Pith. sign in

REVIEW 4 major objections 5 minor 79 references

Bridging Information Asymmetry: A Hierarchical Framework for Deterministic Blind Face Restoration

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Blind face restoration can be made deterministic by anchoring a diffusion generator with semantic tokens and pruning its output space with preference-aware reinforcement learning.

desk verdict The RL-plus-AR-token system is a plausible engineering combination, but the headline 'deterministic, identity-consistent' claim rests on variance measured on the very rewards they optimized, and their own Table II shows the RL-tuned variant loses identity. read the letter →

arxiv 2601.19506 v4 pith:LTCMVKEU submitted 2026-01-27 cs.CV

Zhengjian Yao , Jiakui Hu , Kaiwen Li , Hangzhou He , Xinliang Zhang , Shuang Zeng , Lei Zhu , Yanye Lu This is my paper · ORCID
classification cs.CV
keywords BlindfacerestorationInformationasymmetryAutoregressivemodelsDiffusionOn-policyreinforcementlearningPreferencealignmentDeterministicIdentitypreservation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that blind face restoration fails not because generators are weak but because the task is information-asymmetric: a severely degraded face can map to many plausible but identity-wrong faces. It proposes Pref-Restore, which attacks both ends of the pipeline—an autoregressive semantic branch turns image and text cues into discrete tokens that anchor the high-level structure, while a preference-aware reinforcement-learning stage prunes the diffusion output distribution toward human-aligned, identity-consistent results. The central claim is that this two-ended design makes restoration effectively deterministic, cutting the median standard deviation of preference scores across repeated sampling by about 50 percent while matching or beating prior methods on perceptual and face-specific metrics. A sympathetic reader would care because deterministic, identity-faithful restoration is what forensic, medical, and archival applications require.

What carries the argument

Three coordinated components carry the argument. (1) An AR-based semantic integrator: a quantized visual-language model trained by next-token prediction turns degraded pixels plus a text caption into discrete semantic tokens that serve as a stable high-level anchor. (2) A continuous diffusion generator built on conditional flow matching, trained in two sub-stages (semantic-to-diffusion alignment, then texture-to-diffusion alignment) so the anchor and low-level VAE features jointly condition the velocity field. (3) Preference-aware fine-tuning via an on-policy RL scheme applied to the forward diffusion flow (the paper calls this DiffusionNFT): group rollouts produce K candidates scored by a f

What would settle it

Run N=16 restorations of the same degraded face under different random seeds and compute the standard deviation of identity metrics (face-embedding angle and landmark distance) rather than preference scores. If those spreads do not shrink alongside the preference-score spread—or if the mean identity error grows, as the paper's own Table II already shows for stage 2—then the claimed determinism is a property of the reward metric, not of the restored identity.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the ill-posedness of blind face restoration can be re-balanced rather than merely regularized. Pref-Restore augments the degraded input with dense semantic tokens generated by an autoregressive visual-language model, then uses on-policy reinforcement learning on the forward diffusion flow to make the generator prefer high-fidelity trajectories and avoid hallucinated ones. The paper reports that the resulting model matches or exceeds prior methods on synthetic and real benchmarks, with stronger identity-sensitive fidelity after stage one and higher perceptual quality after stage two, and that across 16 repeated restorations the spread of prefere

Load-bearing premise

The load-bearing premise is that the drop in variance of preference scores (human-preference, CLIP-alignment, and aesthetic-quality rewards) across repeated sampling measures a genuine reduction in restoration uncertainty; if the reward model simply makes the model consistently produce a pleasing but wrong face, the deterministic claim collapses.

Editorial extensions

If this is right

  • If the central claim is right, a single degraded face will restore to the same identity-consistent output across repeated sampling runs, which is what forensic, archival, and medical uses demand.
  • Text captions act as a steerable semantic channel: richer captions improve semantic fidelity (the paper reports double-digit gains in CLIP-image similarity and DreamSim), so the same restoration engine can be directed by language at inference time.
  • The two variants make the fidelity–quality trade-off explicit and controllable: stopping after stage 1 preserves identity best, while stage 2 maximizes perceptual scores but measurably worsens identity metrics—so users can choose where to sit on the curve.
  • Because the RL stage needs only about 30 fine-tuning steps and operates on the forward velocity field rather than the reverse sampling policy, the same distribution-pruning recipe can be attached to other diffusion-based restoration models without re-deriving policy gradients.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A stronger test of 'deterministic restoration' than the paper reports would measure the spread of identity metrics (face-embedding angle and landmark distance) across the same 16 runs; if those spreads do not tighten alongside the preference-score spread, the 50% reduction may describe reward collapse rather than faithful reconstruction.
  • The paper's own Table II shows stage 2 degrades identity metrics (embedding angle from 54.06 to 73.99), so 'preference-aligned' currently means aesthetically preferred, not identity-faithful; adding an explicit identity-reward term would be the natural next step.
  • Because captions are generated by a separate visual-language model in deployment, caption errors become a new failure mode: a wrong attribute in the text could anchor the restoration to the wrong identity. A natural stress test is corrupting or dropping captions and measuring identity fidelity and determinism.
  • The recipe—autoregressive semantic anchoring plus preference-aware forward-flow RL—generalizes beyond faces to other ill-posed inverse problems such as blind super-resolution, deblurring, or medical image restoration where structure matters more than aesthetic plausibility.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Pref-Restore, a hierarchical blind face restoration framework combining an auto-regressive semantic integrator (image + text to discrete tokens) with a diffusion-based generator, trained in two stages: Stage 1 performs semantic-to-diffusion and texture-to-diffusion alignment, and Stage 2 fine-tunes the diffusion velocity field via DiffusionNFT-style on-policy RL with a composite reward (HPSv2 + Clip Score + Pick Score). The authors report SOTA results on synthetic and real-world benchmarks and, as the central claim, assert that Stage 2 reduces restoration stochasticity by about 50% (Sec. V-H, Fig. 8) and yields deterministic, identity-consistent restoration. Two variants are offered: Pref-Restore Fidelity (after Stage 1) and Pref-Restore Quality (after Stage 2).

