pith. sign in

arxiv: 2606.29198 · v1 · pith:5KWBWDXBnew · submitted 2026-06-28 · 💻 cs.CV

DTI: Dynamic Trajectory Initialization for Generative Face Video Super-Resolution

Yingwei Tang , Chen Yan , Wendi Liu , Qiang Hu , Xiaoyun Zhang This is my paper

Pith reviewed 2026-06-30 07:48 UTC · model grok-4.3

classification 💻 cs.CV
keywords Generative Face Video Super-ResolutionDiffusion ModelsDynamic Trajectory InitializationDiscriminative GuideSignal-to-Noise Ratio AlignmentDiT BackbonePerception-Distortion Trade-offFidelity Improvement
0
0 comments X

The pith

Dynamic Trajectory Initialization reformulates generative face video super-resolution as input-driven directional restoration to improve fidelity with a pretrained diffusion transformer.

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

The paper presents a Dynamic Trajectory Initialization paradigm that turns full generative face video super-resolution into an input-driven directional task. A novel enhancement-and-injection conditioning mechanism is applied to a pretrained DiT backbone to raise fidelity while preserving perceptual quality. A Discriminative Guide trained through objective SNR alignment dynamically chooses the starting sampling point, avoiding fixed sampling and the need for large auxiliary training. Only minor model adaptation and fine-tuning are required to reach state-of-the-art results across multiple metrics and benchmarks. The work also examines how common perceptual metrics relate to overall quality and identifies LPIPS as the most reliable indicator in this setting.

Core claim

By reformulating GFVSR as input-driven directional restoration, the DTI paradigm with enhancement-and-injection conditioning on a pretrained DiT backbone and a Discriminative Guide set via SNR alignment delivers improved fidelity without loss of perceptual quality, reaching SOTA overall performance through only minor adaptation and fine-tuning.

What carries the argument

The Discriminative Guide, trained via objective Signal-to-Noise Ratio alignment, which dynamically selects the starting sampling point for the pretrained DiT backbone.

If this is right

  • Fidelity rises significantly while perceptual quality is maintained.
  • Inference avoids fixed sampling trajectories and large auxiliary training costs.
  • State-of-the-art results appear across diverse metrics and benchmarks after only minor adaptation.
  • LPIPS emerges as the most convincing metric for evaluating comprehensive quality in this domain.
  • The perception-distortion trade-off is shown to limit simultaneous gains in all standard metrics.

Where Pith is reading between the lines

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

  • The same dynamic initialization approach could be tested on non-face video restoration tasks that use diffusion backbones.
  • Combining the conditioning mechanism with other forms of temporal guidance might further reduce artifacts in long sequences.
  • Re-evaluating existing GFVSR benchmarks with LPIPS as the primary metric could change which methods are considered state-of-the-art.
  • If the SNR alignment generalizes, similar guides could be derived for other generative restoration problems without retraining the core model.

Load-bearing premise

The Discriminative Guide trained via SNR alignment can dynamically and effectively set the starting sampling point without large-scale auxiliary training.

What would settle it

A controlled benchmark comparison in which the DTI method fails to exceed prior GFVSR approaches on fidelity measures while matching or exceeding their perceptual scores.

Figures

Figures reproduced from arXiv: 2606.29198 by Chen Yan, Qiang Hu, Wendi Liu, Xiaoyun Zhang, Yingwei Tang.

