Pith. sign in

REVIEW 3 major objections 4 minor 300 references

Routing by restoration difficulty—hard images to a detail-preserving 4×-VAE diffusion expert, easy ones to a fast 8×-VAE expert—yields better perceptual quality at lower compute than one-size-fits-all diffusion super-resolution.

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 ↗

T0 review · deepseek-v4-flash

2026-08-01 22:31 UTC pith:Y5BKEGMF

load-bearing objection The routing idea is real and the easy/hard analysis is nice, but as written the inference-time router needs the ground-truth image, so the central claim doesn't hold until a learned difficulty predictor is added. the 3 major comments →

arxiv 2607.15711 v1 pith:Y5BKEGMF submitted 2026-07-17 cs.CV cs.AI

Efficient Difficulty-Aware Dynamic Routing for Diffusion-Based Real-World Image Super-Resolution

Xue Wu , Kang Zhao , Kafeng Wang , Jianfei Chen , Jingwei Xin , Nannan Wang , Xinbo Gao This is my paper
classification cs.CV cs.AI
keywords real-world image super-resolutiondifficulty-aware routingdynamic routingdiffusion modelsVAE compression ratiohigh-frequency energyone-step diffusionlow-rank adaptation
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.

Real-world image super-resolution models currently process every input with the same network, even though some images are far harder to restore than others. DDR-SR proposes to route each input to one of two diffusion experts: a high-capacity model with a 4×-downsampling VAE that preserves fine details for hard images, and an efficient model with the standard 8×-downsampling VAE for easy images. Difficulty is measured as the loss of high-frequency Fourier energy between the degraded image and its ground-truth counterpart. The paper reports that this dynamic routing beats state-of-the-art diffusion super-resolution methods on perceptual metrics while cutting FLOPs, and that the routing threshold gives users a tunable quality/speed knob. A sympathetic reading takes the central claim to be that the VAE compression ratio, not just the denoising network, is the right place to modulate capacity in diffusion-based restoration.

Core claim

The paper's core discovery is that a pair of expert diffusion networks differing only in their VAE's spatial compression ratio—4× versus 8×—can be trained with an 80/20 mixed-sampling strategy to produce complementary strengths, and a simple threshold on the high-frequency energy attenuation score can then assign each input to the right expert. On DIV2K-val, RealSR, and DRealSR, the routed system ('DDR-SR-1') reports the best SSIM, LPIPS, and CLIPIQA among compared one-step and multi-step diffusion methods, and the 4× expert alone runs at 1.81T FLOPs versus 2.27T for the strongest one-step baseline. The author's claim is that the rigid 8× VAE compression in standard latent diffusion irrevers

What carries the argument

The load-bearing mechanism is the difficulty-aware router. It computes, via the 2D Fourier transform and a high-frequency mask, the score d = 1 − E_high(I_deg)/E_high(I_gt), which falls in [0,1] and measures how much high-frequency spectral energy the degradation has removed. That score is compared to a user-settable threshold τ (default: the training set's median difficulty) to send the input to either Expert-D4, which uses a 4×-downsampling VAE and is LoRA-tuned on 80% hard plus 20% easy samples, or Expert-D8, which uses the standard 8×-downsampling VAE and is LoRA-tuned on the complementary mix. Both experts share the same diffusion UNet and one-step denoising, so the routing decision is

Load-bearing premise

The difficulty score d is computed from both the degraded image and its ground-truth high-resolution counterpart; in real-world deployment the ground truth does not exist, and the paper does not describe a learned network that estimates d from the degraded image alone, so the automatic routing cannot be executed as specified without oracle access.

What would settle it

Compute d for all test images using Eq. (2) with ground truth, then train a neural network that takes only I_deg as input and tries to predict d. If the predicted routing (using that proxy's scores) yields quality or efficiency no better than always sending every image to Expert-D4 (or always to Expert-D8), the central claim that difficulty-aware routing helps would be refuted. A second, simpler check: inspect whether the authors' reported benchmark numbers were produced using the ground-truth-based oracle d, since if they were, the same numbers would not be reproducible at deployment time whe

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

If this is right

  • Diffusion super-resolution no longer needs a single fixed VAE compression ratio; capacity can be matched to input difficulty, improving perceptual fidelity on hard samples.
  • Since easy inputs dominate many real-world collections, average computational cost per image can drop below what a single high-capacity model would need.
  • The mixed-training recipe (80/20 in opposing directions) keeps each expert robust to routing mistakes without sacrificing specialization.
  • The user-adjustable threshold turns the quality/efficiency trade-off into a deployable knob, so practitioners can select a point on the curve at runtime.
  • The routing logic is model-agnostic in principle and could attach to any latent-diffusion restoration system whose VAE compression ratio can be varied.