Significance. If the deterministic-restoration claim were fully supported, the paper would make a valuable contribution: using AR-derived semantic anchors and preference-based RL to stabilize diffusion restoration is a plausible route toward mitigating hallucination and identity drift in blind face restoration. The paper deserves credit for its extensive ablations (Table IX, Fig. 3, Table VII), the staged training design, and the decision to provide two clearly characterized variants. However, the central claim is not currently supported: the variance reduction is measured on the very reward function being optimized, and the authors' own Table II shows the Stage 2 'Quality' variant degrading identity metrics substantially. As written, the evidence supports 'preference-aligned aesthetic stabilization' rather than 'deterministic, identity-consistent restoration.' A major revision is needed to reframe the claim or add evidence from identity-based uncertainty metrics.

major comments (4)
  1. [Sec. V-H and Eq. (15)] The '50% reduction in restoration stochasticity' is measured as the standard deviation of the Sum Score (HPSv2 + PickScore + ClipScore) over 16 runs. This is exactly the reward R_pref optimized in Eq. (15) and tracked in Fig. 6. Reduced variance of the training objective across seeds is expected when the model converges to a reward-maximizing mode, and it can also result from mode collapse. It does not measure uncertainty over the restored identity or output distribution. The paper needs independent uncertainty metrics (e.g., std of ArcFace embeddings, landmark positions, or pixel-space distance between restorations) before the determinism claim can be accepted.
  2. [Table II and Sec. V-B/G] The abstract claims 'stronger identity-sensitive fidelity' and the Method section claims Stage 2 'prunes trajectories that deviate from human aesthetic and fidelity standards.' Yet Table II shows Pref-Restore Quality (the Stage 2 variant) has LMD 8.3588 and ArcFace Deg 73.9882, much worse than Pref-Restore Fidelity (5.1337 / 54.0623), and worse than several baselines. Table IX attributes the identity gains to Stage 1.2, not to preference RL. Thus the stage that produces the claimed stochasticity reduction is also the stage that degrades identity. The deterministic-identity reading is contradicted by the authors' own data; the claim should be restricted to aesthetic stabilization, or Stage 2 must be shown to preserve identity via an identity-aware reward.
  3. [Tables III-VI] On the four real-world datasets, the reported metrics are almost entirely no-reference aesthetic/quality metrics (MUSIQ, CLIPIQA+, MANIQA, topiq, DSL-FIQA) plus FID. Identity-sensitive metrics (ArcFace Deg, LMD) that appear in the synthetic CelebA-Test table are omitted. Since the paper's headline contribution is 'identity-sensitive fidelity' and 'lower restoration uncertainty,' real-world identity metrics (or a clear explanation for their omission) are necessary to support the generalization claim.
  4. [Sec. V-E and Table VIII] The text reporting reward gains is internally inconsistent: 'the Pick Score rises from 76.98 to 86.66 (+12.6%)' while Table VIII lists Pick Score 20.01 to 22.53, and HPSv2 is listed as 22.28 to 30.22 in the table but appears in a different scale in the text. If a rescaling or normalization is used, it must be stated explicitly; otherwise the main quantitative evidence for Stage 2's benefit is not verifiable.
minor comments (5)
  1. [Eq. (6) and Appendix A] The formal derivation in Appendix A is a standard MAP-plus-energy-based-prior rewrite; Eq. (7) is the elementary 'conditioning reduces entropy' inequality. The paper should not present this as a new theoretical proof. The actual contribution is the empirical architecture, which is fine, but the framing should be modest.
  2. [Table I / Eqs. (12)-(14)] The values of gamma, lambda, Z, and the reward composition weights are not reported. Table I lists learning rates and batch size but not these Stage-2-specific hyperparameters. Please provide them for reproducibility.
  3. [Fig. 8] The figure and text do not specify which model is the 'Base Model' (presumably Stage 1?) and how the Sum Score is computed/scaled. The boxplot axes are unlabeled beyond 'Std of Scores'; clarify the normalization of HPSv2/PickScore/ClipScore in this analysis.
  4. [Table V caption] The caption reads 'Wilder-Test' but the dataset is WIDER-Test. Fix the typo.
  5. [Related Work / Contributions] Given DiffusionReward [46], IRPO [49], RealSR-R1 [50], and TTPO [51] already apply RL/preference methods to restoration, the claim to 'pioneer' on-policy RL for BFR should be softened or carefully differentiated from these prior works.

Circularity Check

2 steps flagged · score 6.0 of 10

Deterministic claim is self-referential: the '50% reduction in restoration stochasticity' is computed on the std of HPSv2+PickScore+ClipScore—the exact reward optimized in Eq. 15—while Stage-2 identity metrics degrade.

  1. fitted input called prediction [Sec. V-E (Reward Composition) and Sec. V-H (Deterministic Analysis), Eq. (15), Fig. 8]
    "Our preference reward R_pref is a synergistic ensemble of three state-of-the-art multi-modal scoring models: HPSv2 [72] (Human Preference Score), Clip Score [73] (Semantic alignment), and Pick Score [74] (General aesthetic and quality prior). ... We calculate the standard deviation of various perceptual metrics, including HPSv2, PickScore, ClipScore, and their Sum Score, to quantify the sampling variance. ... the median standard deviation of the Sum Score drops significantly from approximately 0.032 to 0.016, representing a 50% reduction in restoration stochasticity."

