SwinIR (Denoise) — Denoise AI Model
SwinIR strips noise, grain and speckle out of a photograph with a Swin Transformer, at a preset matched to how bad it is — 15, 25 or 50. Built for low-light frames, high-ISO shots and scans. Five to fifteen seconds.
What it does
SwinIR Denoiser removes noise from a degraded image. It is a unified restoration framework built on the Swin Transformer, and it handles grayscale and color at configurable noise levels.
Noise level is a setting rather than something detected, so it is chosen to match the source: 15 for light noise, 25 by default for moderate, 50 for heavy. A mismatch either under-denoises or over-smooths.
When to use it
- Removing sensor noise and grain from photographs
- Grayscale denoising at a chosen noise level of 15, 25 or 50
- Color denoising at a chosen noise level of 15, 25 or 50
- Restoring real-world noisy images rather than synthetic test cases
- Batch cleaning a collection of noisy images in one pass
Details
- CategoryDenoise
- Year2021
- LicenseApache-2.0
Compliance & Provenance
- ProviderOpen-source · Specialized
- EU AI Act RiskMinimal Risk
- Art. 50 TransparencyNot applicable
Inputs & Outputs
- ImageInputimage
Noisy source image
- ImageOutputimage
Denoised output image
Parameters
- noise_levelintegerdefault 25options: 15, 25, 50
Noise level (15/25/50)
Severity preset matching the input's noise level. 15 — light noise (clean smartphone photo). 25 (default) — moderate noise. 50 — heavy noise (low-light, high-ISO). Pick the closest preset; mismatched levels can either under-denoise or over-smooth.
Tags
Alternatives in Image Restoration
- Image Colorization
Auto-colorize B&W photos (DISCO). 10-30s.
- JPEG Quality Restoration
Remove JPEG compression artifacts (SwinIR). 5-15s.
- Motion Blur Removal (MSSNet)
Balanced motion-blur removal (MSSNet). 5-15s.
- Real-World Blur Removal
Diffusion-based deblur (highest quality). 1-3 min.