Figure 1
Figure 1. Figure 1: Left: Qualitative comparison of video restoration results. Right: comparison using normalized metrics (max-participant-based normalization). Our method achieves the best balance between fidelity and perceptual quality. Abstract. As the most perceptually powerful Face Video Super-Reso￾lution (FVSR) method, existing works in Generative FVSR (GFVSR) mainly exploit the generative prior of pretrained diffusion … view at source ↗
Figure 2
Figure 2. Figure 2: (a)(b) some conventional conditioning methods; (c) our novel condition injec￾tion method; (d) represents the real calculated attention parts in adapted DiT, orange ones are only calculated once in a forward. model due to input shifting; ControlNet requires heavy auxiliary components and training; concatenation-flattening-MLP only performs channel-level infor￾mation exchange, resulting in relatively low qua… view at source ↗
Figure 3
Figure 3. Figure 3: (a) represents that LQ preserves perceptually more low-frequency information than high-frequency; (b) visualizes the SNR-aligned timesteps corresponding to real￾world degradations on a test subset. Inspired by ideas of related studies [43] in vision field, the signal-to-noise ratio (SNR) is used in this work to objectively estimate the relationship between a diffusion timestep which represents the injectio… view at source ↗
Figure 4
Figure 4. Figure 4: Model Structure & Pipeline 4 Methodology 4.1 Conditioning and Architecture of GFVSR Model Human faces exhibit a clearly defined subject, an overall structured composi￾tion and pronounced edges. Given the need to extract as much useful visual information as possible from LQ, fine-grained visual feature extractors are our preferred choice. Compared to semantic-aligned contrastive learning models like CLIP [2… view at source ↗
Figure 5
Figure 5. Figure 5: Supervised Training for Discriminative Guide generation process starting from pure noise. But it can’t be treated directly as an intermediate state of noisy latent neither, as LQ is in untractable real-world degradation distribution, not the specific parameterized distribution (Gaussian distribution) learned by diffusion model. So, if we can transfer LQ to a reason￾able nearest point in the probability flo… view at source ↗
Figure 6
Figure 6. Figure 6: Loss comparison Single Double PSNR ↑ 22.15 26.35 SSIM ↑ 0.61 0.74 LPIPS ↓ 0.28 0.17 MUSIQ ↑ 69.20 71.54 CLIP-IQA ↑ 0.56 0.56 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Metrics Discussion. (a): At the top, higher MUSIQ sample has evident visual artifacts; at the bottom, higher PSNR sample is blurry. (b): The relative normalized score trend among LPIPS, fidelity metrics and perceptual metrics [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hyperparameter λ for Perception-Distortion Trade-off 6.3 Controllable Perception-Distortion Trade-off In [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

As the most perceptually powerful Face Video Super-Resolution (FVSR) method, existing works in Generative FVSR (GFVSR) mainly exploit the generative prior of pretrained diffusion models. However, viewed as full generation, they suffer from fixed sampling and expensive inference costs if without large-scale auxiliary training. Furthermore, an excessive pursuit of generic perceptual metrics often results in low fidelity. To address these issues, we present Dynamic Trajectory Initialization (DTI) paradigm for GFVSR, which reformulates GFVSR as an input-driven directional restoration. With a novel enhancement-and-injection conditioning mechanism for pretrained DiT backbone, fidelity of our model has been significantly improved without compromising perceptual quality. To dynamically set the starting sampling point, we propose a Discriminative Guide (DG) trained via objective Signal-to-Noise Ratio (SNR) alignment. With only minor model adaptation and fine-tuning, our method achieves a SOTA overall performance across diverse metrics and benchmarks. An analysis of relationship between actual comprehensive quality and common metrics is also conducted, which demonstrates the perception-distortion trade-off and that the LPIPS is the most convincing metric in our case.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes Dynamic Trajectory Initialization (DTI) for Generative Face Video Super-Resolution (GFVSR). It reformulates GFVSR as input-driven directional restoration of a pretrained DiT backbone via an enhancement-and-injection conditioning mechanism to improve fidelity without sacrificing perceptual quality. A lightweight Discriminative Guide (DG) is introduced, trained through objective Signal-to-Noise Ratio (SNR) alignment, to dynamically select the starting sampling point. The central claim is that only minor model adaptation and fine-tuning suffice to reach SOTA performance across metrics and benchmarks; an auxiliary analysis of the perception-distortion trade-off is also presented, concluding that LPIPS is the most reliable metric in this setting.

