Image Super Resolution
Upscale opaque photographs no larger than 224×224 pixels with a fixed 3× ONNX Model Zoo model. The neural model predicts luminance detail; browser interpolation supplies chroma. Newly inferred details are not verified original content.
Key features
- Run an Apache-2.0 licensed 3× luminance model in a browser worker
- Upscale opaque 16–224 px PNG, JPEG, or WebP images with aspect ratio preserved
- Terminate the worker on cancellation to release inference memory
- Compare source and output, then save PNG and versioned JSON report
- Show mean luminance difference from interpolation without calling it a quality score
How to use
- Choose an opaque 16–224 px photograph or run the synthetic example.
- Run 3× upscale to load the local model and infer in a browser worker.
- Inspect source, result, and luminance difference from interpolation.
- Save the PNG and processing report and compare directly with the original.
Use cases
- Compare learned enlargement with interpolation for a small photo
- Inspect edge artifacts and color shifts produced by inference
- Try a 3× upscale locally without uploading an image
Frequently asked questions
Does this recover the true missing detail?
No. The model predicts luminance patterns from training. It cannot verify or restore the true content of text, faces, or evidentiary photos.
Why are images above 224 pixels rejected?
The tested model has a fixed 224×224 input. Shrinking a larger photo first would not add information, so this version rejects it.
Which scale is available?
The model has a fixed horizontal and vertical 3× scale.
Is my image sent to a server?
No. Only the model and runtime download from this site. Pixel processing and export happen locally in your browser.
Does transparent PNG work?
No. The model uses luminance from opaque photographs, so transparent pixels are rejected explicitly.
Privacy
Your photo and inference result stay in your browser. The model and runtime are fetched as static files from this site; your photo is never uploaded.
Comments & questions