Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Super-Resolution Generative Adversarial Networks based Video Enhancement

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that adding 3D non-local blocks to SRGAN and training on video patches rather than single frames yields video super-resolution models that outperform single-image methods on PSNR, SSIM, and LPIPS.

desk verdict Headline claim invalidated by uncontrolled SISR baselines and loss-evaluation circularity; the architecture is plausible but the evidence is not there yet. read the letter →

arxiv 2505.10589 v4 pith:T6X2JUJQ submitted 2025-05-14 cs.CV cs.AIeess.IV

Kağan Çeti̇n , Hacer Akça , Ömer Nezih Gerek This is my paper · ORCID
classification cs.CVcs.AIeess.IV
keywords videosuper-resolutionSRGANnon-localblocksspatio-temporalfeatureextractionpatch-basedtraininggenerativeadversarialnetworkstemporalcoherenceartificialintelligence
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 show that a video super-resolution model built by adding 3D Non-Local Blocks to the single-image SRGAN architecture, and trained on video patches rather than whole frames, outperforms frame-by-frame single-image super-resolution methods in both objective metrics and temporal consistency. The authors report two trained variants: an RRDB-based model with 27 million parameters and a residual-block model with 5 million parameters, evaluated on the BVI-AOM, BVI-HOMTex, and REDS video datasets. On the reported numbers, the larger model reaches PSNR up to 39.494, SSIM up to 0.973, and LPIPS as low as 0.140, with advantages claimed over SRCNN, EDSR, and Real-ESRGAN. The smaller model is reported to achieve higher SSIM on texture-rich or fast-moving datasets, suggesting a complexity-perception trade-off. A sympathetic reader would care because practical video upscaling for streaming, gaming, or archive restoration needs inter-frame coherence that single-image methods do not model.

What carries the argument

The load-bearing mechanism is the 3D Non-Local Block, a dot-product attention unit that computes a normalized weighted sum over all spatial and temporal positions of an input sequence and adds that as a residual. It is inserted into generators built from Residual-in-Residual Dense Blocks (27M parameters) or plain residual blocks (5M parameters), so the network can read correspondences between frames rather than processing them independently. Around this sit the training innovations: video frames are cropped into 16×16 or 32×32 patches, downsampled for 2× and cascaded 2×→4× tasks, processed with gradient accumulation and clipping, and degraded with a battery of augmentations such as Gaussian blur, Gaussian noise, JPEG compression, and content-aware blur. The discriminator is a U-Net, and the loss set mixes pixel losses, perceptual losses such as LPIPS, and edge-aware losses such as Laplacian, Sobel, Ricker, and gradient terms.

What would settle it

Retrain SRCNN, EDSR, and Real-ESRGAN on the same training splits and the same 2× bicubic and bilinear degradation used for the proposed models, then evaluate all models on the same test frames; if the PSNR gap shrinks from roughly ten decibels to a few tenths of a decibel, the central claim of superiority is not supported.

Watch

Extended reading notes

Core claim

The central claim is that extending SRGAN from two-dimensional single images to three-dimensional frame sequences lets the generator learn relationships across both space and time, and that this spatio-temporal modeling is what produces the reported gains. The paper states that the spatio-temporal RRDB-based model outperforms single-image super-resolution methods such as EDSR, Real-ESRGAN, and SRCNN on PSNR, SSIM, and LPIPS, and that it delivers improved temporal coherence, sharper textures, and fewer visual artifacts than traditional single-image methods. The reported numbers include a PSNR of 39.494, SSIM of 0.973, and LPIPS of 0.140 on BVI-AOM for 2× upscaling, roughly ten decibels above the experimentally obtained Real-ESRGAN result on the same dataset. The paper also reports a trade-off between the two variants: the 27M-parameter RRDB model wins on pixel-wise and edge-aware metrics, while the 5M-parameter residual model attains higher SSIM on BVI-HOMTex and REDS, which the authors attribute to reduced overfitting to complex textures and better temporal coherence in dynamic scenes.

Load-bearing premise

The load-bearing premise is that the SISR baselines were trained and evaluated under the same degradation, scale, and test conditions as the proposed models, but the paper's own comparison-table caption says the baseline numbers were obtained experimentally by rebuilding and manually training those models and may vary across sessions, which weakens that assumption.

