Pith. sign in

REVIEW 6 major objections 8 minor 1 cited by

Distribution-aware Dataset Distillation for Efficient Image Restoration

T0 review · 6 major / 8 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read TripleD claims that training image restoration models on a 1-5% complexity-selected subset preserves about 90-95% of full-data quality and fits 4K training on one consumer GPU in under eight hours.

desk verdict A plausible distillation pipeline for image restoration that is undone by the absence of a random-subset baseline; the central selection mechanism is unverified. read the letter →

arxiv 2504.14826 v1 pith:VRRT3OTO submitted 2025-04-21 cs.CV

Zhuoran Zheng , Xin Su , Chen Wu , Xiuyi Jia This is my paper · ORCID
classification cs.CV
keywords datasetdistillationimagerestorationsubsetselectionvisiontransformerentropy-basedcomplexitylatentdiffusiondistributionmatchingefficienttraining
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

This paper tries to establish that dataset distillation, previously applied almost entirely to classification, works for image restoration, a dense pixel-prediction task with no classification margin to absorb errors. It proposes a pipeline named TripleD that scores image complexity with a pretrained vision Transformer, keeps the top 1-2% of samples, synthesizes extra compact training pairs in latent space with a diffusion model, and fine-tunes the small dataset's feature distribution with an 8-layer CNN. On deraining, deblurring, denoising, all-in-one restoration, and 4K low-light benchmarks, training on this tiny subset reaches about 90-95% of full-dataset PSNR/SSIM and fits on one consumer GPU in under eight hours. The practical reason to care is that, if correct, researchers can iterate on large restoration models without large GPU clusters, and the framework points a way to distill other dense prediction datasets.

What carries the argument

The load-bearing object is the complexity-selected, distribution-aligned mini-dataset produced by three cooperating components. A ViT-16 with a sigmoid head, trained on ImageNet with manually revised entropy labels, assigns each downsampled image a complexity score in [0,1], and the top 1-2% form the real sub-dataset. A diffusion model generates synthetic degraded/clean pairs whose latent codes match the selected real samples, replacing the GAN used in prior latent distillation. An 8-layer CNN over 3x3 convolutions then adjusts the feature distribution of the combined small set, supervised by L2 and KL terms plus a gradient-matching loss, while the restoration model trains. The two-stage curriculum, easy low-complexity samples first and harder ones later, organizes this small dataset into a training schedule.

What would settle it

Train the same restoration network on a randomly chosen 2% subset, run it through the same CNN fine-tuning and diffusion synthesis, and compare PSNR/SSIM on Rain100L and GoPro; if the random subset matches or beats the complexity-selected one, the entropy scorer is not doing the work the paper assigns it.

Watch

Extended reading notes

Core claim

The central claim is that a 2% subset of training images, chosen by a ViT-based complexity score and adjusted in feature space by a CNN, is enough to train restoration networks to near full-data performance: about 90-95% in PSNR/SSIM across ten benchmarks including Rain100L deraining, GoPro deblurring, SIDD denoising, and the UHD-LOL4K low-light set. The paper also reports that this holds for all-in-one models that must handle several degradations at once, and that the 2% operating point is near the knee of the curve, since raising the subset to 5% or 10% yields only small gains. The mechanism is not simple subsampling: the distilled subset is augmented by diffusion-synthesized pairs and passed through an 8-layer CNN trained jointly with the restoration network, so that the feature distribution of the small set is pulled toward the original dataset.

Load-bearing premise

The method's gains depend on the assumption that a vision Transformer's entropy-style complexity score really identifies the most training-worthy images, but no random 2% subset is tested as a baseline, so the selection mechanism could be contributing little beyond aggressive downsampling.

