Denoising Textures is a Different Ballgame Altogether!

Saptarshi Mandal; Oindrila Haldar; Chirayata Bhattacharya; Chandra Sekhar Seelamantula; Zhiwei Xu; Thrasyvoulos N. Pappas; Gaurav Sharma

eess.IVOct 15, 2026

Denoising Textures is a Different Ballgame Altogether!

Authors: Saptarshi Mandal, Oindrila Haldar, Chirayata Bhattacharya, Chandra Sekhar Seelamantula, Zhiwei Xu, Thrasyvoulos N. Pappas, Gaurav Sharma

Organizations: Department of Electrical Engineering, Indian Institute of Science, Bangalore, India · Department of Electrical and Computer Engineering, Northwestern University, Evanston, Illinois, USA · Department of Electrical and Computer Engineering, University of Rochester, Rochester, NY, USA

Abstract

The human brain effortlessly deciphers colors and textures from visual stimuli and reconstructs the world around us, yet preserving such details remains challenging in image denoising. While significant progress has been made in generic image denoising, the problem of texture denoising has remained relatively under-explored. Fine details, such as intricate patterns and repetitive structures, often suffer from over-smoothing even in deep learning based approaches for denoising. In this paper, we analyze the behavior of state-of-the-art (SOTA) denoisers on texture datasets and show that performance improvements are possible only by fine-tuning the denoisers on the texture datasets. Surprisingly, fine-tuning on texture datasets result in the models catastrophically forgetting the original dataset. These findings highlight the need for specialized treatment of textures in denoising. Since denoising quality directly impacts downstream tasks such as generative modelling, we argue that dedicated texture aware denoisers are essential for faithful texture reconstruction and synthesis.

Figures & tables

Panel of Figure 1 — Fig. 1 : Comparison of a noisy and denoised texture using pretrained MIRNet [ 23 ] and its finetuned version (MIRNet † ). PSNR and SSIM values improve after finetuning.

Panel of Figure 2 — Fig. 2 : Visual comparison of various denoising models on the NUUR-Texture500 [ 14 ] dataset. Top row : noisy and pre-trained model results, bottom row : reference and (Model † ) fine-tuned models. PSNR is shown for each result relative to the reference.

Panel of Figure 3 — Fig. 3 : Visual comparison of denoising models on the SIDD [ 1 ] . Top row : noisy and pre-trained models, bottom row : reference and (Model † ) fine-tuned models. PSNR (dB) shown w.r.t. the reference.

Model	PSNR $\uparrow$	SSIM $\uparrow$	MS-SSIM $\uparrow$	CW-SSIM $\uparrow$	DISTS $\downarrow$	STSIM1 $\uparrow$	STSIM2 $\uparrow$	LPIPS $\downarrow$
Noisy	26.23 $\pm$ 0.36	0.860 $\pm$ 0.123	0.976 $\pm$ 0.024	0.962 $\pm$ 0.077	0.157 $\pm$ 0.076	0.989 $\pm$ 0.023	0.941 $\pm$ 0.052	0.1286 $\pm$ 0.1488
CAT [ 4 ]	27.73 $\pm$ 4.542	0.834 $\pm$ 0.149	0.956 $\pm$ 0.074	0.931 $\pm$ 0.127	0.165 $\pm$ 0.077	0.954 $\pm$ 0.092	0.926 $\pm$ 0.062	0.1904 $\pm$ 0.1778
CAT †	34.17 $\pm$ 3.389	0.959 $\pm$ 0.032	0.990 $\pm$ 0.009	0.973 $\pm$ 0.054	0.062 $\pm$ 0.034	0.997 $\pm$ 0.006	0.949 $\pm$ 0.050	0.0304 $\pm$ 0.0283
CycleISP [ 22 ]	25.52 $\pm$ 3.759	0.781 $\pm$ 0.192	0.933 $\pm$ 0.099	0.872 $\pm$ 0.212	0.239 $\pm$ 0.089	0.929 $\pm$ 0.136	0.911 $\pm$ 0.077	0.3047 $\pm$ 0.2425
CycleISP †	34.20 $\pm$ 3.346	0.959 $\pm$ 0.032	0.998 $\pm$ 0.005	0.973 $\pm$ 0.053	0.062 $\pm$ 0.034	0.998 $\pm$ 0.005	0.948 $\pm$ 0.050	0.0306 $\pm$ 0.0296
MIRNet [ 23 ]	25.23 $\pm$ 5.765	0.721 $\pm$ 0.204	0.917 $\pm$ 0.104	0.913 $\pm$ 0.144	0.232 $\pm$ 0.081	0.909 $\pm$ 0.127	0.909 $\pm$ 0.070	0.3425 $\pm$ 0.2464