Editorial extensions

If this is right

  • If the reported results hold, frame-by-frame single-image upscalers such as SRCNN, EDSR, and Real-ESRGAN are not the right tool for video, and the extra parameters spent on temporal modeling are justified.
  • The 5M-parameter model's higher SSIM on texture-rich and fast-moving datasets implies that model complexity is not monotone in perceptual quality, so smaller architectures can be the better choice for dynamic content.
  • Because the paper's pipeline separates patch-based multi-scale training from the non-local mechanism, the same training environment can be carried over to other video restoration tasks such as deblurring and denoising.
  • The cascaded 2× then 4× procedure is designed to extend to 8× upscaling, so the framework could serve larger enlargement factors without a new architecture.

Reading between the lines

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

  • Editorial inference: the reported roughly ten-decibel PSNR advantage over Real-ESRGAN on BVI-AOM is much larger than typical super-resolution gains, so a matched re-run with identically trained baselines is the experiment that would separate an architecture effect from a training-setup effect.
  • Editorial inference: if the gains replicate, ablating the non-local block would still be needed to know how much of the improvement comes from temporal attention versus the patch-based degradation-augmented training pipeline alone.
  • Editorial inference: the lighter model's SSIM advantage suggests that distilling the 27M model into a compact network, or pruning it, could preserve most of the perceptual benefit at lower inference cost.
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

4 major / 5 minor

Summary. The paper proposes a video super-resolution framework that extends SRGAN to spatio-temporal input by feeding 3D frame sequences through 3D Non-Local Blocks. A patch-based training pipeline downsamples sub-patches (16x16 or 32x32) by 2 and 4, applies several degradations, accumulates gradients, and uses a cascade for 4x upscaling. Two models are built, an RRDB-based model with 27M parameters and a Residual-based model with 5M parameters. The models are evaluated on BVI-AOM, BVI-HOMTex, and REDS under bicubic/bilinear downsampling. The headline claim, made in the abstract and in Section 4.3.1, is that the RRDB-based model substantially outperforms SISR baselines (EDSR, Real-ESRGAN, SRCNN) on PSNR, SSIM, and LPIPS and achieves better temporal coherence. The main comparison table (Table 3) reports baseline numbers obtained by 'training manually' with no experimental protocol, and several evaluation metrics overlap with training losses.

Significance. If the quantitative claims were established under a controlled protocol, the contribution would be a practical and lightweight alternative to frame-by-frame SISR for video, with two model variants spanning different capacity regimes. The architectural idea of using 3D non-local blocks inside an SRGAN for video is plausible, and the paper documents a rich set of degradations and loss functions. However, the experimental evidence is currently not credible because the baseline comparison is uncontrolled, the evaluation shares training objectives, and no temporal metric is reported. The significance of the paper is therefore conditional on fixing these issues.