Editorial extensions

If this is right

  • A single mid-range GPU can train restoration transformers that currently require multi-GPU setups, because the 2% subset keeps memory and training steps low.
  • UHD (4K) low-light restoration models can be trained in under eight hours instead of days, making dataset-scale iteration practical for small labs.
  • The same distilled-subset recipe works across degradation types such as rain, blur, noise, haze, and low light, so users do not need a separate distillation scheme for each task.
  • Because gains saturate between 2% and 10% subset size, the 2% point is close to the optimal trade-off between fidelity and training cost.
  • Models trained on the distilled subset still transfer to downstream object detection and segmentation, so reduced training data does not lock out downstream users.

Reading between the lines

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

  • The authors leave implicit that the entropy scorer's contribution is untested against random sampling; a natural next experiment is a random 2% subset passed through the same CNN and diffusion pipeline, which would separate the selection mechanism from the distribution-alignment machinery.
  • The manual revision of entropy labels makes the scorer closer to a learned perceptual-complexity measure than to true information-theoretic entropy, which would explain why cosine distance, standard deviation, and KL divergence performed poorly as complexity scores.
  • Because distribution matching happens in latent space rather than task-specific pixels, the recipe could transfer to other dense regression problems such as super-resolution, inpainting, or depth estimation.
  • The easy-to-hard curriculum is a testable hypothesis the ablations do not isolate: with the same 2% subset, shuffling the order would show whether the ordering itself contributes or only the selection matters.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 8 minor

Summary. The paper proposes TripleD, a distribution-aware dataset distillation framework for image restoration. TripleD selects a small subset (1-5%) of training pairs using a ViT-16 that predicts image-complexity scores from manually revised ImageNet entropy labels, then fine-tunes the selected subset's feature distribution with an 8-layer CNN and also synthesizes latent-space samples with a diffusion model (SD3). The distilled data are used to train restoration backbones (Restormer, PromptIR, UHDFormer) on a single RTX 3090 GPU. Experiments cover multi-task restoration (deraining, deblurring, denoising), all-in-one restoration, and UHD low-light enhancement, reporting that 2% of the data preserves 90-95% of full-training PSNR/SSIM.

Significance. The problem addressed - reducing the training cost of large image-restoration models - is important, and the paper demonstrates a concrete resource-saving result: training UHDFormer on a single consumer GPU. The experimental scope is broad, with 11 tables and ablations of CNN depth, downsampling resolution, feature extractors, and subset proportion. These are genuine strengths. However, the paper's central attribution claim is not yet supported: no random-subsample baseline is reported, so the results may reflect simple data downsampling rather than the proposed complexity-aware selection. The reliance on a manually annotated scorer and the inconsistent loss definitions also prevent reproducibility. The contribution is potentially significant, but the evidence presented is insufficient as it stands.