    The 'Sum Score' whose variance is reported as 'restoration stochasticity' is the sum of HPSv2, PickScore, and ClipScore—exactly the reward R_pref optimized in Eq. (15) through the normalized r^(k) in Eq. (12). The RL loss is built from this reward, so reporting low variance of the same reward is measuring the training objective itself, not an independent posterior-entropy estimate over true identity. The paper's own Table II shows the Stage-2 Quality variant degrades identity metrics (ArcFace Deg 54.06→73.99, LMD 5.13→8.36), so the reduced variance is consistent with aesthetic mode collapse rather than identity-consistent determinism.

  2. self definitional [Sec. V-E, Table VIII and Fig. 6]
    "As summarized in Table VIII, comparing the PrefRestore-Base (end of Stage 1) with the PrefRestore-RL (end of Stage 2) reveals substantial improvements across all preference-related dimensions."

    Table VIII's 'preference-related dimensions' are HPSv2, Clip Score, and Pick Score—the exact additive components of R_pref in Eq. (15). The evaluation therefore reports the training reward's own components; the same frozen scorers provide both the RL gradient and the reported 'preference' gains. While mean-reward improvement is an expected sanity check, presenting it as evidence of general 'preference-aligned restoration' or 'solution-space pruning' is self-referential: no external or held-out preference signal is used.

full rationale

The circularity is concentrated in Stage 2 and in the paper's deterministic claim. Eq. (15) defines the RL loss using normalized reward r from R_pref, which is explicitly composed of HPSv2 + ClipScore + PickScore. Section V-H then measures 'restoration stochasticity' as the standard deviation of exactly those scores (the Sum Score) across N=16 runs and reports a 50% drop. Thus the headline uncertainty reduction is computed on the training objective itself; it is a self-referential metric, not an independent measure of posterior entropy over the true identity. Table VIII's preference gains are likewise the training reward's own components. The identity-consistency part of the claim is further weakened by the paper's own Table II, where the Stage-2 Quality variant increases ArcFace Deg from 54.06 to 73.99 and LMD from 5.13 to 8.36, and by Appendix I(b), which lists identity-preserving rewards as future work, stating the authors 'primarily utilized an aesthetic reward model.' Stage 1 has independent, non-circular content: the AR semantic anchoring and Texture-to-Diffusion Alignment are validated by external metrics (LMD, ArcFace Deg, LPIPS, FID), and Table IX shows large identity gains from Stage 1.2. No load-bearing self-citation chain or imported uniqueness theorem appears; DiffusionNFT is external prior work and the authors' own AR papers are cited only as related work. The score reflects that the central 'deterministic, identity-consistent restoration' claim partially reduces to a self-referential reward-variance measure, while Stage 1 retains substantial independent evidence.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests on several hand-tuned constants and on the assumptions that captions and preference scores capture identity fidelity. No new physical or ontological entities are introduced; 'semantic anchor', 'implicit velocity proxies', and 'solution space pruning' are mathematical or architectural constructs, not independent entities.

free parameters (6)
  • lambda in Eq. (6) (preference constraint weight) = not reported
    Balances the augmented likelihood against R_pref; no value or schedule is given, yet it controls the fidelity-quality trade-off central to Stage 2.
  • gamma in Eqs. (13)-(14) (implicit velocity interpolation) = not reported
    Sets the strength of positive/negative velocity proxies in the DiffusionNFT loss; no value is reported.
  • Z in Eq. (12) (reward normalization scale) = not reported
    Scales group-centered reward clipping to r in [0,1]; this directly affects which samples become positive or negative exemplars.
  • Reward composition weights for HPSv2 + Clip Score + Pick Score = equal/unspecified
    R_pref is described as a summation, but no calibration or weights are given; this determines exactly what Stage 2 optimizes.
  • p_txt = 0.95, p_rec = 0.1 (training dropouts) = 0.95 caption / 0.05 null; 0.1 degradation bypass
    Chosen by hand in Appendix F; controls how much the model relies on text and on identity-preserving auto-encoding.
  • K candidate rollouts in Stage 2 = not reported
    Group size for reward centering in Eq. (12) is not specified, despite being essential to the on-policy RL update.
assumptions (6)
  • standard math Conditional entropy reduction: H(x|y,S_AR) <= H(x|y)
    Invoked in Eq. (7) to argue semantic augmentation shrinks the posterior. Mathematically true on average, but it does not guarantee the mode becomes identity-correct.
  • domain assumption Text captions are valid and available identity anchors at deployment
    The AR integrator consumes image plus text (Sec. IV-B1, Appendix D); experiments provide detailed captions to Pref-Restore while baselines receive no captions.
  • domain assumption Reward models HPSv2, CLIP Score, and Pick Score are good proxies for human perceptual fidelity
    R_pref in Sec. V-E is optimized and then used as the main success metric; there is no correction for reward overoptimization.
  • domain assumption DiffusionNFT forward-flow contrast retains its guarantees under conditional restoration
    Eqs. (1)-(4) and (15) are adopted from [29] for conditional BFR; no proof or validation is given that velocity-field refinement prunes hallucinations rather than overfitting rewards.
  • domain assumption Frozen diffusion backbone allows semantic token-diffusion alignment through backprop
    Stage 1.1 propagates L_diff through a frozen diffusion backbone to train the projector and AR module; assumes the frozen cross-attention subspace can absorb discrete semantic tokens.
  • ad hoc to paper Reduced std of reward scores over 16 seeds equals deterministic restoration
    Sec. V-H treats lower HPSv2/PickScore/ClipScore variance as a '50% reduction in restoration stochasticity' without a formal definition of restoration uncertainty or identity-level stability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging Information Asymmetry: A Hierarchical Framework for Deterministic Blind Face Restoration." pith.science (2026). https://pith.science/paper/LTCMVKEU

