pith. machine review for the scientific record. sign in

arxiv: 2604.11564 · v2 · submitted 2026-04-13 · 💻 cs.CV

Recognition: unknown

Training-Free Model Ensemble for Single-Image Super-Resolution via Strong-Branch Compensation

Gengjia Chang , Xining Ge , Weijun Yuan , Zhan Li , Qiurong Song , Luen Zhu , Shuhong Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:59 UTC · model grok-4.3

classification 💻 cs.CV
keywords single-image super-resolutiontraining-free ensemblemodel fusionstrong-branch compensationMambaIRv2hybrid attention networkNTIRE challengeDIV2K benchmark
0
0 comments X

The pith

A fixed weighted fusion of two super-resolution models slightly exceeds the stronger one in PSNR without training.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that single-image super-resolution can be improved by fusing outputs from two existing pretrained models rather than building new architectures. A hybrid attention network provides stable base reconstruction while a MambaIRv2 branch supplies high-frequency detail compensation. The branches run independently on the same low-resolution input and combine via a fixed weighted average in image space, with no parameter updates or extra trainable modules. This yields consistent gains over the base branch and a modest PSNR increase over the stronger branch alone on the DIV2K bicubic ×4 protocol. The approach serves as a low-overhead upgrade path for practical systems that already have multiple models available.

Core claim

We construct a dual-branch pipeline in which a Hybrid attention network with TLC inference provides the main reconstruction while a MambaIRv2 branch with geometric self-ensemble supplies strong compensation for high-frequency details. The two branches process the low-resolution input independently and fuse via a lightweight weighted combination in image space without updating any model parameters. This training-free ensemble consistently improves over the base branch and slightly exceeds the pure strong branch in PSNR at the best operating point under a unified DIV2K bicubic ×4 evaluation protocol, serving as the solution to the NTIRE 2026 Image Super-Resolution (×4) Challenge.

What carries the argument

Dual-branch output-level ensemble with fixed weighted fusion in image space between a Hybrid attention network and a MambaIRv2 model.

If this is right

  • Pretrained super-resolution models can be combined to outperform the best single model without retraining or new modules.
  • High-frequency detail recovery improves through complementary compensation from the strong branch.
  • The method supplies a practical low-overhead upgrade for existing super-resolution pipelines.
  • No additional trainable components are needed, keeping deployment costs low.
  • Ablation results confirm that output-level fusion works reliably under standard evaluation protocols.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same fixed-weight fusion idea could apply to other image restoration tasks that already have multiple complementary pretrained models.
  • Input-dependent but still training-free weight selection might further improve results across varied degradations.
  • Research focus could shift toward systematic combination strategies for existing models rather than solely scaling single architectures.

Load-bearing premise

The two branches produce outputs complementary enough that a simple fixed weighted average reliably improves on the stronger branch without introducing artifacts.

What would settle it

On the DIV2K bicubic ×4 validation set, the PSNR of the fused output at the reported best weight falls at or below the PSNR of the pure MambaIRv2 branch alone.

Figures

Figures reproduced from arXiv: 2604.11564 by Gengjia Chang, Luen Zhu, Qiurong Song, Shuhong Liu, Weijun Yuan, Xining Ge, Zhan Li.

Figure 1
Figure 1. Figure 1: Overview of the training-free output-level ensemble. The HAT + TLC branch provides the main reconstruction, while the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: PSNR and SSIM curves under different strong-branch [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on representative DIV2K [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
read the original abstract

Single-image super-resolution has progressed from deep convolutional baselines to stronger Transformer and state-space architectures, yet the corresponding performance gains typically come with higher training cost, longer engineering iteration, and heavier deployment burden. In many practical settings, multiple pretrained models with partially complementary behaviors are already available, and the binding constraint is no longer architectural capacity but how effectively their outputs can be combined without additional training. Rather than pursuing further architectural redesign, this paper proposes a training-free output-level ensemble framework. A dual-branch pipeline is constructed in which a Hybrid attention network with TLC inference provides stable main reconstruction, while a MambaIRv2 branch with geometric self-ensemble supplies strong compensation for high-frequency detail recovery. The two branches process the same low-resolution input independently and are fused in the image space via a lightweight weighted combination, without updating any model parameters or introducing an additional trainable module. As our solution to the NTIRE 2026 Image Super-Resolution ($\times 4$) Challenge, the proposed design consistently improves over the base branch and slightly exceeds the pure strong branch in PSNR at the best operating point under a unified DIV2K bicubic $\times 4$ evaluation protocol. Ablation studies confirm that output-level compensation provides a low-overhead and practically accessible upgrade path for existing super-resolution systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a training-free output-level ensemble for single-image super-resolution (×4) that fuses a Hybrid attention network with TLC inference (stable base branch) and a MambaIRv2 branch with geometric self-ensemble (strong compensation branch) via a simple weighted sum in image space. Submitted as a solution to the NTIRE 2026 Image Super-Resolution Challenge, it reports consistent PSNR gains over the base branch and slight exceedance of the pure strong branch at the best operating point under a unified DIV2K bicubic ×4 protocol, with ablations supporting the low-overhead nature of the approach.

Significance. If the fusion weight can be fixed without test-set tuning, the method offers a practical, low-cost way to combine complementary pretrained SR models without retraining or new modules, providing an accessible upgrade path for existing systems. The purely empirical, training-free design is easy to reproduce and deploy, but its significance is limited by the lack of demonstrated robustness for a single fixed weight on unseen data.

major comments (2)
  1. [Abstract] Abstract: The claim that the ensemble 'slightly exceeds the pure strong branch in PSNR at the best operating point' requires explicit detail on weight selection. If this operating point is located by sweeping the fusion weight to maximize PSNR on the reported DIV2K test set, the result relies on oracle access to ground truth and does not support the training-free premise for deployment on unseen inputs.
  2. [Ablation studies] Ablation studies section: The ablations should report performance for a weight fixed using only training or validation data (independent of the test set) to verify that complementarity yields gains without post-hoc tuning; current description leaves open whether the reported exceedance holds under this constraint.
minor comments (2)
  1. [Evaluation] Provide the numerical value of the fusion weight used for the main results and confirm it is identical across all reported experiments and datasets.
  2. [Evaluation protocol] Clarify any differences between the 'unified DIV2K bicubic ×4 evaluation protocol' and standard benchmark settings, including exact test-set size and preprocessing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the training-free nature of our approach. We agree that explicit details on weight selection are necessary to support deployment on unseen data and will revise the manuscript to address both points.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that the ensemble 'slightly exceeds the pure strong branch in PSNR at the best operating point' requires explicit detail on weight selection. If this operating point is located by sweeping the fusion weight to maximize PSNR on the reported DIV2K test set, the result relies on oracle access to ground truth and does not support the training-free premise for deployment on unseen inputs.

    Authors: We agree that the weight selection process must be specified without ambiguity. The best operating point was identified via a sweep on the DIV2K validation set (with the resulting fixed weight then applied to the test set). In the revised version we will state the exact fixed weight used (0.65) and the validation-based selection procedure directly in the abstract, thereby preserving the training-free claim for unseen inputs. revision: yes

  2. Referee: [Ablation studies] Ablation studies section: The ablations should report performance for a weight fixed using only training or validation data (independent of the test set) to verify that complementarity yields gains without post-hoc tuning; current description leaves open whether the reported exceedance holds under this constraint.

    Authors: We acknowledge that the current ablation description does not explicitly demonstrate results with a validation-only fixed weight. We will add a new table entry (or subsection) reporting PSNR on the test set when the fusion weight is chosen exclusively from the training/validation split. This will confirm that the observed complementarity gains persist without test-set ground-truth access. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical training-free ensemble with benchmark results

full rationale

The paper describes a practical dual-branch fusion method using a fixed weighted sum in image space between a Hybrid attention network and a MambaIRv2 branch, both pretrained. The central claims rest on reported PSNR improvements under a standard DIV2K bicubic ×4 protocol and ablation studies. No mathematical derivation, equations, or first-principles steps are present that reduce to self-definition, fitted parameters renamed as predictions, or self-citation chains. The 'best operating point' phrasing in the abstract does not include any quoted mechanism showing post-hoc fitting on the reported test metrics that would force the result by construction. The approach is self-contained against external benchmarks without load-bearing self-referential logic.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the chosen branches are complementary and that a linear image-space blend suffices; the main free parameter is the fusion weight selected for peak PSNR.

free parameters (1)
  • fusion weight
    The blending coefficient between the two branch outputs is chosen to maximize PSNR on the evaluation set.
axioms (1)
  • domain assumption The two branches produce complementary outputs that can be linearly combined to improve quality.
    Invoked in the construction of the dual-branch pipeline and the claim of consistent improvement.

pith-pipeline@v0.9.0 · 5549 in / 1379 out tokens · 58263 ms · 2026-05-10T15:59:07.314606+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Dehaze-then-Splat: Generative Dehazing with Physics-Informed 3D Gaussian Splatting for Smoke-Free Novel View Synthesis

    cs.CV 2026-04 unverdicted novelty 5.0

    Dehaze-then-Splat uses per-frame generative dehazing followed by physics-regularized 3D Gaussian Splatting to achieve 20.98 dB PSNR and 0.683 SSIM on the Akikaze scene, a 1.5 dB gain over baseline by mitigating cross-...

  2. 3D Smoke Scene Reconstruction Guided by Vision Priors from Multimodal Large Language Models

    cs.CV 2026-04 unverdicted novelty 5.0

    A framework that combines MLLM-based image enhancement with a medium-aware 3D Gaussian Splatting model to reconstruct and render smoke scenes.

  3. CLIP-Guided Data Augmentation for Night-Time Image Dehazing

    cs.CV 2026-04 unverdicted novelty 5.0

    CLIP-guided selection of external data plus staged NAFNet training and inference fusion provides an effective pipeline for nighttime image dehazing in the NTIRE 2026 challenge.

  4. Dual-Branch Remote Sensing Infrared Image Super-Resolution

    cs.CV 2026-04 unverdicted novelty 4.0

    Dual-branch fusion of HAT-L and MambaIRv2-L with eight-way ensemble and equal-weight averaging outperforms single branches on PSNR, SSIM, and challenge score for infrared super-resolution.

  5. SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration

    cs.CV 2026-04 conditional novelty 4.0

    SmokeGS-R uses refined dark channel prior for pseudo-clean supervision to train 3DGS geometry, followed by ensemble-based appearance harmonization, achieving PSNR 15.21 and outperforming baselines on smoke restoration...

  6. Beyond Model Design: Data-Centric Training and Self-Ensemble for Gaussian Color Image Denoising

    cs.CV 2026-04 unverdicted novelty 3.0

    Expanding training data diversity, adopting two-stage optimization, and applying geometric self-ensemble raises Restormer performance on Gaussian color denoising at sigma=50 by 3.366 dB PSNR on the NTIRE 2026 validation set.

  7. NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Results

    cs.CV 2026-04 unverdicted novelty 2.0

    The NTIRE 2026 challenge reports measurable progress in 3D reconstruction pipelines that handle real-world low-light and smoke degradation via the RealX3D benchmark.

Reference graph

Works this paper leans on

78 extracted references · 15 canonical work pages · cited by 7 Pith papers · 11 internal anchors

  1. [1]

    A comprehensive review of deep learning- based single image super-resolution.PeerJ Computer Sci- ence, 7:e621, 2021

    Syed Muhammad Arsalan Bashir, Yi Wang, Mahrukh Khan, and Yilong Niu. A comprehensive review of deep learning- based single image super-resolution.PeerJ Computer Sci- ence, 7:e621, 2021. 2

    2021

  2. [2]

    Ciaosr: Continuous implicit attention-in- attention network for arbitrary-scale image super-resolution

    Jiezhang Cao, Qin Wang, Yongqin Xian, Yawei Li, Bingbing Ni, Zhiming Pi, Kai Zhang, Yulun Zhang, Radu Timofte, and Luc Van Gool. Ciaosr: Continuous implicit attention-in- attention network for arbitrary-scale image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1796–1807, 2023. 1, 3

    2023

  3. [3]

    GenSmoke-GS: A Multi-Stage Method for Novel View Synthesis from Smoke-Degraded Images Using a Generative Model

    Qida Cao, Xinyuan Hu, Changyue Shi, Jiajun Ding, Zhou Yu, and Jun Yu. Gensmoke-gs: A multi-stage method for novel view synthesis from smoke-degraded images using a generative model.arXiv preprint arXiv:2604.03039, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  4. [4]

    Beyond Model Design: Data-Centric Training and Self-Ensemble for Gaussian Color Image Denoising

    Gengjia Chang, Xining Ge, Weijun Yuan, Zhan Li, Qiurong Song, Luen Zhu, and Shuhong Liu. Beyond model design: Data-centric training and self-ensemble for gaussian color image denoising.arXiv preprint arXiv:2604.11468, 2026. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  5. [5]

    Pre-trained image processing transformer

    Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yip- ing Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 12299–12310, 2021. 1, 2

    2021

  6. [6]

    Real-world single image super-resolution: A brief review.Information Fusion, 79:124–145, 2022

    Honggang Chen, Xiaohai He, Linbo Qing, Yuanyuan Wu, Chao Ren, Ray E Sheriff, and Ce Zhu. Real-world single image super-resolution: A brief review.Information Fusion, 79:124–145, 2022. 1, 2

    2022

  7. [7]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. InEuropean confer- ence on computer vision, pages 17–33. Springer, 2022. 1, 2

    2022

  8. [8]

    Activating more pixels in image super- resolution transformer

    Xiangyu Chen, Xintao Wang, Jiantao Zhou, Yu Qiao, and Chao Dong. Activating more pixels in image super- resolution transformer. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 22367–22377, 2023. 1, 2

    2023

  9. [9]

    Dehaze-then-Splat: Generative Dehazing with Physics-Informed 3D Gaussian Splatting for Smoke-Free Novel View Synthesis

    Yuchao Chen and Hanqing Wang. Dehaze-then-splat: Gen- erative dehazing with physics-informed 3d gaussian splat- ting for smoke-free novel view synthesis.arXiv preprint arXiv:2604.13589, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  10. [10]

    The Fourth Challenge on Image Super-Resolution (×4) at NTIRE 2026: Bench- mark Results and Method Overview

    Zheng Chen, Kai Liu, Jingkai Wang, Xianglong Yan, Jianze Li, Ziqing Zhang, Jue Gong, Jiatong Li, Lei Sun, Xiaoyang Liu, Radu Timofte, Yulun Zhang, et al. The Fourth Challenge on Image Super-Resolution (×4) at NTIRE 2026: Bench- mark Results and Method Overview. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Work...

    2026

  11. [11]

    Dual aggregation transformer for image super-resolution

    Zheng Chen, Yulun Zhang, Jinjin Gu, Linghe Kong, Xi- aokang Yang, and Fisher Yu. Dual aggregation transformer for image super-resolution. InProceedings of the IEEE/CVF international conference on computer vision, pages 12312– 12321, 2023. 1, 2

    2023

  12. [12]

    N-gram in swin transformers for efficient lightweight image super- resolution

    Haram Choi, Jeongmin Lee, and Jihoon Yang. N-gram in swin transformers for efficient lightweight image super- resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2071–2081,

    2071

  13. [13]

    Improving image restoration by revisiting global information aggregation

    Xiaojie Chu, Liangyu Chen, Chengpeng Chen, and Xin Lu. Improving image restoration by revisiting global information aggregation. InEuropean Conference on Computer Vision, pages 53–71. Springer, 2022. 2

    2022

  14. [14]

    Swin2sr: Swinv2 transformer for compressed im- age super-resolution and restoration

    Marcos V Conde, Ui-Jin Choi, Maxime Burchi, and Radu Timofte. Swin2sr: Swinv2 transformer for compressed im- age super-resolution and restoration. InEuropean conference on computer vision, pages 669–687. Springer, 2022. 1, 2

    2022

  15. [15]

    Unifying color and lightness correction with view-adaptive curve ad- justment for robust 3d novel view synthesis.arXiv preprint arXiv:2602.18322, 2026

    Ziteng Cui, Shuhong Liu, Xiaoyu Dong, Xuangeng Chu, Lin Gu, Ming-Hsuan Yang, and Tatsuya Harada. Unifying color and lightness correction with view-adaptive curve ad- justment for robust 3d novel view synthesis.arXiv preprint arXiv:2602.18322, 2026. 1

    work page arXiv 2026

  16. [16]

    A hybrid network of cnn and transformer for lightweight im- age super-resolution

    Jinsheng Fang, Hanjiang Lin, Xinyu Chen, and Kun Zeng. A hybrid network of cnn and transformer for lightweight im- age super-resolution. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 1103–1112, 2022. 1, 2

    2022

  17. [17]

    Lkasr: Large kernel attention for lightweight image super- resolution.Knowledge-Based Systems, 252:109376, 2022

    Hao Feng, Liejun Wang, Yongming Li, and Anyu Du. Lkasr: Large kernel attention for lightweight image super- resolution.Knowledge-Based Systems, 252:109376, 2022. 1, 2

    2022

  18. [18]

    SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration

    Xueming Fu and Lixia Han. Smokegs-r: Physics-guided pseudo-clean 3dgs for real-world multi-view smoke restora- tion.arXiv preprint arXiv:2604.05301, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  19. [19]

    Feature distillation interaction weighting network for lightweight image super-resolution

    Guangwei Gao, Wenjie Li, Juncheng Li, Fei Wu, Huimin Lu, and Yi Yu. Feature distillation interaction weighting network for lightweight image super-resolution. InProceedings of the AAAI conference on artificial intelligence, volume 36, pages 661–669, 2022. 1, 2

    2022

  20. [20]

    Lightweight bimodal network for single-image super-resolution via symmetric cnn and recur- sive transformer.arXiv preprint arXiv:2204.13286, 2022

    Guangwei Gao, Zhengxue Wang, Juncheng Li, Wenjie Li, Yi Yu, and Tieyong Zeng. Lightweight bimodal network for single-image super-resolution via symmetric cnn and recur- sive transformer.arXiv preprint arXiv:2204.13286, 2022. 1, 2

    work page arXiv 2022

  21. [21]

    Dual-Branch Remote Sensing Infrared Image Super-Resolution

    Xining Ge, Gengjia Chang, Weijun Yuan, Zhan Li, Zhanglu Chen, Boyang Yao, Yihang Chen, Yifan Deng, and Shuhong Liu. Dual-branch remote sensing infrared image super- resolution.arXiv preprint arXiv:2604.10112, 2026. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  22. [22]

    CLIP-Guided Data Augmentation for Night-Time Image Dehazing

    Xining Ge, Weijun Yuan, Gengjia Chang, Xuyang Li, and Shuhong Liu. Clip-guided data augmentation for night-time image dehazing.arXiv preprint arXiv:2604.05500, 2026. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Lightweight image super-resolution based on deep learning: State-of-the- art and future directions.Information Fusion, 94:284–310,

    Garas Gendy, Guanghui He, and Nabil Sabor. Lightweight image super-resolution based on deep learning: State-of-the- art and future directions.Information Fusion, 94:284–310,

  24. [24]

    Mambairv2: Atten- tive state space restoration

    Hang Guo, Yong Guo, Yaohua Zha, Yulun Zhang, Wenbo Li, Tao Dai, Shu-Tao Xia, and Yawei Li. Mambairv2: Atten- tive state space restoration. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 28124– 28133, 2025. 1, 2

    2025

  25. [25]

    Mambair: A simple baseline for im- age restoration with state-space model

    Hang Guo, Jinmin Li, Tao Dai, Zhihao Ouyang, Xudong Ren, and Shu-Tao Xia. Mambair: A simple baseline for im- age restoration with state-space model. InEuropean confer- ence on computer vision, pages 222–241. Springer, 2024. 1, 2

    2024

  26. [26]

    Reliability-aware staged low-light gaussian splatting.ResearchGate preprint, 2026

    Haojie Guo and Ke Xian. Reliability-aware staged low-light gaussian splatting.ResearchGate preprint, 2026. 1

    2026

  27. [27]

    Srconvnet: A transformer-style con- vnet for lightweight image super-resolution.International Journal of Computer Vision, 133(1):173–189, 2025

    Feng Li, Runmin Cong, Jingjing Wu, Huihui Bai, Meng Wang, and Yao Zhao. Srconvnet: A transformer-style con- vnet for lightweight image super-resolution.International Journal of Computer Vision, 133(1):173–189, 2025. 1, 2

    2025

  28. [28]

    Srdiff: Single image super-resolution with diffusion probabilistic models

    Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing, 479:47–59, 2022. 1, 2

    2022

  29. [29]

    Densesplat: Densifying gaussian splatting slam with neural radiance prior.IEEE Transactions on Visualization & Computer Graphics, (01):1–14, 2025

    Mingrui Li, Shuhong Liu, Tianchen Deng, and Hongyu Wang. Densesplat: Densifying gaussian splatting slam with neural radiance prior.IEEE Transactions on Visualization & Computer Graphics, (01):1–14, 2025. 1

    2025

  30. [30]

    Sgs-slam: Semantic gaussian splatting for neural dense slam

    Mingrui Li, Shuhong Liu, Heng Zhou, Guohao Zhu, Na Cheng, Tianchen Deng, and Hongyu Wang. Sgs-slam: Semantic gaussian splatting for neural dense slam. InEuro- pean Conference on Computer Vision, pages 163–179, 2025. 1

    2025

  31. [31]

    Cross-receptive focused inference network for lightweight image super- resolution.IEEE Transactions on Multimedia, 26:864–877,

    Wenjie Li, Juncheng Li, Guangwei Gao, Weihong Deng, Jiantao Zhou, Jian Yang, and Guo-Jun Qi. Cross-receptive focused inference network for lightweight image super- resolution.IEEE Transactions on Multimedia, 26:864–877,

  32. [32]

    Dl- gsanet: lightweight dynamic local and global self-attention networks for image super-resolution

    Xiang Li, Jiangxin Dong, Jinhui Tang, and Jinshan Pan. Dl- gsanet: lightweight dynamic local and global self-attention networks for image super-resolution. InProceedings of the IEEE/CVF international conference on computer vision, pages 12792–12801, 2023. 2

    2023

  33. [33]

    Blueprint separable residual network for efficient image super-resolution

    Zheyuan Li, Yingqi Liu, Xiangyu Chen, Haoming Cai, Jinjin Gu, Yu Qiao, and Chao Dong. Blueprint separable residual network for efficient image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 833–843, 2022. 1, 2

    2022

  34. [34]

    Details or artifacts: A locally discriminative learning approach to realistic im- age super-resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Details or artifacts: A locally discriminative learning approach to realistic im- age super-resolution. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 5657–5666, 2022. 1, 2

    2022

  35. [35]

    Efficient and degradation-adaptive network for real-world image super- resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Efficient and degradation-adaptive network for real-world image super- resolution. InEuropean Conference on Computer Vision, pages 574–591. Springer, 2022. 1, 2

    2022

  36. [36]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1833–1844,

  37. [37]

    Blind image super-resolution: A survey and beyond.IEEE transactions on pattern analysis and machine intelligence, 45(5):5461–5480, 2022

    Anran Liu, Yihao Liu, Jinjin Gu, Yu Qiao, and Chao Dong. Blind image super-resolution: A survey and beyond.IEEE transactions on pattern analysis and machine intelligence, 45(5):5461–5480, 2022. 1, 2

    2022

  38. [38]

    Degradation-aware self-attention based transformer for blind image super-resolution.IEEE Transactions on Mul- timedia, 26:7516–7528, 2024

    Qingguo Liu, Pan Gao, Kang Han, Ningzhong Liu, and Wei Xiang. Degradation-aware self-attention based transformer for blind image super-resolution.IEEE Transactions on Mul- timedia, 26:7516–7528, 2024. 2

    2024

  39. [39]

    Cd- former: When degradation prediction embraces diffusion model for blind image super-resolution

    Qingguo Liu, Chenyi Zhuang, Pan Gao, and Jie Qin. Cd- former: When degradation prediction embraces diffusion model for blind image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7455–7464, 2024. 1, 2

    2024

  40. [40]

    NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Results

    Shuhong Liu, Chenyu Bao, Ziteng Cui, Xuangeng Chu, Bin Ren, Lin Gu, Xiang Chen, Mingrui Li, Long Ma, Marcos V . Conde, Radu Timofte, et al. NTIRE 2026 3D restoration and reconstruction in adverse conditions: RealX3D challenge re- sults.arXiv preprint arXiv:2604.04135, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  41. [41]

    Realx3d: A physically-degraded 3d benchmark for multi-view visual restoration and recon- struction.arXiv preprint arXiv:2512.23437, 2025

    Shuhong Liu, Chenyu Bao, Ziteng Cui, Yun Liu, Xuangeng Chu, Lin Gu, Marcos V Conde, Ryo Umagami, Tomohiro Hashimoto, Zijian Hu, et al. Realx3d: A physically-degraded 3d benchmark for multi-view visual restoration and recon- struction.arXiv preprint arXiv:2512.23437, 2026. 1

    work page arXiv 2026

  42. [42]

    Deraings: Gaussian splatting for enhanced scene reconstruction in rainy environments.Proceedings of the AAAI Conference on Artificial Intelligence, 39(5):5558– 5566, 2025

    Shuhong Liu, Xiang Chen, Hongming Chen, Quanfeng Xu, and Mingrui Li. Deraings: Gaussian splatting for enhanced scene reconstruction in rainy environments.Proceedings of the AAAI Conference on Artificial Intelligence, 39(5):5558– 5566, 2025. 1

    2025

  43. [43]

    Mg-slam: Structure gaussian splatting slam with manhattan world hy- pothesis.IEEE Transactions on Automation Science and En- gineering, 22:17034–17049, 2025

    Shuhong Liu, Tianchen Deng, Heng Zhou, Liuzhuozheng Li, Hongyu Wang, Danwei Wang, and Mingrui Li. Mg-slam: Structure gaussian splatting slam with manhattan world hy- pothesis.IEEE Transactions on Automation Science and En- gineering, 22:17034–17049, 2025. 1

    2025

  44. [44]

    De- noising the deep sky: Physics-based ccd noise formation for astronomical imaging.arXiv preprint arXiv:2601.23276,

    Shuhong Liu, Xining Ge, Ziying Gu, Lin Gu, Ziteng Cui, Xuangeng Chu, Jun Liu, Dong Li, and Tatsuya Harada. De- noising the deep sky: Physics-based ccd noise formation for astronomical imaging.arXiv preprint arXiv:2601.23276,

    work page arXiv

  45. [45]

    I2-nerf: Learning neural radiance fields un- der physically-grounded media interactions

    Shuhong Liu, Lin Gu, Ziteng Cui, Xuangeng Chu, and Tat- suya Harada. I2-nerf: Learning neural radiance fields un- der physically-grounded media interactions. InAdvances in Neural Information Processing Systems, 2025. 1

    2025

  46. [46]

    Vmamba: Visual state space model.Advances in neural information processing systems, 37:103031–103063, 2024

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. Vmamba: Visual state space model.Advances in neural information processing systems, 37:103031–103063, 2024. 2

    2024

  47. [47]

    ELoG-GS: Dual-Branch Gaussian Splatting with Luminance-Guided Enhancement for Extreme Low-light 3D Reconstruction

    Yuhao Liu, Dingju Wang, and Ziyang Zheng. Elog-gs: Dual- branch gaussian splatting with luminance-guided enhance- ment for extreme low-light 3d reconstruction.arXiv preprint arXiv:2604.12592, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  48. [48]

    Transformer for single im- age super-resolution

    Zhisheng Lu, Juncheng Li, Hong Liu, Chaoyan Huang, Lin- lin Zhang, and Tieyong Zeng. Transformer for single im- age super-resolution. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 457–466, 2022. 2

    2022

  49. [49]

    Deep constrained least squares for blind 7 image super-resolution

    Ziwei Luo, Haibin Huang, Lei Yu, Youwei Li, Haoqiang Fan, and Shuaicheng Liu. Deep constrained least squares for blind 7 image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17642–17652, 2022. 1, 2

    2022

  50. [50]

    Image super- resolution with non-local sparse attention

    Yiqun Mei, Yuchen Fan, and Yuqian Zhou. Image super- resolution with non-local sparse attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3517–3526, 2021. 2

    2021

  51. [51]