Significance. If the empirical claims hold, the work would demonstrate a practical route to harness strong generative priors for GFVSR with low additional training cost, potentially lowering inference expense while mitigating the fidelity loss common in pure generative approaches. The metric-analysis component could also inform evaluation practices in the broader perceptual restoration literature.

major comments (2)
  1. [Abstract] Abstract: the claim that the method 'achieves a SOTA overall performance across diverse metrics and benchmarks' is asserted without any quantitative tables, baseline comparisons, ablation results, or error bars. Because the central contribution is an empirical performance improvement, this absence prevents verification of the claim from the supplied text.
  2. [Abstract] Abstract (DG description): the statement that the Discriminative Guide is 'trained via objective Signal-to-Noise Ratio (SNR) alignment' is given at a high level with no equation, loss formulation, or training protocol. This detail is load-bearing for the 'minor adaptation' claim, yet no concrete implementation is visible.
minor comments (2)
  1. [Abstract] Abstract: 'a SOTA' is grammatically awkward; standard usage is 'SOTA' or 'state-of-the-art'.
  2. [Abstract] Abstract: the final sentence on the metric analysis refers to 'our case' without defining the evaluation protocol or dataset, reducing clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed feedback on the abstract. We address the two major comments point by point below, noting that the abstract is a concise summary while the full manuscript contains the supporting details.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the method 'achieves a SOTA overall performance across diverse metrics and benchmarks' is asserted without any quantitative tables, baseline comparisons, ablation results, or error bars. Because the central contribution is an empirical performance improvement, this absence prevents verification of the claim from the supplied text.

    Authors: Abstracts are subject to strict length limits and conventionally omit tables or detailed quantitative results; the SOTA claim is substantiated by the full manuscript, which includes baseline comparisons (Table 1), ablation studies (Section 4.3), and benchmark results with metrics across multiple datasets (Section 5). Error bars are reported where statistical significance is assessed. The supplied text for this review appears to have been limited to the abstract, but the empirical evidence is present in the complete paper. revision: no

  2. Referee: [Abstract] Abstract (DG description): the statement that the Discriminative Guide is 'trained via objective Signal-to-Noise Ratio (SNR) alignment' is given at a high level with no equation, loss formulation, or training protocol. This detail is load-bearing for the 'minor adaptation' claim, yet no concrete implementation is visible.

    Authors: The abstract provides only a high-level overview due to space constraints. The full manuscript details the SNR alignment objective, including the loss formulation (Equation 4) and training protocol for the lightweight Discriminative Guide (Section 3.2), which is trained independently to enable the claimed minor adaptation of the pretrained DiT backbone. This separation keeps the core model changes minimal while supporting dynamic initialization. revision: partial

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The abstract and available description present a method proposal (DTI paradigm, DG module trained on SNR alignment) without any equations, derivations, or first-principles claims. No load-bearing steps reduce predictions or results to inputs by construction, self-citation chains, or fitted parameters renamed as outputs. The SOTA claim is framed as an empirical outcome after minor adaptation, which is externally falsifiable on benchmarks and does not rely on internal self-definition. This is the expected outcome for a methods paper lacking visible mathematical derivation chains.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are described.

pith-pipeline@v0.9.1-grok · 5740 in / 997 out tokens · 32613 ms · 2026-06-30T07:48:42.174583+00:00 · methodology

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

54 extracted references · 12 canonical work pages · 8 internal anchors

  1. [1]

    In: International Conference on Learning Representations (ICLR) (2026)

    Bai,H.,Chen,X.,Yang,C.,He,Z.,Deng,S.,Chen,Y.:Vivid-vr:Distillingconcepts from text-to-video diffusion transformer for photorealistic video restoration. In: International Conference on Learning Representations (ICLR) (2026)

    2026

  2. [2]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Blau, Y., Michaeli, T.: The perception-distortion tradeoff. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 6228-6237, 2018 (2018)

    2018

  3. [3]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2021)

    Chan, K.C., Wang, X., Xu, X., Gu, J., Loy, C.C.: Glean: Generative latent bank for large-factor image super-resolution. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2021)

    2021

  4. [4]

    arXiv preprint arXiv:2104.13371 (2021)

    Chan, K.C., Zhou, S., Xu, X., Loy, C.C.: Basicvsr++: Improving video su- per resolution with enhanced propagation and alignment". arXiv preprint arXiv:2104.13371 (2021)

    work page arXiv 2021

  5. [5]

    In: IEEE Conference on Computer Vision and Pattern Recogni- tion (2022)

    Chan, K.C., Zhou, S., Xu, X., Loy, C.C.: Investigating tradeoffs in real-world video super-resolution. In: IEEE Conference on Computer Vision and Pattern Recogni- tion (2022)

    2022

  6. [6]

    In: IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) (2021)