@misc{pith2026260119506,
  author       = {Pith},
  title        = {Pith review of: Bridging Information Asymmetry: A Hierarchical Framework for Deterministic Blind Face Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LTCMVKEU}},
  note         = {Machine review of arXiv:2601.19506}
}
read the original abstract

Blind face restoration remains a persistent challenge due to the inherent ill-posedness of reconstructing holistic structures from severely constrained observations. Current generative paradigms, while capable of synthesizing realistic facial details, remain limited by the under-constrained nature of blind restoration, where severely degraded inputs can be mapped to plausible yet identity-inconsistent outputs. To address this issue, we present Pref-Restore, a hierarchical framework for deterministic BFR. Our design is organized around three complementary principles: (1) Semantic Information Augmentation, where an auto-regressive semantic branch converts image and text cues into structured tokens that provide a stable high-level anchor; (2) Texture-level Fidelity Alignment, where the diffusion generator is trained under this anchor to recover identity-relevant details; and (3) Fidelity-constrained Preference Optimization, where a face-aware reward refines the diffusion trajectory while controlling the quality-fidelity trade-off. Extensive experiments on synthetic and real-world benchmarks show that Pref-Restore achieves state-of-the-art performance, with stronger identity-sensitive fidelity and lower restoration uncertainty across repeated sampling. Systematic ablations further attribute these gains to the proposed hierarchical design, showing the necessity of staged training, the robustness of the text pathway under deployment-faithful conditions, and the benefit of fidelity-constrained preference optimization.

Figures

Figures reproduced from arXiv: 2601.19506 by the authors.

Figure 1
Figure 1. Conceptual illustration of (a) the conventional generative prior paradigm versus (b) our proposed Pref-Restore framework. Existing methods suffer from information asymmetry, where ill-posed priors and sparse inputs lead to stochastic outcomes such as hallucinations or identity loss. Our Pref-Restore resolves this by augmenting input density through AR￾based semantic modeling and pruning the output distribution via o… view at source ↗
Figure 2
Figure 2. The overall framework of Pref-Restore. (a) Hierarchical Restoration Architecture: Our model decouples the blind face restoration task into a discrete semantic stream and a continuous texture stream. The AR-based Semantic Integrator processes degraded observations y and textual instructions T to generate discrete semantic tokens S via Next-Token Prediction (NTP), acting as a global structural anchor. Simultaneously, … view at source ↗
Figure 3
Figure 3. Evolution of FID scores during the hierarchical training process. The curves illustrate the FID (HQ) and FID (FFHQ) results on the CelebA-Test dataset. Stage 1 comprises two critical steps: Step 1 (Semantic-to-Diffusion Alignment) and Step 2 (Texture-to-Diffusion Alignment). The sustained decrease in FID during Stage 1 highlights the scaling effect of our AR-based semantic integrator in aligning distributions. Our f… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of face restoration on the CelebA-HQ dataset. We compare our Pref-Restore (F) and (Q) variants with state-of-the￾art methods. Our models demonstrate superior performance in recovering structural details and high-frequency hair textures compared t…
Figure 6
Figure 6. Figure 6: Evolution of preference-aware rewards during DiffusionNFT fine￾tuning. We employ a dual-axis visualization to demonstrate the synchronized optimization of diverse metrics. The Total Reward (purple line, left axis) represents the aggregate alignment, while HPSv2, Clip S…
Figure 7
Figure 7. Figure 7: Qualitative comparison between Pref-Restore(F) and Pref￾Restore(Q). (a): Situations where Pref-Restore(F) is preferred for its superior preservation of original identity details, whereas the (Q) variant exhibits slight over-beautification. (b): Cases where Pref-Restore…
Figure 8
Figure 8. Figure 8: Quantitative analysis of solution space reduction. We report the standard deviation of scores across N = 16 independent sampling runs for (a) HPSv2, (b) PickScore, (c) ClipScore, and (d) the aggregate Sum Score. Compared to the Base Model, our Pref-Restore consistently…
Figure 9
Figure 9. Figure 9: Progress of the compound reward score on the validation set during Stage 2 RL training. The dashed red line indicates the performance of CodeFormer (21.37). Our Pref-Restore model (dark blue line) rapidly surpasses the baseline within the first 30 steps and consistentl…
Figure 10
Figure 10. Figure 10: Additional qualitative comparison on the CelebA-HQ dataset (Part I). Compared with SOTA methods, Pref-Restore(F) demonstrates superior structural anchors in preserving identity, while Pref-Restore(Q) achieves the highest perceptual quality. For instance, in rows with …
Figure 11
Figure 11. Figure 11: Additional qualitative comparison on challenging cases (Part II). This figure highlights our model’s robustness against complex facial components and occlusions, such as tightly curled hair (row 3) and large accessories like cowboy hats (row 4). By internalizing solut…
Figure 12
Figure 12. Figure 12: Qualitative comparison on real-world face degradation datasets. Each row presents a challenging real-world case characterized by severe noise, blur, and low resolution. Our method, Pref-Restore, consistently produces the clearest and most structurally accurate results…
Figure 13
Figure 13. Figure 13: Analysis of holistic structural consistency and global artifacts. (a) Comparison of accessory restoration: Baseline methods often focus on local facial sharpness at the expense of global geometry, leading to warped and unrealistic hat shapes. (b) Restoration under ext…
Figure 14
Figure 14. Figure 14: Visual comparison demonstrating the necessity of our hierarchical alignment. The third column (Stage 1.1) illustrates the results using only Semantic-to-Diffusion Alignment (equivalent to the training paradigm of Blip-3o Next), where the lack of fine-grained structura…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 24 linked inside Pith

