Super-Resolution
Reconstructing high-resolution images from low-resolution inputs—turn 480p into sharp 4K.
Think of it like
A detective using CSI enhancement to zoom into a grainy security video.
Example
Real-ESRGAN upscales degraded photos (JPEG artifacts, noise) to 4K with plausible detail recovery.
How it actually works
Combines upsampling with artifact removal. Training on paired degraded/clean images or blind SR that handles any degradation. Perceptual loss (VGG features) outperforms pixel-level MSE for visual quality. GAN-based methods (ESRGAN) produce sharper results than diffusion but can hallucinate details.
For product teams
Recover missing pixels from old videos, low-res security footage, scans.
For engineers
Learned upsampling (sub-pixel convolution) plus residual blocks; trained with perceptual + adversarial loss.
Related
- Upscaling — Variant of.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome