Pith. sign in

REVIEW 2 major objections 1 minor 38 references

Profile-specific rewards on a shared candidate pool let one super-resolution base model produce either faithful or aesthetic outputs via separate LoRA adapters.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

FoA-SR trains separate Faithful and Aesthetic LoRA adapters by ranking a shared stochastic candidate pool with profile-specific rewards after initial supervised Flux2SR training.

T0 review reviewed 2026-06-27 challenge →

load-bearing objection FoA-SR gives a practical pipeline for separate faithful and aesthetic LoRAs from one base adapter, but the reward models lack any shown validation against the claimed definitions. the 2 major comments →

arxiv 2606.10275 v1 pith:25XO5XAU submitted 2026-06-09 cs.CV

FoA-SR: Faithful or Aesthetic? Profile-Aware Preference Optimization for Real-World Image Super-Resolution

Amjad Mahdi Alqarni , Peizhong Ju This is my paper
classification cs.CV
keywords image super-resolutionpreference optimizationLoRA adaptersfaithful restorationaesthetic restorationprofile-aware optimizationreal-world SRreward models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Real-world image super-resolution often faces a choice between staying true to a reference image and producing visually pleasing results. The paper develops FoA-SR to train separate adapters on the same base model for these two profiles. It begins with a shared supervised Flux2SR adapter trained on paired data. For each input it then draws a stochastic pool of candidates and ranks them separately with Faithful and Aesthetic reward models to form winner-loser pairs. These pairs fine-tune distinct LoRA adapters while the base stays frozen. Experiments on RealSR and DIV2K confirm that the resulting Faithful adapter lifts reference-based metrics and the Aesthetic adapter lifts no-reference perceptual metrics.

Core claim

FoA-SR demonstrates that a shared stochastic candidate pool ranked by profile-specific Faithful and Aesthetic reward models supplies winner-loser pairs that fine-tune separate LoRA adapters for distinct restoration objectives from the same frozen base model.

What carries the argument

Profile-specific Faithful and Aesthetic reward models that rank candidates from a shared stochastic pool to mine winner-loser pairs for preference optimization of separate LoRA adapters.

Load-bearing premise

The profile-specific Faithful and Aesthetic reward models correctly rank candidates according to the intended definitions of faithfulness versus aesthetics.

What would settle it

If the Faithful adapter fails to improve reference-consistent metrics more than the Aesthetic adapter (or the reverse for no-reference perceptual metrics) on the same test set after identical training steps.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • The Faithful adapter improves reference-consistent metrics while the Aesthetic adapter boosts metrics that measure perceptual quality without reference.
  • Faithful and Aesthetic rewards frequently select different winners from the same candidate pool.
  • Collapsing both profiles into one hybrid reward produces an implicit compromise rather than explicit profile control.
  • The same base model can be steered toward distinct restoration objectives by swapping the LoRA adapter.

Where Pith is reading between the lines

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

  • The method could be tested on other conditional generation tasks where multiple valid outputs exist, such as image inpainting.
  • End-user tools could expose a profile switch that loads the corresponding LoRA without retraining the base model.
  • The value of diverse candidate pools may generalize to other preference-optimization pipelines that need controllable trade-offs.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The paper introduces FoA-SR, a profile-aware preference optimization method for real-world image super-resolution. It begins with a supervised Flux2SR adapter trained via flow matching and reconstruction losses, then generates a shared stochastic candidate pool per input and uses separate Faithful (reference-consistent) and Aesthetic (perceptual) reward models to mine winner-loser pairs for fine-tuning distinct LoRA adapters on the frozen base. Experiments on RealSR and DIV2K report that the Faithful adapter improves reference-based metrics while the Aesthetic adapter improves no-reference perceptual metrics; candidate-pool analysis shows the rewards often select different winners, and a Hybrid-LoRA ablation collapses to a compromise.

Significance. If the central claim holds, the work provides a practical mechanism for controllable SR that decouples faithfulness from aesthetics via separate LoRAs rather than a single objective, which could be useful for applications with varying user priorities. The shared stochastic pool and explicit comparison to Hybrid-LoRA are strengths that allow direct measurement of profile divergence.

major comments (2)
  1. [Methods (reward model and pair mining)] The abstract and method description provide no details on the training data, architecture, loss functions, or validation procedure for the Faithful and Aesthetic reward models. This is load-bearing because the winner-loser pairs (and thus the LoRA updates) are derived directly from these models; without evidence that the Faithful reward ranks by reference consistency/structure/hallucination suppression and the Aesthetic reward ranks by perceptual naturalness, the reported metric separations on RealSR/DIV2K cannot be attributed to explicit profile control rather than reward-model artifacts.
  2. [Experiments and candidate-pool analysis] No statistical tests, confidence intervals, or ablation controls on the metric improvements are reported, and the candidate-pool analysis is described only qualitatively. This weakens support for the claim that the adapters achieve distinct objectives, as it is unclear whether the observed differences exceed what would arise from random selection within the pool or from the base Flux2SR adapter alone.
minor comments (1)
  1. [Method] The notation for the stochastic candidate pool generation and the exact form of the preference optimization objective (e.g., DPO-style loss) should be formalized with equations for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important gaps in methodological transparency and statistical rigor that we will address in revision. We respond to each major comment below.