major comments (6)
  1. [Section 4.2, Tables 1, 9-11] The central claim that the proposed selection mechanism enables 90-95% of full-dataset performance with 1-5% of data is never tested against a random or uniform subsample. Table 1 compares TripleD only with multi-GPU and single-GPU full-data training; Tables 9-11 compare models trained with TripleD with pre-trained or retrained baselines. The ablations in Tables 2-8 vary components of the pipeline but always operate on the ViT-selected subset, so they do not isolate the selection step. Please add a random-2%-subset baseline (using the same training budget, optimizer, and gradient-accumulation settings) and a 'first-2%' or uniform-sampling baseline for each task, and report PSNR/SSIM with standard deviations over multiple seeds.
  2. [Section 3.2-3.4, Figure 2] The paper does not specify the composition and size of the final training set. Section 3.2 selects a subset B of size 1-2%; Sections 3.3-3.4 generate a synthetic dataset Bs; Figure 2 states that 'both the distilled real and synthetic subsets are used to train the restoration network.' The abstract and Figure 1 claim 'only 2% of the data,' but if the model trains on B union Bs, the effective training data may be larger than 2%. Please report the exact number of training pairs used in each experiment and clarify how the 1-5% claim is computed.
  3. [Section 3.4, Eq. (4)] The loss used for the CNN fine-tuning step is not consistently defined. Eq. (3) defines LDC as a gradient-matching cosine-similarity loss, but the text after Eq. (4) says 'LDC denotes KL dispersion.' The notation IFa and IFb is also undefined. Please reconcile these definitions, specify the exact objective and its weights, and describe how the CNN is trained jointly with the restoration model.
  4. [Section 1 and Section 3.2] The complexity scorer requires manual revision of ImageNet entropy labels ('we manually conduct a significant amount of revision annotation', Section 1), but the paper does not describe what the revision consists of, how many labels were revised, or whether the revised labels are released. This is a load-bearing component because the selection mechanism is the core contribution; without this information the method is not fully specified or reproducible. Please provide the annotation protocol and make the revised labels available.
  5. [Section 4.6, Table 4] The feature-extractor ablation in Table 4 shows that ResNet-50, Mamba, and MLP-Mixer all achieve PSNR within 1.58 dB of ViT (30.50-32.08 dB), suggesting that the specific choice of complexity scorer has a minor effect. Combined with the absence of a random baseline, this leaves open the possibility that the gains come from the reduced training set size and the CNN fine-tuning rather than from the entropy-based selection. Please discuss this and, if possible, compare against a no-selection (random) control.
  6. [Section 3.3 and Figure 5] Figure 5 introduces a comparison with 'GSDD' in the diversity analysis, but GSDD is never defined or cited, and no quantitative GSDD results appear in the tables. In addition, Table 7 compares SD3 against StyleGAN2 but not against the selected real subset without synthetic data, so the contribution of the latent distillation module is not established. Please define GSDD and include the missing no-synthetic baseline.
minor comments (8)
  1. [Abstract] The abstract says '500 savings in computing resources'; this should be written as '500x' or '500% savings' to be meaningful.
  2. [Table 3] In Table 3, the row labeled 'Ours' should be labeled '128x128', which is the downsampling resolution used by TripleD; the current label is not a resolution.
  3. [Table 9 caption] Table 9's caption says 'pre-trained PromptIR model without TripleD' while the text says 'retrained'; please align the wording with the actual protocol.
  4. [Figure 8] Figure 8 lists per-image PSNR values (e.g., 'Ours 43.80') that differ from the dataset-average PSNR in Table 10 (35.01); please state explicitly that the figure shows per-sample PSNR, not the benchmark average.
  5. [Section 3.1, Eq. (2)] Eq. (2) uses class-based IPC notation (M = C x IPC) that is not meaningful for restoration datasets without classes; please define how the subset size is determined for regression-style restoration tasks.
  6. [Section 4.1 vs. Section 3.2] Section 4.1 says 'dynamically selecting 2% of the data in each epoch,' but Section 3.2 describes a one-time static selection of the top 1-2% by entropy; please clarify whether the selection changes across training epochs.
  7. [Reference [10]] The text refers to YOLOv5 in Section 4.7, but reference [10] is YOLOX; please correct the citation or the reference.
  8. [Table 1] Table 1 lists identical PSNR/SSIM values for RealBlur-J and RealBlur-R under each method; please verify whether this is a copy-paste error.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TripleD is an empirical distillation method validated against external benchmarks; the missing random-subset baseline is a validation gap, not a circular derivation.

full rationale

TripleD is an empirical dataset-distillation method rather than a formal derivation, and its central claim is tested against external reference points (official multi-GPU Restormer, PromptIR, UHDFormer) instead of being derived from its own definitions. The ViT-based complexity scorer is trained on ImageNet with manually revised entropy labels, independent of the downstream restoration PSNR/SSIM, so the selected subset is not defined in terms of the reported performance. The CNN distribution-matching module in Section 3.4 is co-optimized with the restoration model's loss, which is a standard differentiable data-selection/fitting loop rather than a logical circularity, because the final evaluation is on held-out external test sets (Rain100L, GoPro, SIDD, DND, UHD-LOL4K, UHD-LL, etc.). The absence of a random-2%-subset baseline is a genuine empirical weakness: it prevents attributing the gains specifically to the complexity-based selection mechanism. However, that is a question of experimental control, not a case where the claimed prediction reduces by construction to the fitted input. The cited prior work on latent distillation (GLaD) is an external method with independent validation, and no load-bearing self-citation or imported uniqueness theorem appears in the argument. Therefore no circular step is identified.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim depends on several unproven modeling choices: the validity of entropy as a complexity proxy, the effectiveness of feature-space distribution matching, and the usefulness of diffusion-synthesized data. There are no new postulated physical or mathematical entities.