  1. [1]

    Towards real-world blind face restoration with generative facial prior,

    X. Wang, Y . Li, H. Zhang, and Y . Shan, “Towards real-world blind face restoration with generative facial prior,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 9168– 9178

    2021

  2. [2]

    Gan prior embedded network for blind face restoration in the wild,

    T. Yang, P. Ren, X. Xie, and L. Zhang, “Gan prior embedded network for blind face restoration in the wild,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 672– 681

    2021

  3. [3]

    Vqfr: Blind face restoration with vector-quantized dictionary and parallel decoder,

    Y . Gu, X. Wang, L. Xie, C. Dong, G. Li, Y . Shan, and M.-M. Cheng, “Vqfr: Blind face restoration with vector-quantized dictionary and parallel decoder,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 126–143

    2022

  4. [4]

    Towards robust blind face restoration with codebook lookup transformer,

    S. Zhou, K. Chan, C. Li, and C. C. Loy, “Towards robust blind face restoration with codebook lookup transformer,”Advances in Neural Information Processing Systems, vol. 35, pp. 30 599–30 611, 2022

    2022

  5. [5]

    Difface: Blind face restoration with diffused error contraction,

    Z. Yue and C. C. Loy, “Difface: Blind face restoration with diffused error contraction,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  6. [6]

    Dr2: Diffusion-based robust degradation remover for blind face restoration,

    Z. Wang, Z. Zhang, X. Zhang, H. Zheng, M. Zhou, Y . Zhang, and Y . Wang, “Dr2: Diffusion-based robust degradation remover for blind face restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1704–1713

    2023

  7. [7]

    Dual asso- ciated encoder for face restoration,

    Y .-J. Tsai, Y .-L. Liu, L. Qi, K. C. Chan, and M.-H. Yang, “Dual asso- ciated encoder for face restoration,”arXiv preprint arXiv:2308.07314, 2023

    arXiv 2023

  8. [8]

    Blind image restoration by anisotropic regularization,

    Y .-L. You and M. Kaveh, “Blind image restoration by anisotropic regularization,”IEEE Transactions on Image Processing, vol. 8, no. 3, pp. 396–407, 1999. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

    1999

Show all 79 references
  1. [9]

    Blind image deconvolution,

    D. Kundur and D. Hatzinakos, “Blind image deconvolution,”IEEE signal processing magazine, vol. 13, no. 3, pp. 43–64, 2002

    2002

  2. [10]

    Deep variational network toward blind image restoration,

    Z. Yue, H. Yong, Q. Zhao, L. Zhang, D. Meng, and K.-Y . K. Wong, “Deep variational network toward blind image restoration,”IEEE Trans- actions on Pattern Analysis and Machine Intelligence, vol. 46, no. 11, pp. 7011–7026, 2024

    2024

  3. [11]

    Diffbir: Towards blind image restoration with generative diffusion prior,

    X. Lin, J. He, Z. Chen, Z. Lyu, B. Dai, F. Yu, W. Ouyang, Y . Qiao, and C. Dong, “Diffbir: Towards blind image restoration with generative diffusion prior,”arXiv preprint arXiv:2308.15070, 2023

    2023 arXiv

  4. [12]

    Universal image restora- tion pre-training via masked degradation classification,

    J. Hu, Z. Yao, L. Jin, Y . Chen, and Y . Lu, “Universal image restora- tion pre-training via masked degradation classification,”arXiv preprint arXiv:2510.13282, 2025

    2025

  5. [13]

    A survey on all-in- one image restoration: Taxonomy, evaluation and future trends,

    J. Jiang, Z. Zuo, G. Wu, K. Jiang, and X. Liu, “A survey on all-in- one image restoration: Taxonomy, evaluation and future trends,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  6. [14]

    Photo-realistic single image super-resolution using a generative adversarial network,

    C. Ledig, L. Theis, F. Husz ´ar, J. Caballero, A. Cunningham, A. Acosta, A. Aitken, A. Tejani, J. Totz, Z. Wanget al., “Photo-realistic single image super-resolution using a generative adversarial network,” in Proceedings of the IEEE conference on computer vision and pattern r...

    2017

  7. [15]

    Esrgan: Enhanced super-resolution generative adversar- ial networks,

    X. Wang, K. Yu, S. Wu, J. Gu, Y . Liu, C. Dong, Y . Qiao, and C. Change Loy, “Esrgan: Enhanced super-resolution generative adversar- ial networks,” inProceedings of the European conference on computer vision (ECCV) workshops, 2018, pp. 0–0

    2018

  8. [16]

    Real-world super- resolution via kernel estimation and noise injection,

    X. Ji, Y . Cao, Y . Tai, C. Wang, J. Li, and F. Huang, “Real-world super- resolution via kernel estimation and noise injection,” inproceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 466–467

    2020

  9. [17]

    Auto-encoding variational bayes,

    D. P. Kingma and M. Welling, “Auto-encoding variational bayes,”arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  10. [18]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” Advances in neural information processing systems, vol. 27, 2014

    2014

  11. [19]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840– 6851, 2020

    2020

  12. [20]

    Generative modeling by estimating gradients of the data distribution,

    Y . Song and S. Ermon, “Generative modeling by estimating gradients of the data distribution,”Advances in neural information processing systems, vol. 32, 2019

    2019

  13. [21]