    Chen, C., Li, X., Lingbo, Y., Lin, X., Zhang, L., Wong, K.Y.K.: Progressive semantic-aware style transformation for blind face restoration. In: IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) (2021)

    2021

  7. [7]

    Chen, X., Tan, J., Wang, T., Zhang, K., Luo, W., Cao, X.: Towards real-world blind face restoration with generative diffusion prior (2023)

    2023

  8. [8]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2018)

    Chen, Y., Tai, Y., Liu, X., Shen, C., Yang, J.: Fsrnet: End-to-end learning face super-resolution with facial priors. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2018)

    2018

  9. [9]

    In: INTERSPEECH (2018)

    Chung, J.S., Nagrani, A., Zisserman, A.: Voxceleb2: Deep speaker recognition. In: INTERSPEECH (2018)

    2018

  10. [10]

    In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Deng, J., Guo, J., Xue, N., Zafeiriou, S.: Arcface: Additive angular margin loss for deep face recognition. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4685–4694 (2019)

    2019

  11. [11]

    Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., Podell, D., Dockhorn, T., English, Z., Lacey, K., Goodwin, A., Marek, Y., Rombach, R.: Scaling rectified flow transformers for high-resolution image synthesis. arXiv preprint arXiv:2403.03206 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  12. [12]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Fang, Y., Chen, Y., Yin, S., Hu, Q., Yao, J., Zhang, Y., Zhang, X., Wang, Y.: One-step diffusion transformer for controllable real-world image super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 23440–23450 (2026)

    2026

  13. [13]

    In: European Conference on Computer Vision (ECCV) (2024)

    Feng, R., Li, C., Loy, C.C.: Kalman-inspired feature propagation for video face super-resolution. In: European Conference on Computer Vision (ECCV) (2024)

    2024

  14. [14]

    In: CVPR (2020)

    Gu, J., Shen, Y., Zhou, B.: Image processing using multi-code gan prior. In: CVPR (2020)

    2020

  15. [15]

    In: ECCV (2022)

    Gu, Y., Wang, X., Xie, L., Dong, C., Li, G., Shan, Y., Cheng, M.M.: Vqfr: Blind face restoration with vector-quantized dictionary and parallel decoder. In: ECCV (2022)

    2022

  16. [16]

    Ho,J.,Jain,A.,Abbeel,P.:Denoisingdiffusionprobabilisticmodels.arXivpreprint arxiv:2006.11239 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  17. [17]

    CVPR (2018) 16 Y

    Jo, Y., Oh, S.W., Kang, J., Kim, S.J.: Deep video super-resolution network using dy namic upsampling filters without explicit motion compensa tion. CVPR (2018) 16 Y. Tang et al

    2018

  18. [18]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Ke, J., Wang, Q., Wang, Y., Milanfar, P., Yang, F.: Musiq: Multi-scale image quality transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 5148–5157 (2021)

    2021

  19. [19]

    HunyuanVideo: A Systematic Framework For Large Video Generative Models

    Kong, W., Tian, Q., Zhang, Z., Min, R., Dai, Z., Zhou, J., Xiong, J., Li, X., Wu, B., Zhang, J., et al.: Hunyuanvideo: A systematic framework for large video generative models. arXiv preprint arXiv:2412.03603 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  20. [20]

    In: ECCV (2020)

    Li, X., Chen, C., Zhou, S., Lin, X., Zuo, W., Zhang, L.: Blind face restoration via deep multi-scale component dictionaries. In: ECCV (2020)

    2020

  21. [21]

    In: CVPR (2020)

    Li, X., Li, W., Ren, D., Zhang, H., Wang, M., Zuo, W.: Enhanced blind face restoration with multi-exemplar images and adaptive spatial feature fusion. In: CVPR (2020)

    2020

  22. [22]

    In: The European Conference on Computer Vision (ECCV) (September 2018)

    Li, X., Liu, M., Ye, Y., Zuo, W., Lin, L., Yang, R.: Learning warped guidance for blind face restoration. In: The European Conference on Computer Vision (ECCV) (September 2018)

    2018

  23. [23]

    Flow Matching for Generative Modeling

    Lipman, Y., Chen, R.T.Q., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  24. [24]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  25. [25]

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H.V., Szafraniec, M., Khalidov, V., Fernandez,P.,Haziza,D.,Massa,F.,El-Nouby,A.,Howes,R.,Huang,P.Y.,Xu,H., Sharma, V., Li, S.W., Galuba, W., Rabbat, M., Assran, M., Ballas, N., Synnaeve, G., Misra, I., Jegou, H., Mairal, J., Labatut, P., Joulin, A., Bojanowski, P.: Dinov2: Learning robust visual features without ...