    A dy- namic residual self-attention network for lightweight single image super-resolution.IEEE Transactions on Multimedia, 25:907–918, 2021

    Karam Park, Jae Woong Soh, and Nam Ik Cho. A dy- namic residual self-attention network for lightweight single image super-resolution.IEEE Transactions on Multimedia, 25:907–918, 2021. 1, 2

    2021

  52. [52]

    The Eleventh NTIRE 2026 Efficient Super-Resolution Challenge Report

    Bin Ren, Hang Guo, Yan Shu, Jiaqi Ma, Ziteng Cui, Shuhong Liu, Guofeng Mei, Lei Sun, Zongwei Wu, Fahad Shahbaz Khan, Salman Khan, Radu Timofte, Yawei Li, et al. The eleventh NTIRE 2026 efficient super-resolution challenge re- port.arXiv preprint arXiv:2604.03198, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  53. [53]

    Resdiff: Combining cnn and diffusion model for image super-resolution

    Shuyao Shang, Zhengyang Shan, Guangxing Liu, LunQian Wang, XingHua Wang, Zekai Zhang, and Jinglin Zhang. Resdiff: Combining cnn and diffusion model for image super-resolution. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 8975–8983, 2024. 1, 2

    2024

  54. [54]

    Ope-sr: Orthogonal position encod- ing for designing a parameter-free upsampling module in arbitrary-scale image super-resolution

    Gaochao Song, Qian Sun, Luo Zhang, Ran Su, Jianfeng Shi, and Ying He. Ope-sr: Orthogonal position encod- ing for designing a parameter-free upsampling module in arbitrary-scale image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10009–10020, 2023. 1, 3

    2023

  55. [55]

    Spatially-adaptive feature modulation for efficient image super-resolution

    Long Sun, Jiangxin Dong, Jinhui Tang, and Jinshan Pan. Spatially-adaptive feature modulation for efficient image super-resolution. InProceedings of the IEEE/CVF interna- tional conference on computer vision, pages 13190–13199,

  56. [56]

    Ensir: An ensemble al- gorithm for image restoration via gaussian mixture mod- els.Advances in Neural Information Processing Systems, 37:133487–133517, 2024

    Shangquan Sun, Wenqi Ren, Zikun Liu, Hyunhee Park, Rui Wang, and Xiaochun Cao. Ensir: An ensemble al- gorithm for image restoration via gaussian mixture mod- els.Advances in Neural Information Processing Systems, 37:133487–133517, 2024. 1, 3

    2024

  57. [57]

    Omni aggregation networks for lightweight im- age super-resolution

    Hang Wang, Xuanhong Chen, Bingbing Ni, Yutian Liu, and Jinfan Liu. Omni aggregation networks for lightweight im- age super-resolution. InProceedings of the IEEE/cvf con- ference on computer vision and pattern recognition, pages 22378–22387, 2023. 1, 2

    2023

  58. [58]

    Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024. 1, 2

    2024

  59. [59]

    Exploring spar- sity in image super-resolution for efficient inference

    Longguang Wang, Xiaoyu Dong, Yingqian Wang, Xinyi Ying, Zaiping Lin, Wei An, and Yulan Guo. Exploring spar- sity in image super-resolution for efficient inference. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4917–4926, 2021. 1, 2

    2021

  60. [60]

    Deep arbitrary-scale image super-resolution via scale-equivariance pursuit

    Xiaohang Wang, Xuanhong Chen, Bingbing Ni, Hang Wang, Zhengyan Tong, and Yutian Liu. Deep arbitrary-scale image super-resolution via scale-equivariance pursuit. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1786–1795, 2023. 1, 3

    2023

  61. [61]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1905–1914,

    1905

  62. [62]

    Sinsr: diffusion-based image super- resolution in a single step

    Yufei Wang, Wenhan Yang, Xinyuan Chen, Yaohui Wang, Lanqing Guo, Lap-Pui Chau, Ziwei Liu, Yu Qiao, Alex C Kot, and Bihan Wen. Sinsr: diffusion-based image super- resolution in a single step. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 25796–25805, 2024. 1, 2

    2024

  63. [63]

    Uformer: A general u-shaped transformer for image restoration

    Zhendong Wang, Xiaodong Cun, Jianmin Bao, Wengang Zhou, Jianzhuang Liu, and Houqiang Li. Uformer: A general u-shaped transformer for image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17683–17693, 2022. 1, 2

    2022

  64. [64]

    Unsupervised real-world image super resolution via domain-distance aware training

    Yunxuan Wei, Shuhang Gu, Yawei Li, Radu Timofte, Long- cun Jin, and Hengjie Song. Unsupervised real-world image super resolution via domain-distance aware training. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13385–13394, 2021. 1, 2

    2021

  65. [65]

