pith. sign in

arxiv: 2607.02131 · v1 · pith:5MJGONJNnew · submitted 2026-07-02 · 💻 cs.CV · cs.LG

AbsoluteDegradation: A Physics-Inspired Synthetic Film-Degradation Pipeline and Archival Film Restoration Benchmark

Miko{\l}aj Jastrz\k{e}bski , Dawid Glinkowski , Dawid Zieli\'nski , Daniel Borkowski , Wojciech Koz{\l}owski , Kamil Adamczewski This is my paper

Pith reviewed 2026-07-03 15:30 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords archival film restorationsynthetic degradation pipelinefilm artifactsrestoration benchmarkphysics-inspired synthesisgeneralizationtemporal coherencesignal-dependent grain
0
0 comments X

The pith

A physics-inspired modular pipeline for film degradations lets restoration models generalize better to real archival footage.

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

The paper presents AbsoluteDegradation, a pipeline that builds synthetic film degradations by composing separate artifact families such as signal-dependent grain, parametric scratches, and temporally coherent motion. It pairs this generator with a new benchmark of 81,576 real high-resolution archival frames for evaluation. Existing methods rely on synthetic data that does not match real degradation patterns and on small or inaccessible test sets, which limits both training and comparison. Models trained on the new pipeline show improved performance on actual deteriorated footage, and the benchmark makes systematic weaknesses in current architectures visible. The work supplies a single framework for generating training data and measuring progress under realistic conditions.

Core claim

AbsoluteDegradation models the analog-to-digital process as a structured composition of artifact families that includes signal-dependent grain, parametric scratches, and temporally coherent camera motion. This composition supports controlled generation of diverse degradation regimes. When used for training, the resulting models generalize better to real-world archival footage. The accompanying benchmark of 81,576 curated high-resolution frames from real deteriorated film enables consistent evaluation and exposes failure modes that prior small or inaccessible datasets could not reveal.

What carries the argument

The AbsoluteDegradation pipeline, a modular composition of artifact families that models the analog-to-digital transition and generates controlled, temporally coherent degradations.

If this is right

  • Models trained with the pipeline outperform prior synthetic-data approaches on real archival footage.
  • The benchmark dataset supports standardized, reproducible comparison across restoration methods.
  • Systematic failure modes of existing architectures become measurable under controlled real-world conditions.
  • Controlled variation of individual artifact families allows targeted study of which degradations are hardest to restore.
  • A unified training-and-evaluation framework reduces reliance on scarce paired real data.

Where Pith is reading between the lines

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

  • The modular structure could be adapted to simulate degradations in other imaging domains that involve similar physical processes, such as old photographs or medical scans.
  • Parameter sweeps over the artifact families might allow matching the pipeline to the specific characteristics of particular film stocks or eras.
  • Adding explicit chemical or mechanical simulation steps for each artifact could further close the remaining gap to real footage.
  • The benchmark could serve as a fixed test set for comparing restoration methods across independent research groups without data-access issues.

Load-bearing premise

The modular composition of separate artifact families accurately reproduces the complex, temporally coherent degradations present in real archival film.

What would settle it

An experiment in which models trained on AbsoluteDegradation show no improvement over models trained on earlier synthetic degradations when both are tested on the 81,576-frame real archival benchmark.

Figures

Figures reproduced from arXiv: 2607.02131 by Daniel Borkowski, Dawid Glinkowski, Dawid Zieli\'nski, Kamil Adamczewski, Miko{\l}aj Jastrz\k{e}bski, Wojciech Koz{\l}owski.