    2023

  26. [26]

    In: European Conference on Computer Vision (ECCV) (2020)

    Pan, X., Zhan, X., Dai, B., Lin, D., Loy, C.C., Luo, P.: Exploiting deep generative prior for versatile image restoration and manipulation. In: European Conference on Computer Vision (ECCV) (2020)

    2020

  27. [27]

    Scalable Diffusion Models with Transformers

    Peebles, W., Xie, S.: Scalable diffusion models with transformers. arXiv preprint arXiv:2212.09748 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  28. [28]

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning transferable visual models from natural language supervision (2021)

    2021

  29. [29]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models (2021)

    2021

  30. [30]

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khalidov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., Massa, F., Haziza, D., Wehrstedt, L., Wang, J., Darcet, T., Moutakanni, T., Sentana, L., Roberts, C., Vedaldi, A., Tolan, J., Brandt, J., Couprie, C., Mairal, J., Jégou, H., Labatut, P., Bojanowski, P.: DINOv3 (2025)

    2025

  31. [31]

    In: Interna- tional Conference on Learning Representations (2021)

    Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score- based generative modeling through stochastic differential equations. In: Interna- tional Conference on Learning Representations (2021)

    2021

  32. [32]

    In: CVPR (2019)

    Tero Karras, Samuli Laine, T.A.: A style-based generator architecture for genera- tive adversarial networks. In: CVPR (2019)

    2019

  33. [33]

    In: 2021 Smart Technologies, Communication and Robotics (STCR)

    Varma, K., Reddy, G.S., Subramanyam, N.: Face image super resolution using a generative adversarial network. In: 2021 Smart Technologies, Communication and Robotics (STCR). pp. 1–8 (2021)

    2021

  34. [34]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Wan, T., Wang, A., Ai, B., Wen, B., Mao, C., Xie, C.W., Chen, D., Yu, F., Zhao, H., Yang, J., Zeng, J., Wang, J., Zhang, J., Zhou, J., Wang, J., Chen, J., Zhu, K., Zhao, K., Yan, K., Huang, L., Feng, M., Zhang, N., Li, P., Wu, P., Chu, R., Feng, R., Zhang, S., Sun, S., Fang, T., Wang, T., Gui, T., Weng, T., Shen, T., Lin, W., DTI 17 Wang, W., Wang, W., Zh...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    In: AAAI (2023)

    Wang, J., Chan, K.C., Loy, C.C.: Exploring clip for assessing the look and feel of images. In: AAAI (2023)

    2023

  36. [36]

    In: ICLR (2026)

    Wang, J., Lin, S., Lin, Z., Ren, Y., Wei, M., Yue, Z., Zhou, S., Chen, H., Zhao, Y., Yang, C., Xiao, X., Loy, C.C., Jiang, L.: Seedvr2: One-step video restoration via diffusion adversarial post-training. In: ICLR (2026)

    2026

  37. [37]

    Wang, J., Lin, Z., Wei, M., Zhao, Y., Yang, C., Loy, C.C., Jiang, L.: Seedvr: Seeding infinityindiffusiontransformertowardsgenericvideorestoration.In:CVPR(2025)

    2025

  38. [38]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops (2019)

    Wang, X., Chan, K.C., Yu, K., Dong, C., Loy, C.C.: Edvr: Video restoration with enhanced deformable convolutional networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops (2019)

    2019

  39. [39]

    In: International Conference on Computer Vision Workshops (ICCVW) (2021)

    Wang,X.,Xie,L.,Dong,C.,Shan,Y.:Real-esrgan:Trainingreal-worldblindsuper- resolution with pure synthetic data. In: International Conference on Computer Vision Workshops (ICCVW) (2021)

    2021

  40. [40]

    In: The IEEE International Conference on Computer Vision (ICCV) (October 2019)