read point-by-point responses

  1. Referee: [Methods (reward model and pair mining)] The abstract and method description provide no details on the training data, architecture, loss functions, or validation procedure for the Faithful and Aesthetic reward models. This is load-bearing because the winner-loser pairs (and thus the LoRA updates) are derived directly from these models; without evidence that the Faithful reward ranks by reference consistency/structure/hallucination suppression and the Aesthetic reward ranks by perceptual naturalness, the reported metric separations on RealSR/DIV2K cannot be attributed to explicit profile control rather than reward-model artifacts.

    Authors: We agree that the current manuscript lacks sufficient detail on the reward models. In the revised version we will add a new subsection under Methods that specifies: (i) the datasets used to train each reward model, (ii) their architectures and input representations, (iii) the exact loss functions and training hyperparameters, and (iv) validation procedures (including correlation with human judgments or reference-based proxies). This addition will directly demonstrate that the Faithful reward emphasizes reference consistency and hallucination suppression while the Aesthetic reward emphasizes perceptual naturalness, thereby grounding the observed metric separations in explicit profile control rather than artifacts. revision: yes

  2. Referee: [Experiments and candidate-pool analysis] No statistical tests, confidence intervals, or ablation controls on the metric improvements are reported, and the candidate-pool analysis is described only qualitatively. This weakens support for the claim that the adapters achieve distinct objectives, as it is unclear whether the observed differences exceed what would arise from random selection within the pool or from the base Flux2SR adapter alone.

    Authors: We acknowledge the absence of statistical rigor. In the revision we will (i) report 95% confidence intervals for all quantitative metrics on RealSR and DIV2K, (ii) add paired statistical tests (e.g., Wilcoxon signed-rank) with p-values comparing each adapter against the base Flux2SR and against each other, and (iii) quantify the candidate-pool analysis with concrete statistics such as the fraction of inputs where Faithful and Aesthetic rewards select different winners, the rank correlation between the two reward scores, and an ablation measuring performance when winners are chosen uniformly at random from the pool. These additions will show that the profile-specific gains exceed random variation and the base model. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external rewards and empirical evaluation

full rationale

The paper's chain begins with a supervised Flux2SR adapter trained on standard losses, generates a stochastic candidate pool, applies separate external profile-specific reward models to mine preference pairs, and fine-tunes distinct LoRAs. The claimed divergence (Faithful adapter on reference metrics, Aesthetic on no-reference perceptual metrics) is presented as an empirical outcome of this process rather than a definitional reduction. No equations equate the final metric improvements to quantities defined by the same fitted parameters, no self-citations bear the central load, and the rewards are treated as independent inputs whose alignment with definitions is an assumption rather than a constructed identity. The method is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so no concrete free parameters, axioms, or invented entities can be extracted. The approach assumes standard concepts such as LoRA adapters and flow matching but supplies no further ledger items.

reviewed 2026-06-27 · how reviews work

0 comments
Cite this review

Pith. "Pith review of FoA-SR: Faithful or Aesthetic? Profile-Aware Preference Optimization for Real-World Image Super-Resolution." pith.science (2026). https://pith.science/paper/25XO5XAU

@misc{pith2026260610275,
  author       = {Pith},
  title        = {Pith review of: FoA-SR: Faithful or Aesthetic? Profile-Aware Preference Optimization for Real-World Image Super-Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/25XO5XAU}},
  note         = {Machine review of arXiv:2606.10275}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Real-world image super-resolution (SR) is often designed with a single restoration objective, despite the current capacity of generative models to produce multiple high-quality reconstructions for the same input. In this paper, we argue that the best restoration strategy is subject to the specific restoration profile: a Faithful restoration prioritizes reference consistency, structure preservation, and hallucination suppression, whereas an Aesthetic restoration prioritizes visually pleasing and natural-looking details. We propose FoA-SR, a novel preference optimization approach to real-world SR based on profiles. To achieve this goal, FoA-SR starts with our supervised FLUX.2-based SR adapter (Flux2SR) trained with LR latent conditioning, flow matching, and image-space reconstruction losses for paired LR-to-HR image super-resolution. Following the development of the shared supervised super-resolution adapter, FoA-SR generates a shared stochastic candidate pool for each input image and ranks the same candidates using profile-specific Faithful and Aesthetic rewards to mine winner-loser pairs. These pairs are used to fine-tune separate LoRA adapters while keeping the base model frozen. Experiments on RealSR and DIV2K show that FoA-SR can steer the same SR adapter towards distinct restoration objectives: a Faithful adapter improves reference-consistent metrics while an Aesthetic adapter boosts metrics that measure perceptual quality without reference. Our candidate-pool analysis shows that Faithful and Aesthetic rewards frequently select different winners, and a Hybrid-LoRA ablation shows that collapsing both profiles into one reward yields an implicit compromise rather than explicit profile control.

