/Adapting-ESRGAN-for-Enhanced-Super-Resolution-Experiments.git
cd ESRGANX

Tools & Technologies

Frameworks

• PyTorch — Core deep learning framework for building and training ESRGAN.

Image Processing

• OpenCV — Image manipulation, resizing, augmentation, and I/O.
• Pillow (PIL) — Flexible image loading and pixel-level preprocessing.

Data Handling

• NumPy — Tensor operations and numerical computation.
• Pandas — Managing metadata, experiment logs, and training statistics.

Visualization & Monitoring

• Matplotlib — Plot training curves (loss, PSNR, SSIM) and visualize outputs.
• Seaborn — Enhances visualization aesthetics.

Experimental Workflow

1. Data Preprocessing

• Resize images to consistent shape.
• Normalize pixel values for stable training.

2. Model Training

• Optimize adversarial loss to improve perceptual quality and fine-detail restoration.

3. Model Evaluation

• Metrics: PSNR and SSIM
• Visual comparisons of generated vs. ground truth images.

Contributing

1. Fork the repository
2. Modify code or add features
3. Submit a Pull Request

Report issues or suggestions via the Issues tab.

Contact

• GitHub: @anthonyhuang19

---