Figure 1
Figure 1. Figure 1: Pipeline for dataset creation. Each clean clip [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative and quantitative comparison of the proposed AbsoluteDegradation and Bringing [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative frames from (a) SRWOV [27] and (b) our archival benchmark. (c) Quantita￾tive comparison of datasets. Unlike SRWOV, our dataset consists of high-resolution, lossless frames of genuine archival footage (rather than mixed content such as cartoons), with no overlay watermarks and substantially richer physical analog degradations, making it more domain-representative. Our benchmark exhibits highe… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative analysis of restored images using our pipeline vs. Bringing Old Films [ [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Degradation pipeline ablations on real dataset with RTN [45]. Ablation studies. To measure individual pipeline contributions, we evaluate RTN on six ablated variants of AbsoluteDegradation ( [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Tier comparison on REDS frame 000 (single fixed crop): RGB ground truth (left) and [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: AbsoluteDegradation operator atlas. Reference crop (top-left), ten single-operator panels, and a chained all degradations preview (bottom-right) D.2.1 Full scene-split dataset. The primary dataset contains every curated single-scene clip, totalling 81,576 frames. Each clip depicts a single continuous shot with no internal cuts, making it directly suitable for evaluating temporal restoration methods that as… view at source ↗
Figure 8
Figure 8. Figure 8: Visual comparison of our pipeline with Bringing Old Films Pipeline [ [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual comparison against Bringing Old Films Pipeline [ [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visual comparison of our pipeline with Bringing Old Films Pipeline [ [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
read the original abstract

Restoring archival film remains a fundamentally challenging problem due to the absence of paired training data and the lack of standardized evaluation benchmarks. Pristine versions of deteriorated footage are physically unrecoverable, requiring supervised methods to rely on synthetic data that often fail to capture the complex, temporally coherent nature of real film degradation. At the same time, existing real-world datasets are limited in scale, quality, and accessibility, hindering reliable evaluation and fair comparison across methods. We address both limitations with AbsoluteDegradation, a physics-inspired, modular pipeline for synthesizing realistic film degradations, and a new large-scale archival benchmark. The proposed pipeline models the analog-to-digital process as a structured composition of artifact families, incorporating signal-dependent grain, parametric scratches, and temporally coherent camera motion, enabling controlled generation of diverse degradation regimes. In parallel, we introduce a curated dataset of 81,576 high-resolution frames sourced from real archival footage, designed for consistent evaluation under real-world conditions. Together, these contributions provide a unified framework for training and benchmarking restoration models. Extensive experiments across multiple architectures show that models trained with AbsoluteDegradation generalize better to real-world footage, while the proposed benchmark reveals systematic failure modes of current methods. We hope this work establishes a foundation for reproducible and domain-authentic evaluation in archival film restoration.

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 manuscript introduces AbsoluteDegradation, a physics-inspired modular pipeline for synthesizing film degradations via composition of artifact families (signal-dependent grain, parametric scratches, temporally coherent camera motion) and a new benchmark of 81,576 high-resolution real archival frames. Experiments across architectures claim that training on the synthetic data yields improved generalization to real footage while the benchmark exposes systematic failure modes of existing methods.

Significance. If the synthetic pipeline's modular outputs are shown to match the joint statistics and temporal correlations of real archival degradations, the work would address the core paired-data scarcity problem and supply a large-scale, accessible benchmark for reproducible evaluation in archival film restoration. The controlled, physics-inspired generation of diverse regimes is a clear strength that could support systematic ablation studies.

major comments (2)
  1. [Abstract] Abstract: The central claim that models trained with AbsoluteDegradation generalize better to real-world footage is load-bearing on the unverified assumption that the modular artifact families reproduce the complex, temporally coherent degradations of real film; no quantitative distribution matching, statistical tests, or perceptual validation against real data is referenced.
  2. [Experiments] Experiments section: The reported generalization improvements lack controls (e.g., comparison against simpler synthetic baselines or ablations isolating higher-order interactions such as grain-scratch coupling under motion) that would isolate the contribution of pipeline fidelity from other factors such as data volume.
minor comments (2)
  1. [Abstract] The abstract would benefit from naming the specific architectures and loss functions used in the experiments to allow immediate assessment of the scope of the generalization results.
  2. Consider adding a supplementary table or figure that tabulates key pipeline parameters and their physical motivation for each artifact family.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments correctly identify gaps in quantitative validation of the synthetic pipeline and in experimental controls. We address each point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that models trained with AbsoluteDegradation generalize better to real-world footage is load-bearing on the unverified assumption that the modular artifact families reproduce the complex, temporally coherent degradations of real film; no quantitative distribution matching, statistical tests, or perceptual validation against real data is referenced.