    Transfer between modalities with metaqueries,

    X. Pan, S. N. Shukla, A. Singh, Z. Zhao, S. K. Mishra, J. Wang, Z. Xu, J. Chen, K. Li, F. Juefei-Xuet al., “Transfer between modalities with metaqueries,”arXiv preprint arXiv:2504.06256, 2025

    2025 arXiv

  14. [22]

    Blip3o-next: Next frontier of native image generation,

    J. Chen, L. Xue, Z. Xu, X. Pan, S. Yang, C. Qin, A. Yan, H. Zhou, Z. Chen, L. Huanget al., “Blip3o-next: Next frontier of native image generation,”arXiv preprint arXiv:2510.15857, 2025

    2025

  15. [23]

    Vision as a dialect: Unifying visual under- standing and generation via text-aligned representations,

    J. Han, H. Chen, Y . Zhao, H. Wang, Q. Zhao, Z. Yang, H. He, X. Yue, and L. Jiang, “Vision as a dialect: Unifying visual under- standing and generation via text-aligned representations,”arXiv preprint arXiv:2506.18898, 2025

    2025 arXiv

  16. [24]

    Auto-regressively gen- erating multi-view consistent images,

    J. Hu, Y . Yang, J. Liu, J. Wu, C. Zhao, and Y . Lu, “Auto-regressively gen- erating multi-view consistent images,”arXiv preprint arXiv:2506.18527, 2025

    2025 arXiv

  17. [25]

    Omni-view: Unlocking how generation facilitates under- standing in unified 3d model based on multiview images,

    J. Hu, S. Zhao, Q.-G. Chen, X. Qiu, J. Liu, Z. Xu, W. Luo, K. Zhang, and Y . Lu, “Omni-view: Unlocking how generation facilitates under- standing in unified 3d model based on multiview images,”arXiv preprint arXiv:2511.07222, 2025

    2025

  18. [26]

    Uniworld: High-resolution semantic encoders for unified visual understanding and generation,

    B. Lin, Z. Li, X. Cheng, Y . Niu, Y . Ye, X. He, S. Yuan, W. Yu, S. Wang, Y . Geet al., “Uniworld: High-resolution semantic encoders for unified visual understanding and generation,”arXiv preprint arXiv:2506.03147, 2025

    2025 arXiv

  19. [27]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models,

    Z. Shao, P. Wang, Q. Zhu, R. Xu, J. Song, X. Bi, H. Zhang, M. Zhang, Y . Li, Y . Wuet al., “Deepseekmath: Pushing the limits of mathematical reasoning in open language models,”arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  20. [28]

    Flow-grpo: Training flow matching models via online rl,

    J. Liu, G. Liu, J. Liang, Y . Li, J. Liu, X. Wang, P. Wan, D. Zhang, and W. Ouyang, “Flow-grpo: Training flow matching models via online rl,” arXiv preprint arXiv:2505.05470, 2025

    2025 arXiv

  21. [29]

    Diffusionnft: Online diffusion reinforcement with forward process,

    K. Zheng, H. Chen, H. Ye, H. Wang, Q. Zhang, K. Jiang, H. Su, S. Ermon, J. Zhu, and M.-Y . Liu, “Diffusionnft: Online diffusion reinforcement with forward process,”arXiv preprint arXiv:2509.16117, 2025

    2025 arXiv

  22. [30]

    Fsrnet: End-to-end learning face super-resolution with facial priors,

    Y . Chen, Y . Tai, X. Liu, C. Shen, and J. Yang, “Fsrnet: End-to-end learning face super-resolution with facial priors,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2492–2501

    2018

  23. [31]

    Pro- gressive semantic-aware style transformation for blind face restoration,

    C. Chen, X. Li, L. Yang, X. Lin, L. Zhang, and K.-Y . K. Wong, “Pro- gressive semantic-aware style transformation for blind face restoration,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 11 896–11 905

    2021

  24. [32]

    Face super-resolution guided by 3d facial priors,

    X. Hu, W. Ren, J. LaMaster, X. Cao, X. Li, Z. Li, B. Menze, and W. Liu, “Face super-resolution guided by 3d facial priors,” inComputer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IV 16. Springer, 2020, pp. 763–780

    2020

  25. [33]

    Exemplar guided face image super- resolution without facial landmarks,

    B. Dogan, S. Gu, and R. Timofte, “Exemplar guided face image super- resolution without facial landmarks,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2019, pp. 0–0

    2019

  26. [34]

    Blind face restoration via deep multi-scale component dictionaries,

    X. Li, C. Chen, S. Zhou, X. Lin, W. Zuo, and L. Zhang, “Blind face restoration via deep multi-scale component dictionaries,” inEuropean conference on computer vision. Springer, 2020, pp. 399–415

    2020

  27. [36]

    Image processing using multi-code gan prior,

    J. Gu, Y . Shen, and B. Zhou, “Image processing using multi-code gan prior,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 3012–3021

    2020

  28. [37]

    Pulse: Self- supervised photo upsampling via latent space exploration of generative models,

    S. Menon, A. Damian, S. Hu, N. Ravi, and C. Rudin, “Pulse: Self- supervised photo upsampling via latent space exploration of generative models,” inProceedings of the ieee/cvf conference on computer vision and pattern recognition, 2020, pp. 2437–2445

    2020

  29. [38]

    Perceive, understand and re- store: Real-world image super-resolution with autoregressive multimodal generative models,

    H. Wei, S. Liu, C. Yuan, and L. Zhang, “Perceive, understand and re- store: Real-world image super-resolution with autoregressive multimodal generative models,”arXiv preprint arXiv:2503.11073, 2025

    2025 arXiv

  30. [40]

