图像超分辨率 Image Super Resolution with SRGAN

Enhanced Deep Residual Networks for Single Image Super-Resolution vs Unsupervised Degradation Representation Learning for Blind Super-Resolution vs Transformer based encoder for Unsupervised Degradation Representation Learning for Blind Super-Resolution

Comparison of classification model accuracy with Low-Resolution input images vs Super-Resolution input images obtained using Deep Convolutional Network

No description available

Image Super Resolution with Deep Learning (Bicubic vs. SwinIR)

Implemented pothole detection using YOLOv9-Tiny with image enhancement via Stable Diffusion Super-Resolution. The experiment compares raw vs. enhanced images to evaluate detection accuracy on road imagery.

A GPU-accelerated super-resolution engine that takes a low-resolution image and outputs a high-resolution version using a pretrained ESRGAN model in PyTorch. It benchmarks inference speed and memory usage on CPU vs GPU, and includes an optional API endpoint.

Latent space fidelity bounds diffusion-based medical image super-resolution. Controlled comparison of domain-specific (MedVAE) vs. generic (SD-VAE) autoencoders across knee MRI, brain MRI, and chest X-ray.

Deep Learning final project exploring manga image super-resolution using SPAN. Includes data preprocessing on Manga109, comparisons with bicubic interpolation, pretrained vs randomly initialized models, and fine-tuning analysis with PSNR/SSIM evaluation.

Developed a Pix2Pix GAN to translate 500+ film negatives into positives using nonlinear mappings. Integrated ESRGAN for super-resolution, improving PSNR by 28%. Achieved 15% higher SSIM and 20% lower error vs. U-Net, and automated the pipeline for large-scale high-res image translation.