free parameters (3)
  • Subset proportion p = 2% (tuned among 1%, 2%, 5%, 10%)
    The paper chooses p=2% as a trade-off between efficiency and accuracy; this value directly determines the reported resource savings.
  • Downsampling resolution = 128x128
    The ViT complexity score uses images downsampled to 128x128; ablation shows resolution affects performance, so the chosen value is a tuned design choice.
  • CNN depth = 8 layers
    Ablation in Table 2 shows 8 layers is the best choice, indicating this is a hand-selected architecture component.
assumptions (3)
  • domain assumption Image complexity, as scored by a ViT trained on ImageNet with manual entropy labels, identifies the most valuable training samples for restoration.
    The entire selection mechanism depends on this correlation, which is never validated independently.
  • domain assumption Feature-space distribution matching (L2 + KL) between the selected subset and the full dataset is an effective surrogate for preserving training value.
    The CNN fine-tuning relies on this assumption, but no analysis links feature alignment to restoration quality.
  • domain assumption Synthetic samples generated by SD3 and then fine-tuned can replace real degraded-clean pairs.
    The method assumes the diffusion model produces useful training data, but the synthesis procedure is not described in enough detail to assess.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distribution-aware Dataset Distillation for Efficient Image Restoration." pith.science (2026). https://pith.science/paper/VRRT3OTO

@misc{pith2026250414826,
  author       = {Pith},
  title        = {Pith review of: Distribution-aware Dataset Distillation for Efficient Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VRRT3OTO}},
  note         = {Machine review of arXiv:2504.14826}
}
read the original abstract

With the exponential increase in image data, training an image restoration model is laborious. Dataset distillation is a potential solution to this problem, yet current distillation techniques are a blank canvas in the field of image restoration. To fill this gap, we propose the Distribution-aware Dataset Distillation method (TripleD), a new framework that extends the principles of dataset distillation to image restoration. Specifically, TripleD uses a pre-trained vision Transformer to extract features from images for complexity evaluation, and the subset (the number of samples is much smaller than the original training set) is selected based on complexity. The selected subset is then fed through a lightweight CNN that fine-tunes the image distribution to align with the distribution of the original dataset at the feature level. To efficiently condense knowledge, the training is divided into two stages. Early stages focus on simpler, low-complexity samples to build foundational knowledge, while later stages select more complex and uncertain samples as the model matures. Our method achieves promising performance on multiple image restoration tasks, including multi-task image restoration, all-in-one image restoration, and ultra-high-definition image restoration tasks. Note that we can train a state-of-the-art image restoration model on an ultra-high-definition (4K resolution) dataset using only one consumer-grade GPU in less than 8 hours (500 savings in computing resources and immeasurable training time).

Figures

Figures reproduced from arXiv: 2504.14826 by the authors.