    Promptir: Prompting for all-in-one image restoration,

    V . Potlapalli, S. W. Zamir, S. H. Khan, and F. Shahbaz Khan, “Promptir: Prompting for all-in-one image restoration,”Advances in Neural Infor- mation Processing Systems, vol. 36, pp. 71 275–71 293, 2023

    2023

  31. [41]

    Instructir: High-quality image restoration following human instructions,

    M. V . Conde, G. Geigle, and R. Timofte, “Instructir: High-quality image restoration following human instructions,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 1–21

    2024

  32. [42]

    Controlling vision-language models for multi-task image restoration,

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sj ¨olund, and T. B. Sch ¨on, “Controlling vision-language models for multi-task image restoration,” arXiv preprint arXiv:2310.01018, 2023

    2023 arXiv

  33. [43]

    Crafting a toolchain for image restoration by deep reinforcement learning,

    K. Yu, C. Dong, L. Lin, and C. C. Loy, “Crafting a toolchain for image restoration by deep reinforcement learning,” inProceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 2443–2452

    2018

  34. [44]

    Path-restore: Learn- ing network path selection for image restoration,

    K. Yu, X. Wang, C. Dong, X. Tang, and C. C. Loy, “Path-restore: Learn- ing network path selection for image restoration,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    2021

  35. [45]

    Training language models to follow instructions with human feedback,

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Rayet al., “Training language models to follow instructions with human feedback,”Advances in neural information processing systems, vol. 35, pp. 27 730–27 744, 2022

    2022

  36. [46]

    Diffusionreward: Enhancing blind face restoration through reward feedback learning,

    B. Wu, W. Wang, Y . Liu, Z. Li, and Y . Zhao, “Diffusionreward: Enhancing blind face restoration through reward feedback learning,” arXiv preprint arXiv:2505.17910, 2025

    2025 arXiv

  37. [47]

    Direct preference optimization: Your language model is secretly a reward model,

    R. Rafailov, A. Sharma, E. Mitchell, C. D. Manning, S. Ermon, and C. Finn, “Direct preference optimization: Your language model is secretly a reward model,”Advances in neural information processing systems, vol. 36, pp. 53 728–53 741, 2023

    2023

  38. [48]

    Dspo: Direct semantic preference optimization for real-world image super- resolution,

    M. Cai, S. Li, W. Li, X. Huang, H. Chen, J. Hu, and Y . Wang, “Dspo: Direct semantic preference optimization for real-world image super- resolution,”arXiv preprint arXiv:2504.15176, 2025

    2025 arXiv

  39. [49]

    Irpo: Boosting image restoration via post-training grpo,

    H. X. Y . Liu, B. Jiang, J. Peng, D. Luo, X. Hu, S. Yan, and H. Li, “Irpo: Boosting image restoration via post-training grpo,”arXiv preprint arXiv:2512.00814, 2025

    2025 arXiv

  40. [50]

    Realsr-r1: Reinforcement learning for real-world image super-resolution with vision-language chain-of-thought,

    J. Qiao, M. Cai, W. Li, Y . Liu, X. Huang, G. He, J. Xie, J. Hu, X. Chen, and S. Lin, “Realsr-r1: Reinforcement learning for real-world image super-resolution with vision-language chain-of-thought,”arXiv preprint arXiv:2506.16796, 2025

    2025 arXiv

  41. [51]

    Test- time preference optimization for image restoration,

    B. Li, X. Li, J. Xu, J. Guo, W. Li, R. Pei, and Z. Chen, “Test- time preference optimization for image restoration,”arXiv preprint arXiv:2511.19169, 2025

    2025

  42. [52]

    Bridging supervised learning and reinforce- ment learning in math reasoning,

    H. Chen, K. Zheng, Q. Zhang, G. Cui, Y . Cui, H. Ye, T.-Y . Lin, M.-Y . Liu, J. Zhu, and H. Wang, “Bridging supervised learning and reinforce- ment learning in math reasoning,”arXiv preprint arXiv:2505.18116, 2025

    2025

  43. [53]

    A mathematical theory of communication,

    C. E. Shannon, “A mathematical theory of communication,”The Bell System Technical Journal, vol. 27, no. 3, pp. 379–423, 1948. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15

    1948

  44. [54]

    A tutorial on energy-based learning,

    Y . LeCun, S. Chopra, R. Hadsell, M. Ranzato, and F. Huang, “A tutorial on energy-based learning,”Predicting structured data, vol. 1, no. 0, 2006

    2006

  45. [55]

    Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features,

    M. Tschannen, A. Gritsenko, X. Wang, M. F. Naeem, I. Alabdul- mohsin, N. Parthasarathy, T. Evans, L. Beyer, Y . Xia, B. Mustafa et al., “Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features,”arXiv preprint arXiv...

    2025 arXiv

  46. [56]

    Sana 1.5: Efficient scaling of training-time and inference-time compute in linear diffusion transformer,

    E. Xie, J. Chen, Y . Zhao, J. Yu, L. Zhu, C. Wu, Y . Lin, Z. Zhang, M. Li, J. Chenet al., “Sana 1.5: Efficient scaling of training-time and inference-time compute in linear diffusion transformer,”arXiv preprint arXiv:2501.18427, 2025

    2025 arXiv

  47. [57]

    A style-based generator architecture for generative adversarial networks,

    T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4401– 4410

    2019

  48. [58]

    Progressive growing of gans for improved quality, stability, and variation,

    T. Karras, “Progressive growing of gans for improved quality, stability, and variation,”arXiv preprint arXiv:1710.10196, 2017

    2017 arXiv

  49. [59]

    Labeled faces in the wild: A database forstudying face recognition in unconstrained environments,