    Wang, X., Bo, L., Fuxin, L.: Adaptive wing loss for robust face alignment via heatmap regression. In: The IEEE International Conference on Computer Vision (ICCV) (October 2019)

    2019

  41. [41]

    arXiv preprint arXiv:2303.06885 (2023)

    Wang, Z., Zhang, X., Zhang, Z., Zheng, H., Zhou, M., Zhang, Y., Wang, Y.: Dr2: Diffusion-based robust degradation remover for blind face restoration. arXiv preprint arXiv:2303.06885 (2023)

    work page arXiv 2023

  42. [42]

    Wang, Z., Chen, X., Xu, C., Zhu, J., Hu, X., Zhang, J., Wang, C., Liu, Y., Zhou, Y., Ji, R.: Svfr: A unified framework for generalized video face restoration (2025)

    2025

  43. [43]

    Wu, Z., Sun, Z., Zhou, T., Fu, B., Cong, J., Dong, Y., Zhang, H., Tang, X., Chen, M., Wei, X.: Omgsr: You only need one mid-timestep guidance for real-world image super-resolution (2025)

    2025

  44. [44]

    In: The IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (2022)

    Xie, L., Wang, X., Zhang, H., Dong, C., Shan, Y.: Vfhq: A high-quality dataset and benchmark for video face super-resolution. In: The IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (2022)

    2022

  45. [45]

    IJCAI 2024 (2024)

    Xu, K., Xu, L., He, G., Yu, W., Li, Y.: Beyond alignment: Blind video face restora- tion via parsing-guided temporal-coherent transformer. IJCAI 2024 (2024)

    2024

  46. [46]

    In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    Yang, T., Ren, P., Xie, X., Zhang, L.: Gan prior embedded network for blind face restoration in the wild. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    2021

  47. [47]

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

    Yang, Z., Teng, J., Zheng, W., Ding, M., Huang, S., Xu, J., Yang, Y., Hong, W., Zhang, X., Feng, G., et al.: Cogvideox: Text-to-video diffusion models with an expert transformer. arXiv preprint arXiv:2408.06072 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  48. [48]

    In: Proceedings of the European Conference on Computer Vision (ECCV) (September 2018)

    Yu, X., Fernando, B., Ghanem, B., Porikli, F., Hartley, R.: Face super-resolution guided by facial component heatmaps. In: Proceedings of the European Conference on Computer Vision (ECCV) (September 2018)

    2018

  49. [49]

    In: CVPR (2018)

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: CVPR (2018)

    2018

  50. [50]

    2025 IEEE International Conference on Multimedia and Expo (ICME) pp

    Zhang, Z., Gao, X., Wang, Z., Hu, Q., Zhang, X.: Td-bfr: Truncated diffusion model for efficient blind face restoration. 2025 IEEE International Conference on Multimedia and Expo (ICME) pp. 1–6 (2025),https://api.semanticscholar. org/CorpusID:277322774 18 Y. Tang et al

    2025

  51. [51]

    arXiv:2507.19138 (2025)

    Zhao1, W., Zhou, J., Zhu, X., Chen, W., Zhang, X.Y., Lei, Z., Wang, F.: Realisvsr: Detail-enhanced diffusion for real-world 4k video super-resolution. arXiv:2507.19138 (2025)

    work page arXiv 2025

  52. [52]

    In: NeurIPS (2022)

    Zhou, S., Chan, K.C., Li, C., Loy, C.C.: Towards robust blind face restoration with codebook lookup transformer. In: NeurIPS (2022)

    2022

  53. [53]

    In: ECCV (2022)

    Zhu, H., Wu, W., Zhu, W., Jiang, L., Tang, S., Zhang, L., Liu, Z., Loy, C.C.: CelebV-HQ: A large-scale video facial attributes dataset. In: ECCV (2022)

    2022

  54. [54]

    Flashvsr: Towards real-time diffusion-based streaming video super-resolution,

    Zhuang, J., Guo, S., Cai, X., Li, X., Liu, Y., Yuan, C., Xue, T.: Flashvsr: To- wards real-time diffusion-based streaming video super-resolution. arXiv preprint arXiv:2510.12747 (2025)

    work page arXiv 2025