TABLE I : Denoising performance (mean

\pm

std) on NUUR-Texture500 dataset [ 14 ] using PSNR, SSIM, MS-SSIM, CW-SSIM, DISTS, STSIM1, STSIM2 and LPIPS . Dagger ( † ) indicates fine-tuned models.

Model	PSNR $\uparrow$	SSIM $\uparrow$
Noisy	27.86 $\pm$ 5.17	0.52 $\pm$ 0.23
MIRNet [ 23 ]	42.59 $\pm$ 4.08	0.96 $\pm$ 0.06
MIRNet †	32.68 $\pm$ 6.75	0.72 $\pm$ 0.26
MPRNet [ 24 ]	42.60 $\pm$ 4.10	0.96 $\pm$ 0.06
MPRNet †	32.48 $\pm$ 6.80	0.72 $\pm$ 0.26
CycleISP [ 22 ]	42.21 $\pm$ 4.12	0.95 $\pm$ 0.06

TABLE II : Original vs. fine-tuned ( † ) model performance on SIDD [ 1 ] .

Image Denoising

Image denoising removes unwanted noise while preserving the edges, textures, and colors needed for faithful images.

See all 157 papers

Generative Texture Filtering
Apr 21, 2026Rongjia Zheng, Shangwei Huang, Lei Zhu +2
Sun Yat-sen University, China · Hong Kong University of Science and Technology (Guangzhou), China
Colorful-Noise: Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image Generation
May 1, 2026Nadav Z. Cohen, Ofir Abramovich, Ariel ShamirGeometry-Conditioned Image Generation Conditional Image Generation
Reichman University, Israel
SONIC: Spectral Optimization of Noise for Inpainting with Consistency
Nov 25, 2025Seungyeon Baek, Erqun Dong, Shadan Namazifard +2Diffusion Model Inversion Image Inpainting
University of British Columbia, Canada · Google DeepMind, USA

Apr 21, 2026J/K move · Enter open · S save

Generative Texture Filtering

Rongjia Zheng, Shangwei Huang, Lei Zhu, Wei-Shi Zheng+1

Sun Yat-sen University, China · Hong Kong University of Science and Technology (Guangzhou), China

Open paper AskPDF arXiv

We present a generative method for texture filtering, which exhibits surprisingly good performance and generalizability. Our core idea is to empower texture filtering by taking full advantage of the strong learned image prior of pre-trained generative models. To this end, we propose to fine-tune a pre-trained generative model via a two-stage strategy. Specifically, we first conduct supervised fine-tuning on a very small set of paired images, and then perform reinforcement fine-tuning on a large-scale unlabeled dataset under the guidance of a reward function that quantifies the quality of texture removal and structure preservation. Extensive experiments show that our method clearly outperforms previous methods, and is effective to deal with previously challenging cases. Our code is available at https://github.com/OnlyZZZZ/Generative_Texture_Filtering.

Abstract

Image Denoising

Explore similar work

Generative Texture Filtering

Colorful-Noise: Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image Generation

SONIC: Spectral Optimization of Noise for Inpainting with Consistency