Where Pith is reading between the lines

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

  • The difficulty estimator as defined requires the ground-truth HR image; to make DDR-SR deployable, the authors would need to train a proxy network that predicts d from the degraded image alone, and the reported gains would then depend on how well that proxy recovers the oracle ranking.
  • The global Fourier energy score may be a coarse difficulty signal: an image with a small, heavily degraded text region could receive a low difficulty score if most of its energy is elsewhere; a spatially localized or region-weighted difficulty predictor might improve routing accuracy.
  • The 4×/8× choice is a two-point discretization of a continuous capacity spectrum; interpolating the VAE compression ratio between 4 and 8, or routing among more than two experts, might smooth the quality cliff at the threshold and yield a finer trade-off.
  • Because the two experts share the UNet, the router could in principle be folded into a single model that selects the VAE at inference time, reducing duplication of the diffusion backbone.

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

3 major / 4 minor

Summary. The paper proposes DDR-SR, a difficulty-aware dynamic routing framework for diffusion-based real-world image super-resolution. It defines a difficulty score based on high-frequency spectral energy attenuation between the degraded image and its ground-truth counterpart, splits training data into easy and hard subsets, and trains two LoRA-adapted expert networks: Expert-D4 with a 4x-compression VAE and Expert-D8 with an 8x-compression VAE. At inference, inputs are routed to one of the two experts according to a threshold on the difficulty score. The method is evaluated on DIV2K-val, RealSR, and DRealSR against several multi-step and one-step diffusion baselines, and the paper claims superior quality and a tunable quality/efficiency trade-off.

Significance. If the method worked as described, the idea of adaptively choosing VAE compression according to per-image difficulty would be useful for the Real-ISR community: it is a plausible way to preserve fine details for hard inputs while saving computation on easy ones. The paper also provides real benchmark numbers and a reasonable experimental setup. However, as written, the central mechanism is not implementable because the routing decision requires the ground-truth image, the efficiency claim is based on an incomplete comparison, and no ablation isolates the effect of routing. None of these issues is merely cosmetic: each concerns the core contribution of the paper.

major comments (3)
  1. [Section 3.2 and Algorithm 2] The routing difficulty score defined in Eq. (2), d = 1 - E_high(I_deg)/E_high(I_gt), requires the ground-truth image I_gt. Algorithm 2 explicitly writes 'd <- E_diff(I_deg, I_gt)'. In real-world deployment I_gt does not exist, yet the paper nowhere specifies, trains, or evaluates a learned estimator that maps only I_deg to d. Without such a proxy, the automatic dynamic routing at inference cannot be executed, and the Table 1 results must have been obtained either with oracle access to ground truth (which would inflate reported gains) or with an undocumented approximation. Since dynamic routing is the paper's stated core contribution, this is a load-bearing flaw. The authors need to provide a fully specified deployable difficulty estimator and re-run the benchmark evaluation without using I_gt at inference.
  2. [Section 4.2, Table 2] The efficiency comparison reports only Expert-D4 (1.81T FLOPs, 1.63B parameters) and omits the full DDR-SR system. The actual inference pipeline also includes the difficulty estimator/router and, under the default median split, routes roughly half the inputs to Expert-D8 and half to Expert-D4. The paper never reports the expected or measured FLOPs/latency of the complete DDR-SR model, nor the router's additional cost. Consequently, the Conclusion's claim of 'a tunable trade-off between quality and speed' is not supported by the efficiency data presented. The authors should report full DDR-SR complexity, including router overhead, for at least a few routing ratios.
  3. [Section 4.3] The ablation study is limited to LoRA rank configurations for Expert-D4 and Expert-D8 (Tables 3 and 4). It does not compare DDR-SR against Expert-D4-only or Expert-D8-only, nor does it vary the routing threshold tau. Without these comparisons, the reported improvement cannot be attributed to dynamic routing; it could come from the higher-capacity Expert-D4 architecture or from the mixed training strategy. The paper needs ablations that isolate the routing decision, e.g., D4-only, D8-only, DDR-SR at tau=median, and DDR-SR at other routing ratios, with matched training budgets.
minor comments (4)
  1. [Section 4.2] Text says 'DiffBIR, SeeSR, PASD, and DiffBIR exhibit better performance' — DiffBIR is listed twice; the intended set is unclear.
  2. [Figure 4 / Section 4.2] In the qualitative discussion, 'DRR-SR' is a typo for 'DDR-SR'.
  3. [Section 4.2 / Table 2] The text says Expert-D4 has '1.65B parameters', but Table 2 lists 1.63B. Please reconcile.
  4. [Eq. (3)] The formula for z_SR is typeset ambiguously; it should be z_SR = (z - sqrt(1 - alpha_bar_t) * eps_8(z, c_t, t)) / sqrt(alpha_bar_t) or written as an explicit one-step denoising expression.