Figures

Figures reproduced from arXiv: 2606.10275 by Amjad Mahdi Alqarni, Peizhong Ju.

Figure 1
Figure 1. Figure 1: Qualitative illustration of profile-dependent restoration. Given the same LR input, FoA [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed FoA-SR framework. Stage 1 adapts FLUX.2-dev into a shared [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on a text-containing RealSR example. FoA-SR-Faithful emphasizes [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example stochastic candidate pool generated by the shared Flux2SR baseline for the same [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Additional qualitative examples on RealSR comparing Flux2SR, FoA-SR-Faithful, FoA [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Additional qualitative examples on DIV2K-Val comparing Flux2SR, FoA-SR-Faithful, [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    Ntire 2017 challenge on single image super-resolution: Dataset and study

    Eirikur Agustsson and Radu Timofte. Ntire 2017 challenge on single image super-resolution: Dataset and study. InCVPR Workshops, 2017

    2017

  2. [2]

    FLUX.2: Next generation image generation

    Black Forest Labs. FLUX.2: Next generation image generation. https://bfl.ai/models/ flux-2, 2025. Accessed: 2026-05-01

    2025

  3. [3]

    FLUX.2-dev model card

    Black Forest Labs. FLUX.2-dev model card. https://huggingface.co/ black-forest-labs/FLUX.2-dev, 2026. Accessed: 2026-05-01

    2026

  4. [4]

    The perception-distortion tradeoff

    Yochai Blau and Tomer Michaeli. The perception-distortion tradeoff. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6228–6237, 2018

    2018

  5. [5]

    Toward real-world single image super-resolution: A new benchmark and a new model

    Jianrui Cai, Hui Zeng, Hongwei Yong, Zisheng Cao, and Lei Zhang. Toward real-world single image super-resolution: A new benchmark and a new model. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3086–3095, 2019

    2019

  6. [6]

    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 conference on computer vision and pattern recognition, pages 22367–22377, 2023

    2023

  7. [7]

    Simoncelli

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P. Simoncelli. Image quality assessment: Unifying structure and texture similarity. InIEEE Transactions on Pattern Analysis and Machine Intelligence, 2020

    2020

  8. [8]

    Learning a deep convolutional network for image super-resolution

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Learning a deep convolutional network for image super-resolution. InEuropean Conference on Computer Vision, 2014. URL https://api.semanticscholar.org/CorpusID:18874645

    2014

  9. [9]

    Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution, 2025

    Linwei Dong, Qingnan Fan, Yihong Guo, Zhonghao Wang, Qi Zhang, Jinwei Chen, Yawei Luo, and Changqing Zou. Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution, 2025. URLhttps://arxiv.org/abs/2411.18263

    work page arXiv 2025

  10. [10]

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

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

    2024

  11. [11]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022

    2022

  12. [12]

    Diffusers welcomes FLUX.2

    Hugging Face. Diffusers welcomes FLUX.2. https://huggingface.co/blog/flux-2,

  13. [13]

    Accessed: 2026-05-01

    2026

  14. [14]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InCVPR, 2019

    2019

  15. [15]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 5148–5157, 2021

    2021

  16. [16]

    Efficient and explicit modelling of image hierarchies for image restoration

    Yawei Li, Yuchen Fan, Xiaoyu Xiang, Denis Demandolx, Rakesh Ranjan, Radu Timofte, and Luc Van Gool. Efficient and explicit modelling of image hierarchies for image restoration. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18278–18289, 2023

    2023

  17. [17]

    Lsdir: A large scale dataset for image restoration.arXiv preprint, 2023

    Yawei Li et al. Lsdir: A large scale dataset for image restoration.arXiv preprint, 2023. 10

    2023

  18. [18]

    Swinir: Image restoration using swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, pages 1833–1844, October 2021

    2021

  19. [19]

    Diffbir: Toward blind image restoration with generative diffusion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Bo Dai, Fanghua Yu, Yu Qiao, Wanli Ouyang, and Chao Dong. Diffbir: Toward blind image restoration with generative diffusion prior. InEuropean Conference on Computer Vision, pages 430–448. Springer, 2024

    2024

  20. [20]