    Authors: We acknowledge that the manuscript does not reference quantitative distribution matching, statistical tests, or perceptual validation of the synthetic degradations against real data. Validation in the current version rests on the physics-inspired modular design and qualitative visual results. We will add Fréchet Inception Distance (FID) comparisons between synthetic and real degraded frame distributions, along with statistical tests on artifact statistics and a small-scale perceptual study, to the revised manuscript. revision: yes

  2. Referee: [Experiments] Experiments section: The reported generalization improvements lack controls (e.g., comparison against simpler synthetic baselines or ablations isolating higher-order interactions such as grain-scratch coupling under motion) that would isolate the contribution of pipeline fidelity from other factors such as data volume.

    Authors: We agree that the reported gains would be more convincingly attributed to pipeline fidelity with additional controls. The current experiments compare AbsoluteDegradation-trained models against existing methods but omit direct baselines using simpler non-modular synthetics or ablations on coupled artifacts. In revision we will include (i) a simpler synthetic baseline with independent artifact addition and (ii) targeted ablations on grain-scratch-motion interactions, while controlling for training data volume. revision: yes

Circularity Check

0 steps flagged

No circularity: pipeline and benchmark are independently constructed

full rationale

The paper defines AbsoluteDegradation as an explicit modular composition of signal-dependent grain, parametric scratches, and temporally coherent motion, then evaluates generalization on a separate curated set of 81,576 real archival frames. No equation or claim reduces a prediction to a fitted parameter by construction, no self-citation is invoked as a uniqueness theorem, and the benchmark is external real footage rather than synthetic outputs. The derivation chain therefore remains self-contained against external data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The pipeline relies on modeling choices for artifact families whose specific parameterizations and validation against real statistics are not detailed in the abstract.

pith-pipeline@v0.9.1-grok · 5800 in / 918 out tokens · 18336 ms · 2026-07-03T15:30:14.898093+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

64 extracted references · 23 canonical work pages

  1. [1]

    Bruni and D

    V . Bruni and D. Vitulano. A generalized model for scratch detection.IEEE Transactions on Image Processing, 13(1):44–50, 2004. doi: 10.1109/tip.2003.817231

    work page doi:10.1109/tip.2003.817231 2004

  2. [2]

    K. C. Chan, X. Wang, K. Yu, C. Dong, and C. C. Loy. BasicVSR: The search for essential components in video super-resolution and beyond. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4947–4956, 2021

    2021

  3. [3]

    K. C. Chan, S. Zhou, X. Xu, and C. C. Loy. BasicVSR++: Improving video super-resolution with enhanced propagation and alignment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5972–5981, 2022

    2022

  4. [4]

    Chang, Y .-L

    R.-C. Chang, Y .-L. Sie, S.-M. Chou, and T. K. Shih. Photo defect detection for image inpainting. InProceedings of the IEEE International Symposium on Multimedia (ISM), pages 403–407,

  5. [5]

    doi: 10.1109/ISM.2005.91

    work page doi:10.1109/ism.2005.91 2005

  6. [6]

    Claus and J

    M. Claus and J. van Gemert. ViDeNN: Deep blind video denoising. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 1843–1852, 2019

    2019

  7. [7]

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, and Y . Wei. Deformable convolutional networks. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 764–773, 2017

    2017

  8. [8]

    DeTone, T

    D. DeTone, T. Malisiewicz, and A. Rabinovich. Superpoint: Self-supervised interest point detection and description. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2018

    2018

  9. [9]

    IEEE Transactions on Image Processing , year =

    M. Elad and M. Aharon. Image denoising via sparse and redundant representations over learned dictionaries.IEEE Transactions on Image Processing, 15(12):3736–3745, 2006. doi: 10.1109/TIP.2006.881969

    work page doi:10.1109/tip.2006.881969 2006

  10. [10]

    Giakoumis, N

    I. Giakoumis, N. Nikolaidis, and I. Pitas. Digital image processing techniques for the detection and removal of cracks in digitized paintings.IEEE Transactions on Image Processing, 15(1): 178–188, 2006. doi: 10.1109/TIP.2005.860311

    work page doi:10.1109/tip.2005.860311 2006

  11. [11]

    H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, and S.-T. Xia. MambaIR: A simple baseline for image restoration with state-space model. InEuropean Conference on Computer Vision (ECCV), pages 222–241, 2024. doi: 10.1007/978-3-031-72649-1_13

    work page doi:10.1007/978-3-031-72649-1_13 2024

  12. [12]

    H. Guo, Y . Guo, Y . Zha, Y . Zhang, W. Li, T. Dai, S.-T. Xia, and Y . Li. MambaIRv2: Attentive state space restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 28124–28133, 2025

    2025

  13. [13]

    Haris, G

    M. Haris, G. Shakhnarovich, and N. Ukita. Recurrent back-projection network for video super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  14. [14]

    Heusel, H

    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, 30, 2017

    2017

  15. [15]

    Huang, X

    Z. Huang, X. Shi, C. Zhang, Q. Wang, K. C. Cheung, H. Qin, J. Dai, and H. Li. FlowFormer: A transformer architecture for optical flow. InEuropean Conference on Computer Vision (ECCV), pages 668–685, 2022. doi: 10.1007/978-3-031-19790-1_40

    work page doi:10.1007/978-3-031-19790-1_40 2022

  16. [16]

    Iizuka and E

    S. Iizuka and E. Simo-Serra. DeepRemaster: Temporal source-reference attention networks for comprehensive video enhancement.ACM Transactions on Graphics (TOG), 38(6):176:1– 176:13, 2019. doi: 10.1145/3355089.3356570

    work page doi:10.1145/3355089.3356570 2019

  17. [17]

    Ivanova, J

    D. Ivanova, J. Williamson, and P. Henderson. Simulating analogue film damage to analyse and improve artefact restoration on high-resolution scans.Computer Graphics Forum, 42(2): 133–148, 2023. 11

    2023

  18. [18]

    Johnson, A

    J. Johnson, A. Alahi, and L. Fei-Fei. Perceptual losses for real-time style transfer and super- resolution. InEuropean Conference on Computer Vision (ECCV), pages 694–711, 2016

    2016

  19. [19]

    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 (ICCV), pages 5148–5157, 2021. doi: 10.1109/ICCV48922.2021.00510

    work page doi:10.1109/iccv48922.2021.00510 2021

  20. [20]

    D. Kim, S. Woo, J.-Y . Lee, and I. S. Kweon. Deep video inpainting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5792–5801, 2019

    2019

  21. [21]

    Lefkimmiatis

    S. Lefkimmiatis. Universal denoising networks: A novel CNN architecture for image denoising. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3204–3213, 2018

    2018

  22. [22]

    Liang, J

    J. Liang, J. Cao, G. Sun, K. Zhang, L. Van Gool, and R. Timofte. SwinIR: Image restoration using swin transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 1833–1844, 2021

    2021

  23. [23]

    Liang, Y

    J. Liang, Y . Fan, X. Xiang, R. Ranjan, E. Ilg, S. Green, J. Cao, K. Zhang, R. Timofte, and L. Van Gool. Recurrent video restoration transformer with guided deformable attention. In Advances in Neural Information Processing Systems (NeurIPS), volume 35, 2022

    2022

  24. [24]

    Liang, J

    J. Liang, J. Cao, Y . Fan, K. Zhang, R. Ranjan, Y . Li, R. Timofte, and L. Van Gool. VRT: A video restoration transformer.IEEE Transactions on Image Processing, 33:2171–2182, 2024. doi: 10.1109/TIP.2024.3372454

    work page doi:10.1109/tip.2024.3372454 2024

  25. [25]

    Lin and E

    S. Lin and E. Simo-Serra. Restoring degraded old films with recursive recurrent transformer networks. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 6718–6728, 2024

    2024

  26. [26]

    Lindenberger, P.-E

    P. Lindenberger, P.-E. Sarlin, and M. Pollefeys. Lightglue: Local feature matching at light speed. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 17627–17638, October 2023

    2023

  27. [27]

    Q. Liu, Y . Liu, L. Wang, F. Yan, Q. Zhang, and H. Ju. Restoration of archival film with large areas of structural damage.npj Heritage Science, 14:272, 2026. doi: 10.1038/s40494-025-02235-3

    work page doi:10.1038/s40494-025-02235-3 2026

  28. [28]

    Y . Mao, H. Luo, Z. Zhong, P. Chen, Z. Zhang, and S. Wang. Making old film great again: Degradation-aware state space model for old film restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 28039–28049, 2025. doi: 10.1109/CVPR52734.2025.02611

    work page doi:10.1109/cvpr52734.2025.02611 2025

  29. [29]