Circularity Check

1 steps flagged

Routing score is computed from ground truth in Eq. (2)/Algorithm 2, making benchmark routing oracle-based rather than predicted.

specific steps
  1. self definitional [Eq. (2), Section 3.2; Algorithm 2 line 2; Section 3.4]
    "The restoration difficulty of a degraded image is defined as: d = 1 − E_high(I_deg)/E_high(I_gt) ... [Algorithm 2] Compute difficulty score: d ← E_diff(I_deg, I_gt)"

    The routing signal d is defined as a normalized comparison against the ground-truth HR image. Algorithm 2 calls this estimator with I_gt at inference: d ← E_diff(I_deg, I_gt). No learned proxy mapping only I_deg to d is introduced or trained anywhere in the paper. Therefore the route choices and benchmark numbers for DDR-SR-1 are not predictions from the degraded input; they are constructed from the target image. The claimed 'automatic assignment' and 'difficulty-aware dynamic routing' thus reduce by construction to oracle access to the label, and the advertised quality/speed trade-off is not demonstrated for deployment.

full rationale

The central derivational flaw is that the difficulty score used to make the routing decision is defined in terms of the ground-truth image (Eq. 2) and is explicitly evaluated with I_gt in Algorithm 2 at inference time. The paper never describes or trains a network that predicts d from the degraded image alone, so the reported DDR-SR-1 results in Table 1 depend on oracle routing rather than an executable sample-adaptive mechanism. This makes the core 'dynamic routing' claim partly circular: the decision variable is not predicted from the input but read off from the target. The other components—two VAE-modulated experts, LoRA tuning, mixed training, and the loss combination—are independent and not circularly justified, and there is no load-bearing self-citation or imported uniqueness theorem. However, because the central contribution is evaluated by a route selector that uses ground truth, the overall circularity score is elevated.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

Central claim rests on two empirical assumptions (high-frequency energy loss equals difficulty; 4×-VAE latents usable by SD UNet), on an unflagged oracle assumption (I_gt available at inference), and on hand-picked hyperparameters (λs, mixing ratios, threshold, LoRA ranks, unspecified frequency mask) not supported by sensitivity analysis.