    Anish Mittal, Rajiv Soundararajan, and Alan C. Bovik. Making a completely blind image quality analyzer.IEEE Signal Processing Letters, 20(3):209–212, 2013

    2013

  21. [21]

    Manning, Stefano Ermon, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. InAdvances in Neural Information Processing Systems, 2023

    2023

  22. [22]

    Improv- ing the stability of diffusion models for content consistent super-resolution.arXiv preprint arXiv:2401.00877, 2024

    Lingchen Sun, Rongyuan Wu, Zhengqiang Zhang, Hongwei Yong, and Lei Zhang. Improv- ing the stability of diffusion models for content consistent super-resolution.arXiv preprint arXiv:2401.00877, 2024

    work page arXiv 2024

  23. [23]

    Diffusion model alignment using direct preference optimization

    Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik. Diffusion model alignment using direct preference optimization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8228–8238, 2024

    2024

  24. [24]

    Jianyi Wang, Kelvin C. K. Chan, and Chen Change Loy. Exploring clip for assessing the look and feel of images. InProceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 2555–2563, 2023

    2023

  25. [25]

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin C. K. Chan, and Chen Change Loy. Ex- ploiting diffusion prior for real-world image super-resolution.arXiv preprint arXiv:2305.07015, 2023

    work page arXiv 2023

  26. [26]

    Spectral and Trajectory Regularization for Diffusion Transformer Super-Resolution

    Jingkai Wang, Yixin Tang, Jue Gong, Jiatong Li, Shu Li, Libo Liu, Jianliang Lan, Yutong Liu, and Yulun Zhang. Spectral and trajectory regularization for diffusion transformer super- resolution.arXiv preprint arXiv:2603.06275, 2026

    work page internal anchor Pith review arXiv 2026

  27. [27]

    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 International Conference on Computer Vision Workshops, pages 1905–1914, 2021

    1905

  28. [28]

    Bovik, Hamid R

    Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: From error visibility to structural similarity. InIEEE Transactions on Image Processing, volume 13, pages 600–612, 2004

    2004

  29. [29]

    One-step effective diffusion network for real-world image super-resolution

    Rongyuan Wu, Lingchen Sun, Zhiyuan Ma, and Lei Zhang. One-step effective diffusion network for real-world image super-resolution. InAdvances in Neural Information Processing Systems, volume 37, pages 92529–92553, 2024

    2024

  30. [30]

    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

    2024

  31. [31]

    Dp2o-sr: Direct perceptual preference optimization for real-world image super-resolution.arXiv preprint arXiv:2510.18851, 2025

    Rongyuan Wu, Lingchen Sun, Zhengqiang Zhang, Shihao Wang, Tianhe Wu, Qiaosi Yi, Shuai Li, and Lei Zhang. Dp2o-sr: Direct perceptual preference optimization for real-world image super-resolution.arXiv preprint arXiv:2510.18851, 2025

    work page arXiv 2025

  32. [32]

    Maniqa: Multi-dimension attention network for no-reference image quality assessment

    Sidi Yang, Tianhe Wu, Shuwei Shi, Shanshan Lao, Yuan Gong, Mingdeng Cao, Jiahao Wang, and Yujiu Yang. Maniqa: Multi-dimension attention network for no-reference image quality assessment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 1191–1200, 2022

    2022

  33. [33]

    Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization.arXiv preprint arXiv:2308.14469, 2023

    Tao Yang, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization.arXiv preprint arXiv:2308.14469, 2023. 11

    work page arXiv 2023

  34. [34]

    Designing a practical degradation model for deep blind image super-resolution

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timofte. Designing a practical degradation model for deep blind image super-resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4791–4800, 2021

    2021

  35. [35]

    Transcending the limit of local window: Advanced super-resolution transformer with adaptive token dictionary

    Leheng Zhang, Yawei Li, Xingyu Zhou, Xiaorui Zhao, and Shuhang Gu. Transcending the limit of local window: Advanced super-resolution transformer with adaptive token dictionary. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2856–2865, 2024

    2024

  36. [36]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unrea- sonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 586–595, 2018

    2018

  37. [37]

    Efficient long-range attention network for image super-resolution, 2022

    Xindong Zhang, Hui Zeng, Shi Guo, and Lei Zhang. Efficient long-range attention network for image super-resolution, 2022. URLhttps://arxiv.org/abs/2203.06697

    work page arXiv 2022

  38. [38]

    Image super- resolution using very deep residual channel attention networks

    Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu. Image super- resolution using very deep residual channel attention networks. InECCV, 2018. 12 A Additional Implementation Details We provide additional details about the Flux2SR supervised training setup, the FoA-SR preference optimization procedure, compute resources, and existing as...

    2018

This paper was first reviewed by grok-4.3 on June 27, 2026.