    Mittal, A

    A. Mittal, A. K. Moorthy, and A. C. Bovik. No-reference image quality assessment in the spatial domain.IEEE Transactions on Image Processing, 21(12):4695–4708, 2012. doi: 10.1109/TIP.2012.2214050

    work page doi:10.1109/tip.2012.2214050 2012

  30. [30]

    S. Nah, S. Baik, S. Hong, G. Moon, S. Son, R. Timofte, and K. M. Lee. NTIRE 2019 challenge on video deblurring and super-resolution: Dataset and study. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 1996– 2005, 2019

    2019

  31. [31]

    B. T. Oh, S.-m. Lei, and C.-C. J. Kuo. Advanced film grain noise extraction and synthesis for high-definition video coding.IEEE Transactions on Circuits and Systems for Video Technology, 19(12):1717–1728, 2009. doi: 10.1109/TCSVT.2009.2026974

    work page doi:10.1109/tcsvt.2009.2026974 2009

  32. [32]

    Ranjan and M

    A. Ranjan and M. J. Black. Optical flow estimation using a spatial pyramid network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4161– 4170, 2017

    2017

  33. [33]

    Salmona, L

    A. Salmona, L. Bouza, and J. Delon. DeOldify: A review and implementation of an automatic colorization method.Image Processing On Line (IPOL), 12:347–368, 2022. 12

    2022

  34. [34]

    M. Seitzer. pytorch-fid: FID Score for PyTorch. https://github.com/mseitzer/ pytorch-fid, August 2020. Version 0.3.0

    2020

  35. [35]

    Stanco, G

    F. Stanco, G. Ramponi, and A. De Polo. Towards the automated restoration of old photographic prints: A survey. InIEEE Region 8 EUROCON 2003: Computer as a Tool, volume B, pages 370–374, 2003. doi: 10.1109/EURCON.2003.1248221

    work page doi:10.1109/eurcon.2003.1248221 2003

  36. [36]

    Stephenson and A

    I. Stephenson and A. Saunders. Simulating film grain using the noise-power spectrum. In Theory and Practice of Computer Graphics (TPCG). Eurographics Association, 2007

    2007

  37. [37]

    S. Su, M. Delbracio, J. Wang, G. Sapiro, W. Heidrich, and O. Wang. Deep video deblurring for hand-held cameras. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1279–1288, 2017

    2017

  38. [38]

    Suganuma, X

    M. Suganuma, X. Liu, and T. Okatani. Attention-based adaptive selection of operations for image restoration in the presence of unknown combined distortions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  39. [39]

    J. Sun, W. Cao, Z. Xu, and J. Ponce. Learning a convolutional neural network for non-uniform motion blur removal. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 769–777, 2015. doi: 10.1109/CVPR.2015.7298677

    work page doi:10.1109/cvpr.2015.7298677 2015

  40. [40]

    Suvorov, E

    R. Suvorov, E. Logacheva, A. Mashikhin, A. Remizova, A. Ashukha, A. Silvestrov, N. Kong, H. Goka, K. Park, and V . Lempitsky. Resolution-robust large mask inpainting with Fourier convolutions. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 2149–2159, 2022

    2022

  41. [41]

    Szegedy, V

    C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the inception archi- tecture for computer vision. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016

    2016

  42. [42]

    Tassano, J

    M. Tassano, J. Delon, and T. Veit. FastDVDnet: Towards real-time deep video denoising without flow estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1354–1363, 2020

    2020

  43. [43]

    Teed and J

    Z. Teed and J. Deng. RAFT: Recurrent all-pairs field transforms for optical flow. InEuropean Conference on Computer Vision (ECCV), pages 402–419, 2020

    2020

  44. [44]

    R. A. Ulichney. Dithering with blue noise.Proceedings of the IEEE, 76(1):56–79, 1988. doi: 10.1109/5.3288

    work page doi:10.1109/5.3288 1988

  45. [45]

    Z. Wan, B. Zhang, D. Chen, P. Zhang, D. Chen, J. Liao, and F. Wen. Bringing old photos back to life. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2747–2757, 2020