free parameters (5)
  • Loss weights λ1, λ2 = λ1=2, λ2=1
    Set in Eq. (5)/Sec. 3.3 without sensitivity analysis.
  • Mixing ratios α, β = α=0.8, β=0.2
    Chosen for mixed training sets in Algorithm 1 without ablation.
  • Routing threshold τ = median of training difficulties
    Default split between Expert-D4 and Expert-D8; no sensitivity analysis, user-tunable.
  • LoRA ranks = VAE-D4 4, UNet-D4 4; VAE-D8 16, UNet-D8 4
    Selected via ablation on RealSR (Tables 3-4); rank choices may be dataset-specific.
  • High-frequency mask threshold M_high = not specified (radial threshold)
    Eq. (1) defines M_high as a binary mask but no concrete frequency cutoff is given; the difficulty scores and training split depend on it.
axioms (4)
  • domain assumption Real-world degradations act predominantly as low-pass filters, so high-frequency energy attenuation (Eq. 2) measures restoration difficulty.
    Justifies the entire difficulty estimator; if degradation also corrupts low frequencies or adds structured noise, the metric misroutes.
  • domain assumption A 4×-compression VAE preserves fine details and is compatible with the pretrained SD 2.1 UNet.
    Latent resolution changes from 64×64 to 128×128; the paper does not explain how the UNet or attention is adapted.
  • ad hoc to paper Ground truth is available when computing the routing score.
    Algorithm 2 line 2 uses d←E_diff(I_deg, I_gt); no learned proxy is provided, so inference depends on an oracle.
  • domain assumption Stable diffusion priors and VAE-D8 from SD 2.1 retain enough prior for one-step SR after LoRA tuning.
    Borrowed from OSEDiff/TVTSR; standard in this literature but assumed.
invented entities (2)
  • Difficulty score d (high-frequency energy attenuation) no independent evidence
    purpose: Routes each input image to one of two experts
    d is defined using ground truth, so it has no falsifiable handle in real deployment; no predictor trained.
  • Expert-D4 / Expert-D8 router no independent evidence
    purpose: Dynamic selection of 4×-VAE or 8×-VAE diffusion network
    Internal to the system; no external validation that routing decisions are correct (no ablation vs fixed experts).

pith-pipeline@v1.3.0-alltime-deepseek · 14932 in / 17312 out tokens · 128351 ms · 2026-08-01T22:31:37.497797+00:00 · methodology

0 comments
read the original abstract

Diffusion-based methods have achieved impressive performance in real-world image super-resolution (Real-ISR) by leveraging large pre-trained stable diffusion (SD) models as powerful generative priors. However, these methods still face two key limitations. First, existing SD-based one-step and multi-step Real-ISR approaches adopt a unified processing paradigm for all input samples, ignoring the varying restoration difficulty across images. Second, the aggressive resolution reduction of the VAE in SD models (e.g., 8x downsampling) leads to irreversible loss of fine-scale details, which cannot be recovered by the subsequent diffusion process. To address these limitations, we propose a Difficulty-aware Dynamic Routing (DDR) strategy that overcomes the rigid, one-size-fits-all processing paradigm. Specifically, we first design a difficulty estimator to predict the restoration cost of each input image, enabling automatic assignment to a network of appropriate capacity. Then, we construct a set of Real-ISR networks with varying model capacities by modulating the spatial downsampling ratio of the VAE in the SD backbone, thereby preserving more high-frequency information for challenging cases while maintaining efficiency for simpler inputs. Extensive experiments have demonstrated the superior efficiency and effectiveness of the proposed model compared to recent state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2607.15711 by Jianfei Chen, Jingwei Xin, Kafeng Wang, Kang Zhao, Nannan Wang, Xinbo Gao, Xue Wu.

Figure 1
Figure 1. Figure 1: Architecture comparison of diffusion-based SR ap￾proaches. (a) Controller-based methods employ conditional mech￾anisms to guide the diffusion process; (b) Standard diffusion pro￾cesses latent representations directly; (c) Our proposed DDR em￾ploys difficulty-aware dynamic routing to adaptively select expert networks with varying VAE compression ratios. degradations. SR is also a challenging ill-posed probl… view at source ↗
Figure 2
Figure 2. Figure 2: The PSNR and SSIM performance of three representa￾tive SOTA methods (OSEDiff (Wu et al., 2024a), S3Diff(Zhang et al., 2024a), and TVTSR(Yi et al., 2025)) on both easy and hard subsets. Notably, the performance gaps among these methods are consistently smaller on the easy subset than on the hard subset across both metrics. on the hard subset, where fine structural details are severely degraded. This observa… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed Difficulty-aware Dynamic Routing (DDR) framework. During training, the dataset is split into “easy” and “hard” subsets using a high-frequency energy–based difficulty estimator. We train two expert models via LoRA: VAE-D4 + diffusion network on 80% hard + 20% easy samples for detail preservation, and VAE-D8 + diffusion network on 80% easy + 20% hard samples for efficiency. The two n… view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparisons of different Real-ISR methods. Please zoom in for a better view. DRealSR (Wei et al., 2020b). The synthetic dataset consists of 3,000 image pairs, which are created by downgrading the 512×512 HQ images cropped from DIV2K-Val to 128×128 LQ images using the Real-ESRGAN (Wang et al., 2021b) degradation pipeline. The real-world datasets comprised 128 × 128 and 512 × 512 HQ image pairs. 7 [P… view at source ↗
Figure 5
Figure 5. Figure 5: provides a visual comparison of various diffusion-based methods applied to real-world datasets, covering both multi￾step and one-step approaches. As shown, none of the competing methods could effectively reconstruct the intricate ruler texture in the first image, the letters and graphics in the second image, or the facial details in the third image. Specifically, these methods either produced blurred or in… 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

