Research map / Image generation / Image restoration

Random Is All You Need: Random Noise Injection on Feature Statistics for Generalizable Deep Image Denoising

ICLR 2025 · Image generation › Image restoration · approach: Self-supervised image denoising

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes random noise injection on feature statistics to train a generalizable deep image denoiser that handles unseen noise types, using only Gaussian noise (sigma=15) during training.

Problem

Generalizable deep image denoising: models trained on Gaussian noise (sigma=15) must denoise across unseen noise types such as speckle and Poisson noise. The abstract cites Masked Training as the current state of the art.

Approach

Injects random noise into feature statistics during training, as an alternative to the masked-input training approach of Masked Training. Only the abstract is available, so the exact mechanism is not described further.

Experimental setup

Datasets: Gaussian noise (sigma=15) training

Baselines: Masked Training

Builds on: Masked Training

Closest papers by meaning

• Masked Image Training for Generalizable Deep Image Denoising (CVPR 2023 · similarity 0.91)
• Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks (ICLR 2020 · similarity 0.89)
• Masked Pre-training Enables Universal Zero-shot Denoiser (NeurIPS 2024 · similarity 0.88)
• Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.88)
• Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable (CVPR 2025 · similarity 0.87)
• Transfer CLIP for Generalizable Image Denoising (CVPR 2024 · similarity 0.87)