cswry

cswry / VOSR

Public

[CVPR2026] VOSR: A Vision-Only Generative Model for Image Super-Resolution

46
1
100% credibility
GitGems finds repos before they trend -- Star growth, AI reviews, and architecture deep-dives -- free with GitHub.
Sign Up Free
AI Analysis
Python
AI Summary

VOSR is a research framework providing models and tools to upscale low-resolution images generatively, excelling at preserving fine details like text.

How It Works

1
📸 Discover VOSR

You hear about this exciting tool that magically sharpens blurry photos, making details like text pop without losing quality.

2
⬇️ Gather the essentials

Download the ready-to-use sharpening packs and example fuzzy pictures to see it in action.

3
🖼️ Add your photos

Drop your own low-quality images into a folder, ready for transformation.

4
Pick your style
⚡
Fast one-step

Get speedy results perfect for everyday use.

⭐
Multi-step detailed

Unlock the highest quality for special photos.

5
✨ Hit go and watch the magic

Start the process and feel the excitement as your pictures come alive sharper than ever.

😍 Admire your sharp masterpieces

Celebrate seeing crisp, detailed images with perfect text and structures restored.

Sign up to see the full architecture

4 more

Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is VOSR?

VOSR is a Python generative image model for vision-only super-resolution, tackling real-world degradation like blur, noise, and compression without text prompts. It upsamples low-res inputs 4x via flow-matching diffusion, outputting crisp, detail-preserving results ideal for text-rich screens. Users run inference on pretrained 0.5B/1.4B models or train custom ones from YAML configs, with tiling for large images.

Why is it gaining traction?

CVPR 2026 acceptance highlights VOSR's edge over GANs and priors: multi-step sampling (25 steps) for fidelity, distilled one-step for speed, and top scores on the new ScreenSR benchmark for screen photos. Python scripts handle RealSR-style eval out-of-box, with CFG knobs tuning faithfulness vs. creativity—devs love the text recovery on diverse artifacts.

Who should use this?

Computer vision engineers building restoration pipelines for apps, researchers evaluating generative super-resolution on custom datasets, and backend devs processing user screenshots or documents where text sharpness matters.

Verdict

Test the ModelScope checkpoints via simple CLI inference—strong potential for production prototypes despite 1.0% credibility from 46 stars and fresh release. Docs and paper are solid; scale up training if your GPUs allow.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.