300 extracted references · 70 linked inside Pith

  1. [1]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Toward real-world single image super-resolution: A new benchmark and a new model , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  2. [2]

    European conference on computer vision , pages=

    Component divide-and-conquer for real-world image super-resolution , author=. European conference on computer vision , pages=

  3. [3]

    Image super-resolution based on dictionary learning and anchored neighborhood regression with mutual incoherence , author=. Proc

  4. [4]

    ECCV , year=

    Accelerating the super-resolution convolutional neural network , author=. ECCV , year=

  5. [5]

    TPAMI , year=

    Image super-resolution using deep convolutional networks , author=. TPAMI , year=

  6. [6]

    CVPR , year=

    Accurate image super-resolution using very deep convolutional networks , author=. CVPR , year=

  7. [7]

    ICCV , year=

    Dual Aggregation Transformer for Image Super-Resolution , author=. ICCV , year=

  8. [8]

    TPAMI , year=

    Deep learning for image super-resolution: A survey , author=. TPAMI , year=

  9. [9]

    ICCV , year=

    Dslr-quality photos on mobile devices with deep convolutional networks , author=. ICCV , year=

  10. [10]

    ICCV , year=

    Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data , author=. ICCV , year=

  11. [11]

    CVPRW , year=

    Real-world super-resolution via kernel estimation and noise injection , author=. CVPRW , year=

  12. [12]

    ECCV , year=

    Component divide-and-conquer for real-world image super-resolution , author=. ECCV , year=

  13. [13]

    CVPR , year=

    Robustifying token attention for vision transformers , author=. CVPR , year=

  14. [14]

    NeurIPS , year=

    Denoising diffusion probabilistic models , author=. NeurIPS , year=

  15. [15]

    ICLR , year=

    Score-based generative modeling through stochastic differential equations , author=. ICLR , year=

  16. [16]

    CVPR , year=

    Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn , title=. CVPR , year=

  17. [17]

    2023 , howpublished=

    Black Forest Labs , title=. 2023 , howpublished=

    2023

  18. [18]

    ICCV , year=

    Adding Conditional Control to Text-to-Image Diffusion Models , author=. ICCV , year=

  19. [19]

    arXiv preprint arXiv:2307.01952 , year=

    Sdxl: Improving latent diffusion models for high-resolution image synthesis , author=. arXiv preprint arXiv:2307.01952 , year=

    Pith/arXiv arXiv

  20. [20]

    ECCV , year=

    Adversarial diffusion distillation , author=. ECCV , year=

  21. [21]

    CVPR , year=

    Ufogen: You forward once large scale text-to-image generation via diffusion gans , author=. CVPR , year=

  22. [22]

    arXiv preprint arXiv:2403.12015 , year=

    Fast high-resolution image synthesis with latent adversarial diffusion distillation , author=. arXiv preprint arXiv:2403.12015 , year=

    Pith/arXiv arXiv

  23. [23]

    ICLR , year=

    Diffusion-gan: Training gans with diffusion , author=. ICLR , year=

  24. [24]

    CVPR , year=

    Swiftbrush: One-step text-to-image diffusion model with variational score distillation , author=. CVPR , year=

  25. [25]

    NeurIPS , year=

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation , author=. NeurIPS , year=

  26. [26]

    ICLR , year=

    Denoising diffusion implicit models , author=. ICLR , year=

  27. [27]