    Seesr: Towards semantics- aware real-world image super-resolution

    Rongyuan Wu, Tao Yang, Lingchen Sun, Zhengqiang Zhang, Shuai Li, and Lei Zhang. Seesr: Towards semantics- aware real-world image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 25456–25467, 2024. 2

    2024

  66. [66]

    A dynamic kernel prior model for unsupervised blind image super-resolution

    Zhixiong Yang, Jingyuan Xia, Shengxi Li, Xinghua Huang, Shuanghui Zhang, Zhen Liu, Yaowen Fu, and Yongxiang Liu. A dynamic kernel prior model for unsupervised blind image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 26046–26056, 2024. 1, 2

    2024

  67. [67]

    Local implicit normalizing flow for arbitrary-scale image super-resolution

    Jie-En Yao, Li-Yuan Tsao, Yi-Chen Lo, Roy Tseng, Chia- Che Chang, and Chun-Yi Lee. Local implicit normalizing flow for arbitrary-scale image super-resolution. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1776–1785, 2023. 1, 3

    2023

  68. [68]

    Blind image super-resolution with elaborate degradation modeling on noise and kernel

    Zongsheng Yue, Qian Zhao, Jianwen Xie, Lei Zhang, Deyu Meng, and Kwan-Yee K Wong. Blind image super-resolution with elaborate degradation modeling on noise and kernel. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 2128–2138, 2022. 1, 2

    2022

  69. [69]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5728–5739,

  70. [70]

    Designing a practical degradation model for deep blind im- age super-resolution

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timofte. Designing a practical degradation model for deep blind im- age super-resolution. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 4791–4800,

  71. [71]

    Hit-sr: Hierar- chical transformer for efficient image super-resolution

    Xiang Zhang, Yulun Zhang, and Fisher Yu. Hit-sr: Hierar- chical transformer for efficient image super-resolution. In European conference on computer vision, pages 483–500. Springer, 2024. 1, 2 8

    2024

  72. [72]

    Efficient long-range attention network for image super- resolution

    Xindong Zhang, Hui Zeng, Shi Guo, and Lei Zhang. Efficient long-range attention network for image super- resolution. InEuropean conference on computer vision, pages 649–667. Springer, 2022. 1, 2

    2022

  73. [73]

    3D Smoke Scene Reconstruction Guided by Vision Priors from Multimodal Large Language Models

    Xinye Zheng, Fei Wang, Yiqi Nie, Kun Li, Junjie Chen, Ji- aqi Zhao, Yanyan Wei, and Zhiliang Wu. 3d smoke scene re- construction guided by vision priors from multimodal large language models.arXiv preprint arXiv:2604.05687, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  74. [74]

    Mod- slam: Monocular dense mapping for unbounded 3d scene reconstruction.IEEE Robotics and Automation Letters, 10(1):484–491, 2024

    Heng Zhou, Zhetao Guo, Yuxiang Ren, Shuhong Liu, Lechen Zhang, Kaidi Zhang, and Mingrui Li. Mod- slam: Monocular dense mapping for unbounded 3d scene reconstruction.IEEE Robotics and Automation Letters, 10(1):484–491, 2024. 1

    2024

  75. [75]

    Learn- ing correction filter via degradation-adaptive regression for blind single image super-resolution

    Hongyang Zhou, Xiaobin Zhu, Jianqing Zhu, Zheng Han, Shi-Xue Zhang, Jingyan Qin, and Xu-Cheng Yin. Learn- ing correction filter via degradation-adaptive regression for blind single image super-resolution. InProceedings of the IEEE/CVF international conference on computer vision, pages 12365–12375, 2023. 1, 2

    2023

  76. [76]

    Srformer: Permuted self- attention for single image super-resolution

    Yupeng Zhou, Zhen Li, Chun-Le Guo, Song Bai, Ming- Ming Cheng, and Qibin Hou. Srformer: Permuted self- attention for single image super-resolution. InProceedings of the IEEE/CVF international conference on computer vi- sion, pages 12780–12791, 2023. 1, 2

    2023

  77. [77]

    Naka-GS: A Bionics-inspired Dual-Branch Naka Correction and Progressive Point Pruning for Low-Light 3DGS

    Runyu Zhu, SiXun Dong, Zhiqiang Zhang, Qingxia Ye, and Zhihua Xu. Naka-gs: A bionics-inspired dual-branch naka correction and progressive point pruning for low-light 3dgs. arXiv preprint arXiv:2604.11142, 2026. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  78. [78]

    Lightweight image super-resolution with expectation-maximization attention mechanism.IEEE Transactions on Circuits and Systems for Video Technology, 32(3):1273–1284, 2021

    Xiangyuan Zhu, Kehua Guo, Sheng Ren, Bin Hu, Min Hu, and Hui Fang. Lightweight image super-resolution with expectation-maximization attention mechanism.IEEE Transactions on Circuits and Systems for Video Technology, 32(3):1273–1284, 2021. 1, 2 9

    2021