    2020

  46. [46]

    Z. Wan, B. Zhang, D. Chen, and J. Liao. Bringing old films back to life. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17694–17703, 2022. doi: 10.1109/CVPR52688.2022.01717

    work page doi:10.1109/cvpr52688.2022.01717 2022

  47. [47]

    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, volume 37, pages 2555–2563,

  48. [48]

    doi: 10.1609/aaai.v37i2.25353

    work page doi:10.1609/aaai.v37i2.25353

  49. [49]

    X. Wang, K. C. Chan, K. Yu, C. Dong, and C. C. Loy. EDVR: Video restoration with enhanced deformable convolutional networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 1954–1963, 2019

    1954

  50. [50]

    X. Wang, L. Xie, C. Dong, and Y . Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InProceedings of the IEEE/CVF International Conference on Com- puter Vision Workshops (ICCVW), pages 1905–1914, 2021. doi: 10.1109/ICCVW54120.2021. 00217

    work page doi:10.1109/iccvw54120.2021 1905

  51. [51]

    Weiss and W

    Y . Weiss and W. T. Freeman. What makes a good model of natural images? InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2007. doi: 10.1109/CVPR.2007.383092. 13

    work page doi:10.1109/cvpr.2007.383092 2007

  52. [52]

    L. Xu, J. S. Ren, C. Liu, and J. Jia. Deep convolutional neural network for image deconvolution. InAdvances in Neural Information Processing Systems (NeurIPS), volume 27, 2014

    2014

  53. [53]

    R. Xu, X. Li, B. Zhou, and C. C. Loy. Deep flow-guided video inpainting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3723–3732, 2019

    2019

  54. [54]

    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 Workshops (CVPRW), pages 1191–1200, 2022. doi: 10.1109/CVPRW56347.2022.00126

    work page doi:10.1109/cvprw56347.2022.00126 2022

  55. [55]

    G. Youk, J. Oh, and M. Kim. FMA-Net: Flow-guided dynamic filtering and iterative feature refinement with multi-attention for joint video super-resolution and deblurring. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 44–55, 2024

    2024

  56. [56]

    J. Yu, Z. Lin, J. Yang, X. Shen, X. Lu, and T. S. Huang. Free-form image inpainting with gated convolution. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4471–4480, 2019

    2019

  57. [57]

    K. Yu, C. Dong, L. Lin, and C. C. Loy. Crafting a toolchain for image restoration by deep reinforcement learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2443–2452, 2018

    2018

  58. [58]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5728–5739, 2022

    2022

  59. [59]

    Zhang, Y

    H. Zhang, Y . Wu, and Z. Kuang. An efficient scratches detection and inpainting algorithm for old film restoration. In2009 International Conference on Information Technology and Computer Science (ITCS), volume 1, pages 75–78. IEEE, 2009

    2009

  60. [60]

    IEEE Transactions on Image Processing , year =

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang. Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising.IEEE Transactions on Image Processing, 26(7): 3142–3155, 2017. doi: 10.1109/TIP.2017.2662206

    work page doi:10.1109/tip.2017.2662206 2017

  61. [61]

    Zhang, W

    K. Zhang, W. Zuo, and L. Zhang. FFDNet: Toward a fast and flexible solution for CNN-based image denoising.IEEE Transactions on Image Processing, 27(9):4608–4622, 2018. doi: 10.1109/TIP.2018.2839891

    work page doi:10.1109/tip.2018.2839891 2018

  62. [62]

    Zhang, J

    K. Zhang, J. Liang, L. Van Gool, and R. Timofte. Designing a practical degradation model for deep blind image super-resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4791–4800, 2021

    2021

  63. [63]

    H. Zhao, L. Tian, X. Xiao, P. Hu, Y . Gou, and X. Peng. AverNet: All-in-one video restoration for time-varying unknown degradations. InAdvances in Neural Information Processing Systems (NeurIPS), volume 37, 2024

    2024

  64. [64]

    out of memory

    K. Zhou, W. Li, L. Lu, X. Han, and J. Lu. Revisiting temporal alignment for video restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6053–6062, 2022. 14 A Related Work A.1 Restoration Methods: From Images to Video Image restoration has evolved from model-based priors over natural image statistics...

    2022