    NeurIPS , year=

    Elucidating the design space of diffusion-based generative models , author=. NeurIPS , year=

  28. [28]

    ICLR , year=

    Pseudo numerical methods for diffusion models on manifolds , author=. ICLR , year=

  29. [29]

    NeurIPS , year=

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps , author=. NeurIPS , year=

  30. [30]

    arXiv preprint arXiv:2211.01095 , year=

    Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models , author=. arXiv preprint arXiv:2211.01095 , year=

    Pith/arXiv arXiv

  31. [31]

    NeurIPS , year=

    Unipc: A unified predictor-corrector framework for fast sampling of diffusion models , author=. NeurIPS , year=

  32. [32]

    arXiv preprint arXiv:2303.04248 , year=

    Tract: Denoising diffusion models with transitive closure time-distillation , author=. arXiv preprint arXiv:2303.04248 , year=

    Pith/arXiv arXiv

  33. [33]

    CVPR , year=

    High-resolution image synthesis with latent diffusion models , author=. CVPR , year=

  34. [34]

    Forty-first International Conference on Machine Learning , year=

    Scaling rectified flow transformers for high-resolution image synthesis , author=. Forty-first International Conference on Machine Learning , year=

  35. [35]

    arXiv preprint arXiv:2209.03003 , year=

    Flow straight and fast: Learning to generate and transfer data with rectified flow , author=. arXiv preprint arXiv:2209.03003 , year=

    Pith/arXiv arXiv

  36. [36]

    CVPR , year=

    On distillation of guided diffusion models , author=. CVPR , year=

  37. [37]

    ICLR , year=

    Progressive distillation for fast sampling of diffusion models , author=. ICLR , year=

  38. [38]

    ICML , year=

    Consistency models , author=. ICML , year=

  39. [39]

    ICML , year=

    Fast sampling of diffusion models via operator learning , author=. ICML , year=

  40. [40]

    CVPR , year=

    One-step diffusion with distribution matching distillation , author=. CVPR , year=

  41. [41]

    ICLR , year=

    Instaflow: One step is enough for high-quality diffusion-based text-to-image generation , author=. ICLR , year=

  42. [42]

    NeurIPS , year=

    One-step diffusion distillation via deep equilibrium models , author=. NeurIPS , year=

  43. [43]

    arXiv preprint arXiv:2405.14867 , year=

    Improved Distribution Matching Distillation for Fast Image Synthesis , author=. arXiv preprint arXiv:2405.14867 , year=

    Pith/arXiv arXiv

  44. [44]

    arXiv preprint arXiv:2405.07510 , year=

    Perflow: Piecewise rectified flow as universal plug-and-play accelerator , author=. arXiv preprint arXiv:2405.07510 , year=

    Pith/arXiv arXiv

  45. [45]

    ICCV , year=

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data , author=. ICCV , year=

  46. [46]

    COMMUN ACM , year=

    Generative adversarial networks , author=. COMMUN ACM , year=

  47. [47]

    NeurIPS , year=

    Visual autoregressive modeling: Scalable image generation via next-scale prediction , author=. NeurIPS , year=

  48. [48]

    CVPR , year=

    Seesr: Towards semantics-aware real-world image super-resolution , author=. CVPR , year=

  49. [49]

    ECCV , year=

    Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization , author=. ECCV , year=

  50. [50]

    ECCV , year=

    Diffbir: Towards blind image restoration with generative diffusion prior , author=. ECCV , year=

  51. [51]

    CVPR , year=

    Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild , author=. CVPR , year=

  52. [52]

    CVPR , year=

    SinSR: Diffusion-based image super-resolution in a single step , author=. CVPR , year=

  53. [53]

    arXiv preprint arXiv:2406.08177 , year=

    One-Step Effective Diffusion Network for Real-World Image Super-Resolution , author=. arXiv preprint arXiv:2406.08177 , year=

    Pith/arXiv arXiv

  54. [54]

    arXiv preprint arXiv:2410.04224 , year=

    Distillation-free one-step diffusion for real-world image super-resolution , author=. arXiv preprint arXiv:2410.04224 , year=

    Pith/arXiv arXiv

  55. [55]

    arXiv preprint arXiv:2408.07476 , year=

    One step diffusion-based super-resolution with time-aware distillation , author=. arXiv preprint arXiv:2408.07476 , year=

