1xDeNoise RealPLKSR OTF

Removes noise from photos without changing their size.

A 1× RealPLKSR model that strips sensor and compression noise from photos while keeping detail. It was fine-tuned from a JPEG-removal model on the Nomos-v2 dataset.

Download ONNX View repository Paper

Denoising Computer Vision

Scale: 1×
Parameters: 7.4M
CPU time per 256 px tile: 3.8s
ONNX file: 30MB

Example output

Drag the handle to compare. Produced by running this model.

After: Input: a portrait with added sensor noise. Output: the model's denoised result at the same size. — Input: a portrait with added sensor noise. Output: the model's denoised result at the same size.

Before: Input: a portrait with added sensor noise. Output: the model's denoised result at the same size. — Input: a portrait with added sensor noise. Output: the model's denoised result at the same size.

Technical details

Architecture: RealPLKSR (partial large kernel CNN)
Fine-tuned from: 1xDeJPG_realplksr_otf
Training data: Nomos-v2, 6,000 photos, on-the-fly degradations
Training: 200,000 iterations
ONNX build: FP32, opset 17, any image size
Input: RGB, NCHW, values 0–1
Benchmark: ONNX Runtime CPU, Apple M2 Pro
Original release: 8 August 2024

Good for

Night and low-light photos
High-ISO shots
Noisy scans

Run it in Flutter

This model runs as-is with our open-source flutter_super_resolution package (MIT), which tiles large images and uses Core ML on Apple devices when available. It works on iOS, Android (ARM devices), macOS and Windows.

pubspec.yaml · Dartdncnn.onnx

dependencies:
  flutter_super_resolution: ^1.1.1

// Load the model, then upscale a ui.Image
final upscaler = FlutterUpscaler(tileSize: 128, overlap: 8);
await upscaler.initializeModel('assets/models/dncnn.onnx');

final result = await upscaler.upscaleImage(
  image,
  scale: 1,
  onProgress: (progress, message) => print('$message ${(progress * 100).round()}%'),
);

upscaler.dispose();

Package on GitHub All models that work with it