Figure 1
Figure 1. Performance and analysis of our TripleD method for image restoration. First, the figure on the left shows that using only 2% of the original dataset allows the model to perform close to training on the full set. The figure on the right demonstrates an ablation where the performance of the image restoration model rises steadily as the proportion of synthetic datasets rises, but the improvement is very limited. expens… view at source ↗
Figure 2
Figure 2. Overview of the TripleD Pipeline. A pretrained IRNet backbone (SwinIR) first processes large-scale real–degraded pairs (Xr → Yr), and a ViT-based entropy scorer selects the top 1-2% most informative samples as the real sub-dataset. Meanwhile, a diffusion￾based latent distillation module (SD3) synthesizes compact training pairs whose latents match those of the selected real samples. Finally, both the distilled real a… view at source ↗
Figure 3
Figure 3. This figure shows the complexity of the image is evalu [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Visual comparisons with state-of-the-art methods on the GoPro, Rain100 and DPDD datasets. All comparison methods are [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: CDFs of pairwise distances between distilled-sample [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison of rain removal results for two [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: All-in-one model with TripleD. All comparison methods are retrained on the PromptIR synthesised dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Comparison of methods on UHD-LOL4K dataset. All comparison methods are retrained on the UHDFormer synthesised dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Comparison of methods on UHD-LL dataset. All comparison methods are retrained on the UHDFormer synthesised dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: The qualitative results of object detection experi [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: The results of segmentation experiments on the real [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Distribution alignment visualization. (a) KDE plot [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Evolution of Dataset Distillation: Toward Scalable and Generalizable Solutions

    cs.CV 2025-02 conditional novelty 3.0 of 10

    A 2023-2025 survey of dataset distillation that organizes matching, generative, decoupling, and selective methods and tabulates ImageNet-scale accuracy comparisons.

Reference graph

Works this paper leans on

44 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [8]

    A study in dataset distillation for image super-resolution.arXiv preprint arXiv:2502.03656, 2025

    Tobias Dietz, Brian B Moser, Tobias Nauen, Federico Raue, Stanislav Frolov, and Andreas Dengel. A study in dataset distillation for image super-resolution.arXiv preprint arXiv:2502.03656, 2025. 2

    work page arXiv 2025

  2. [1]

    A high-quality denoising dataset for smartphone cameras

    Abdelrahman Abdelhamed, Stephen Lin, and Michael S Brown. A high-quality denoising dataset for smartphone cameras. In CVPR, 2018. 4, 5

    work page 2018

  3. [2]

    Defocus deblur- ring using dual-pixel data

    Abdullah Abuolaim and Michael S Brown. Defocus deblur- ring using dual-pixel data. In ECCV, 2020. 4, 5

    work page 2020

  4. [3]

    Scail: Classifier weights scaling for class incremental learning

    Eden Belouadah and Adrian Popescu. Scail: Classifier weights scaling for class incremental learning. In WACV, pages 1266–1275, 2020. 2

    work page 2020

  5. [4]

    Dataset distillation by matching training trajectories

    George Cazenavette, Tongzhou Wang, Mason Morehead, Jonathan Park, Kuno Singh, Dilip Krishnan, Deva Ramanan, and Eli Shechtman. Dataset distillation by matching training trajectories. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7027–7036, 2022. 2

    work page 2022

  6. [5]

    Generalizing dataset dis- tillation via deep generative prior

    George Cazenavette, Tongzhou Wang, Antonio Torralba, Alexei A Efros, and Jun-Yan Zhu. Generalizing dataset dis- tillation via deep generative prior. In CVPR, pages 3739– 3748, 2023. 4

    work page 2023

  7. [6]

    Pre- trained image processing transformer

    Hanting Chen, Yunhe Wang, Tianlong Guo, et al. Pre- trained image processing transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12299–12310, 2021. 2

    work page 2021

  8. [7]

    Trainable nonlinear reac- tion diffusion for image restoration

    Jun Chen and Thomas Pock. Trainable nonlinear reac- tion diffusion for image restoration. IEEE Transactions on Pattern Analysis and Machine Intelligence , 39:1172–1185,

    work page

Show all 44 references
  1. [9]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (ICLR), 2021. 2

    2021

  2. [10]

    Yolox: Exceeding yolo series in 2021

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430, 2021. 8

    2021 arXiv

  3. [11]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016. 6

    2016

  4. [12]

    Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation. In Proc. IEEE CVPR, pages 9924–9935, 2022. 8

    2022

  5. [13]

    Scope of validity of psnr in image/video quality assessment

    Quan Huynh Huynh-Thu and Mohammad Ghanbari. Scope of validity of psnr in image/video quality assessment. Elec- tronics letters, 44(13):800–801, 2008. 4

    2008

  6. [14]

    Deblurgan: Blind motion deblur- ring using conditional adversarial networks

    Orest Kupyn, Viktor Budzan, Taras Mykhailiuk, Dmitry Mishkin, and Ji ˇr´ı Matas. Deblurgan: Blind motion deblur- ring using conditional adversarial networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 8183–8192, 2018. 1

    2018

  7. [15]

    Embedding fourier for ultra-high-definition low-light image 9 enhancement

    Chongyi Li, Chun-Le Guo, Man Zhou, Zhexin Liang, Shangchen Zhou, Ruicheng Feng, and Chen Change Loy. Embedding fourier for ultra-high-definition low-light image 9 enhancement. In International Conference on Learning Rep- resentations (ICLR), 2023. 8

    2023

  8. [16]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. In Proceedings of the IEEE Interna- tional Conference on Computer Vision (ICCV), pages 1833– 1844, 2021. 2

    2021

  9. [17]

    Low-level vi- sion and dynamic sampling for high-fidelity image restora- tion

    Han Liu, Xiaoyang Guo, and Wei Wang. Low-level vi- sion and dynamic sampling for high-fidelity image restora- tion. IEEE Transactions on Image Processing , 30:1575– 1586, 2021. 2

    2021

  10. [18]

    Sgdr: Stochas- tic gradient descent with warm restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochas- tic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983, 2016. 2

    2016 arXiv

  11. [19]

    Deep multi-scale convolutional neural network for dynamic scene deblurring

    Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. In CVPR, 2017. 2, 4, 5

    2017

  12. [20]

    Ramit: Reciprocal atten- tion mixing transformer for lightweight image restoration

    Dai Nguyen, Minfeng Li, et al. Ramit: Reciprocal atten- tion mixing transformer for lightweight image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  13. [21]

    Dataset distillation with infinite ensembles

    Tan M Nguyen, Thang D Bui, Richard E Turner, and Yarin Gal. Dataset distillation with infinite ensembles. InAdvances in Neural Information Processing Systems (NeurIPS), 2022. 2

    2022

  14. [22]

    Continual lifelong learning with neural networks: A review

    German I Parisi, Ronald Kemker, Jose L Part, Christopher Kanan, and Stefan Wermter. Continual lifelong learning with neural networks: A review. Neural Networks, 113:54–71,

  15. [23]

    Freqformer: Frequency- aware transformer for lightweight image super-resolution

    Taeho Park and Sunghyun Kim. Freqformer: Frequency- aware transformer for lightweight image super-resolution. In Proceedings of the International Joint Conference on Artifi- cial Intelligence (IJCAI), 2024. 2

    2024

  16. [24]

    Benchmarking denoising al- gorithms with real photographs

    Tobias Ploetz and Stefan Roth. Benchmarking denoising al- gorithms with real photographs. In CVPR, 2017. 4, 5

    2017

  17. [25]

    Promptir: Prompting for all-in-one image restoration

    Vaishnav Potlapalli, Syed Waqas Zamir, Salman Khan, and Fahad Khan. Promptir: Prompting for all-in-one image restoration. In Thirty-seventh Conference on Neural Infor- mation Processing Systems, 2023. 8

    2023

  18. [26]

    icarl: Incremental classi- fier and representation learning

    Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, and Christoph H Lampert. icarl: Incremental classi- fier and representation learning. InCVPR, pages 2001–2010,

    2001

  19. [27]

    Real- blur: A new dataset for realistic blur synthesis and deblur- ring

    Sunghyun Rim, Hyungseok Son, and Seungyong Lee. Real- blur: A new dataset for realistic blur synthesis and deblur- ring. In CVPR, 2020. 4, 5

    2020

  20. [28]

    Active learning for convo- lutional neural networks: A core-set approach

    Ozan Sener and Silvio Savarese. Active learning for convo- lutional neural networks: A core-set approach. International Conference on Learning Representations (ICLR), 2018. 2

    2018

  21. [29]

    Active learning literature survey

    Burr Settles. Active learning literature survey. University of Wisconsin, Madison, 52(55-66):11, 2009. 2

    2009

  22. [30]

    Active learning literature survey

    Burr Settles. Active learning literature survey. University of Wisconsin, Madison, WI, 52(55-66), p.11, 2010. 2

    2010

  23. [31]

    Adaptive data se- lection strategies for efficient deep learning

    Jiajun Wang, Yiqiao Tang, and Feng Li. Adaptive data se- lection strategies for efficient deep learning. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2385–2394, 2022. 2

    2022

  24. [32]

    Dataset distillation via optimal transport

    Rui Wang, Ankit Goyal, Mohit Bansal, Mar´ıa Rojas-Carulla, Jakub M Tomczak, and Adrian Weller. Dataset distillation via optimal transport. In Proceedings of the 38th Inter- national Conference on Machine Learning (ICML) , pages 11535–11545, 2021. 1, 2

    2021

  25. [33]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 4

    2004

  26. [34]

    Dynamic dataset distillation with uncertainty estimation

    Chen Yang, Jing Liu, Jie Xu, and Lei Wang. Dynamic dataset distillation with uncertainty estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  27. [35]

    Deep joint rain detection and removal from a single image

    Wenhan Yang, Robby T Tan, Jiashi Feng, Zhuwen Liu, Zhid- ing Guo, and Shuicheng Yan. Deep joint rain detection and removal from a single image. In CVPR, 2017. 4, 5

    2017

  28. [36]

    Im- age restoration with deep learning: A review.IEEE Transac- tions on Neural Networks and Learning Systems , 32:1967– 1985, 2020

    Shaohua Yin, Xiaoyang Zhang, Jian Wu, and Jun Chen. Im- age restoration with deep learning: A review.IEEE Transac- tions on Neural Networks and Learning Systems , 32:1967– 1985, 2020. 1

    1967

  29. [37]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 572...

    2022

  30. [38]

    Density-aware single image de-raining using a multi-stream dense network

    He Zhang and Vishal M Patel. Density-aware single image de-raining using a multi-stream dense network. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 695–704, 2018. 2

    2018

  31. [39]

    Image de-raining using a con- ditional generative adversarial network

    He Zhang and Vishal M Patel. Image de-raining using a con- ditional generative adversarial network. In IEEE Transac- tions on Circuits and Systems for Video Technology , 2018. 5

    2018

  32. [40]

    Beyond a gaussian denoiser: Residual learn- ing of deep cnn for image denoising

    Kai Zhang, Wangmeng Zuo, Yunjin Chen, Dongwei Meng, and Lei Zhang. Beyond a gaussian denoiser: Residual learn- ing of deep cnn for image denoising. In IEEE Transactions on Image Processing, pages 3142–3155, 2017. 2

    2017

  33. [41]

    Im- age deraining with feature attention

    Kun Zhang, Wang Zuo, Yandong Chen, and Lei Zhang. Im- age deraining with feature attention. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 50–58, 2019. 1

    2019

  34. [42]

    Im- age restoration: A comprehensive review

    Lei Zhang, Wang Zuo, Yandong Chen, and Kun Zhang. Im- age restoration: A comprehensive review. In IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2021. 1

    2021

  35. [43]

    Dataset condensation with gradi- ent matching

    Bo Zhao and Hakan Bilen. Dataset condensation with gradi- ent matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 708–717, 2021. 2

    2021

  36. [44]

    Dataset distillation: A comprehensive review

    Xin Zhao, Liu Yang, and Hao Zhou. Dataset distillation: A comprehensive review. arXiv preprint arXiv:2301.04866,

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.