    Pith/arXiv arXiv

  56. [56]

    arXiv preprint arXiv:2411.18263 , year=

    TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution , author=. arXiv preprint arXiv:2411.18263 , year=

    Pith/arXiv arXiv

  57. [57]

    IJCV , year=

    Exploiting diffusion prior for real-world image super-resolution , author=. IJCV , year=

  58. [58]

    NeurIPS , year=

    Resshift: Efficient diffusion model for image super-resolution by residual shifting , author=. NeurIPS , year=

  59. [59]

    CVPR , year=

    The unreasonable effectiveness of deep features as a perceptual metric , author=. CVPR , year=

  60. [60]

    2019 , eprint=

    Toward Real-World Single Image Super-Resolution: A New Benchmark and A New Model , author=. 2019 , eprint=

    2019

  61. [61]

    TPAMI , year=

    Image quality assessment: Unifying structure and texture similarity , author=. TPAMI , year=

  62. [62]

    TIP , year=

    A feature-enriched completely blind image quality evaluator , author=. TIP , year=

  63. [63]

    ICCV , year=

    Musiq: Multi-scale image quality transformer , author=. ICCV , year=

  64. [64]

    CVPR , year=

    Maniqa: Multi-dimension attention network for no-reference image quality assessment , author=. CVPR , year=

  65. [65]

    AAAI , year=

    Exploring clip for assessing the look and feel of images , author=. AAAI , year=

  66. [66]

    CVPR , year=

    A Style-Based Generator Architecture for Generative Adversarial Networks , author=. CVPR , year=

  67. [67]

    CVPRW , year=

    Lsdir: A large scale dataset for image restoration , author=. CVPRW , year=

  68. [68]

    CVPRW , year=

    Ntire 2017 challenge on single image super-resolution: Dataset and study , author=. CVPRW , year=

    2017

  69. [69]

    IEEE TIP , year=

    Topiq: A top-down approach from semantics to distortions for image quality assessment , author=. IEEE TIP , year=

  70. [70]

    arXiv preprint arXiv:2312.17090 , year=

    Q-align: Teaching lmms for visual scoring via discrete text-defined levels , author=. arXiv preprint arXiv:2312.17090 , year=

    Pith/arXiv arXiv

  71. [71]

    arXiv preprint arXiv:2010.14701 , year=

    Scaling laws for autoregressive generative modeling , author=. arXiv preprint arXiv:2010.14701 , year=

    Pith/arXiv arXiv 2010

  72. [72]

    arXiv preprint arXiv:2106.09685 , year=

    Lora: Low-rank adaptation of large language models , author=. arXiv preprint arXiv:2106.09685 , year=

    Pith/arXiv arXiv

  73. [73]

    International Conference on Computer Vision (ICCV) , year=

    StableSR: Boosting Real-World Image Super-Resolution via Time-Aware Diffusion , author=. International Conference on Computer Vision (ICCV) , year=

  74. [74]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. ICLR , volume=

  75. [75]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Fine-structure preserved real-world image super-resolution via transfer vae training , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  76. [76]

    Proceedings of the 30th ACM International Conference on Multimedia , pages=

    Real-world blind super-resolution via feature matching with implicit high-resolution priors , author=. Proceedings of the 30th ACM International Conference on Multimedia , pages=

  77. [77]

    arXiv preprint arXiv:2307.12348 , year=

    Resshift: Efficient diffusion model for image super-resolution by residual shifting , author=. arXiv preprint arXiv:2307.12348 , year=

    Pith/arXiv arXiv

  78. [78]

    Journal of Machine Learning Research , volume=

    Visualizing Data using t-SNE , author=. Journal of Machine Learning Research , volume=

  79. [79]

    arXiv preprint arXiv:2503.11073 , year=

    Perceive, Understand and Restore: Real-World Image Super-Resolution with Autoregressive Multimodal Generative Models , author=. arXiv preprint arXiv:2503.11073 , year=

    Pith/arXiv arXiv

  80. [80]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Classsr: A general framework to accelerate super-resolution networks by data characteristic , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Showing first 80 references.