    G. B. Huang, M. Mattar, T. Berg, and E. Learned-Miller, “Labeled faces in the wild: A database forstudying face recognition in unconstrained environments,” inWorkshop on faces in’Real-Life’Images: detection, alignment, and recognition, 2008

    2008

  50. [60]

    Wider face: A face detection benchmark,

    S. Yang, P. Luo, C.-C. Loy, and X. Tang, “Wider face: A face detection benchmark,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 5525–5533

    2016

  51. [61]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    2018

  52. [62]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,”Advances in neural information processing systems, vol. 30, 2017

    2017

  53. [63]

    Musiq: Multi- scale image quality transformer,

    J. Ke, Q. Wang, Y . Wang, P. Milanfar, and F. Yang, “Musiq: Multi- scale image quality transformer,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 5148–5157

    2021

  54. [64]

    Exploring clip for assessing the look and feel of images,

    J. Wang, K. C. Chan, and C. C. Loy, “Exploring clip for assessing the look and feel of images,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 2, 2023, pp. 2555–2563

    2023

  55. [65]

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

    S. Yang, T. Wu, S. Shi, S. Lao, Y . Gong, M. Cao, J. Wang, and Y . Yang, “Maniqa: Multi-dimension attention network for no-reference image quality assessment,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 1191–1200

    2022

  56. [66]

    Arcface: Additive angular margin loss for deep face recognition,

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “Arcface: Additive angular margin loss for deep face recognition,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4690– 4699

    2019

  57. [67]

    IQA-PyTorch: Pytorch toolbox for image quality assessment,

    C. Chen and J. Mo, “IQA-PyTorch: Pytorch toolbox for image quality assessment,” [Online]. Available: https://github.com/chaofengc/ IQA-PyTorch, 2022

    2022

  58. [68]

    Restoreformer++: Towards real-world blind face restoration from undegraded key-value pairs,

    Z. Wang, J. Zhang, T. Chen, W. Wang, and P. Luo, “Restoreformer++: Towards real-world blind face restoration from undegraded key-value pairs,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 12, pp. 15 462–15 476, 2023

    2023

  59. [69]

    Qwen2. 5-vl technical report,

    S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang, S. Wang, J. Tanget al., “Qwen2. 5-vl technical report,”arXiv preprint arXiv:2502.13923, 2025

    2025 arXiv

  60. [70]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PmLR, 2021, pp. 8748–8763

    2021

  61. [71]

    Dreamsim: Learning new dimensions of human visual similarity using synthetic data,

    S. Fu, N. Tamir, S. Sundaram, L. Chai, R. Zhang, T. Dekel, and P. Isola, “Dreamsim: Learning new dimensions of human visual similarity using synthetic data,”arXiv preprint arXiv:2306.09344, 2023

    2023 arXiv

  62. [72]

    Human preference score v2: A solid benchmark for evaluating human pref- erences of text-to-image synthesis,

    X. Wu, Y . Hao, K. Sun, Y . Chen, F. Zhu, R. Zhao, and H. Li, “Human preference score v2: A solid benchmark for evaluating human pref- erences of text-to-image synthesis,”arXiv preprint arXiv:2306.09341, 2023

    2023 arXiv

  63. [73]

    Clipscore: A reference-free evaluation metric for image captioning,

    J. Hessel, A. Holtzman, M. Forbes, R. Le Bras, and Y . Choi, “Clipscore: A reference-free evaluation metric for image captioning,” inProceedings of the 2021 conference on empirical methods in natural language processing, 2021, pp. 7514–7528

    2021

  64. [74]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation,

    Y . Kirstain, A. Polyak, U. Singer, S. Matiana, J. Penna, and O. Levy, “Pick-a-pic: An open dataset of user preferences for text-to-image generation,”Advances in neural information processing systems, vol. 36, pp. 36 652–36 663, 2023

    2023

  65. [75]

    Score-based generative modeling through stochastic differ- ential equations,

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differ- ential equations,” inInternational Conference on Learning Representa- tions, 2021

    2021

  66. [76]

    Qwen-image technical report,

    C. Wu, J. Li, J. Zhou, J. Lin, K. Gao, K. Yan, S.-m. Yin, S. Bai, X. Xu, Y . Chenet al., “Qwen-image technical report,”arXiv preprint arXiv:2508.02324, 2025

    2025 arXiv

  67. [77]

    B. F. Labs, “Flux,” https://github.com/black-forest-labs/flux, 2024

    2024

  68. [78]

    One diffusion step to real- world super-resolution via flow trajectory distillation,

    J. Li, J. Cao, Y . Guo, W. Li, and Y . Zhang, “One diffusion step to real- world super-resolution via flow trajectory distillation,”arXiv preprint arXiv:2502.01993, 2025

    2025 arXiv

  69. [79]

    One-step diffusion with distribution matching distillation,

    T. Yin, M. Gharbi, R. Zhang, E. Shechtman, F. Durand, W. T. Freeman, and T. Park, “One-step diffusion with distribution matching distillation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 6613–6623

    2024

  70. [80]

    Consistency models,

    Y . Song, P. Dhariwal, M. Chen, and I. Sutskever, “Consistency models,” 2023. Zhengjian Yaoreceived the B.S. degree in the School of Mathematics and Statistics from Xi’an Jiaotong University in 2022. He is currently pursuing the Ph.D. degree at the Medical Intelligence Lab, Pe...

    2023

  71. [2021]

    low-reward

    He is currently a joint Ph.D. student of Peking University - Georgia Institute of Technology - Emory University Biomedical Engineering Program. His re- search mainly focuses on self-supervised contrastive learning and medical image processing. Lei Zhureceived the Ph.D. degree ...

    2024

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.