major comments (4)
  1. [Section 4.3.1, Table 3] The central comparison is not controlled. The caption states that SISR results were 'obtained totally experimentally, by building the model again and training it manually' and that 'results may vary according to different sessions and projects.' No training set, degradation pipeline, loss weighting, optimizer, or checkpoint is specified for the SRCNN, EDSR, or Real-ESRGAN baselines. If the baselines were trained on different input resolutions or downsampling methods than the proposed model, the reported PSNR gap (e.g., 39.494 vs. 28.753 on BVI-AOM) is an artifact of the evaluation setup rather than an architectural property. The comparison must be rerun with official pretrained baselines applied to the same LR/HR test pairs, the same bicubic/bilinear degradation, and the same metric implementation, with the protocol fully described; otherwise the 'significantly outperforming' claim in the abstract and Section 4.4 is unsupported.
  2. [Section 3.1.7 and Table 2] The evaluation metrics overlap with the training objectives. LPIPS, Laplacian, Sobel, Ricker, and Gradient losses are all defined as loss functions used for training in Section 3.1.7, and LPIPS is reported in Table 2 while Laplacian and Gradient values are reported as evaluation metrics in Section 4.1. The proposed model is therefore measured in the metric space in which it was optimized, while the SISR baselines were not trained with those losses. This does not demonstrate a general perceptual advantage; it shows that the model optimizes its own loss. The evaluation should be limited to metrics not used in training, or the role of LPIPS and edge losses as training-loss trackers should be clearly separated from the independent evaluation.
  3. [Sections 4 and 4.3.2] The claim of improved temporal coherence is not supported by any temporal metric. All quantitative measurements are per-frame single-image metrics (PSNR, SSIM, LPIPS). The visual and textual claims about 'flickering' and 'motion consistency' in Section 4.3.2 are qualitative only. Without a quantitative temporal consistency measure (e.g., mean absolute frame difference, a temporal flicker metric, or optical-flow error between reconstructed frames), the abstract's assertion of 'improved temporal coherence' is unverifiable. The authors should add such a metric or explicitly restrict the claims to spatial quality.
  4. [Section 4 and Section 3.1] The experimental setup is under-specified. The test procedure is not reproducible: no train/test split is given, no number of test sequences or frames per dataset is reported, and no optimizer, learning-rate schedule, number of training iterations, hardware, or random seed is provided. Table 2 says 'average performance across the test sets' but the composition of the test set is not defined. As a consequence, none of the numerical comparisons can be assessed for statistical significance or variance. This is a load-bearing omission because the headline result depends on the validity of the numeric comparison.
minor comments (5)
  1. [Throughout] The manuscript contains frequent typos and grammatical errors, including 'Artifical Intelligence' in the keywords, 'seqeunces' and 'splitted' in Section 1.4, 'Gradietn loss' in Section 3.1.7, 'Gaussian blue' in Section 3.1.2, and 'compared totraditional' in the Abstract. A careful proofread is needed.
  2. [Equations (4), (6), (9)] Equation (4) uses n in the piecewise conditions although the function is of y; it should read y ≥ 0 and y < 0. The residual output formula in Eq. (6) does not follow from the concatenation equations that precede it, and the cross-references around Eq. (9) are inconsistent because Eq. (9) is called both the output of a non-local block and the unary function g, while the text says g is given in Eq. (12).
  3. [Figure 13 and Section 4.2] Figure 13 is invoked twice in Section 4.2, once for the RRDB-based model and once for the Residual-based model, yet its caption only mentions the Residual-based model. The figure labeling and the associated prose must be clarified.
  4. [Section 4.3] Section 4.3 says the comparison is against 'one of the most well-known' SISR models, Real-ESRGAN, but Table 3 includes SRCNN and EDSR with detailed numbers. The narrative should introduce all three baselines consistently.
  5. [References] The reference list has inconsistencies, such as reference [32] being dated 1981 while the text attributes neighbor embedding to 2004, and reference [33] lacking complete venue and year details. All references should be checked against the original sources.

Circularity Check

1 steps flagged · score 4.0 of 10

LPIPS and edge-aware losses are used both as training objectives and as reported evaluation metrics, so part of the perceptual-quality evidence is forced by construction; PSNR and SSIM provide independent grounding.

  1. fitted input called prediction [Section 3.1.7 (loss definitions) and Section 4.1 / Table 2 (evaluation)]
    "In this context,Learned Perceptual Image Patch Similarity (LPIPS)loss metric is defined to the system. It feeds the image batches into VGG19 pre-trained network (specifically, first36layers of it is used for this purpose), and then compares the results using eitherL1,L2, orCharbonnier loss. [...] The performance of both models was evaluated using Charbonnier Loss (RMSE with penalty), Learned Perceptual Image Patch Similarity (LPIPS), Structural Similarity Index (SSIM), and edge-aware metrics such as Laplacian and Gradient losses."

    The LPIPS functional described in Section 3.1.7 is first introduced as a training loss that 'punishes the generator according to similarity of recognizabilities,' and the same LPIPS score is then reported in Table 2 and Section 4.1 as evidence of superior perceptual quality. The same holds for Laplacian and Gradient losses, which are defined as 'Edge-Detection (Aware) Losses' for the generator and then reused as evaluation metrics. Because the model is explicitly optimized against these exact objectives, low LPIPS and edge-loss values are partially guaranteed by construction rather than being independent measurements. The PSNR and SSIM columns are not training objectives and therefore provide independent support, which keeps this as a partial, not total, circularity.

full rationale

No mathematical derivation in the paper reduces to its inputs: the model is a composition of known SRGAN, RRDB/residual, and non-local blocks, and the reported PSNR/SSIM numbers are not identities with any training loss. The only substantive circularity is evaluative: several metrics used to demonstrate improvement (LPIPS, Laplacian, Gradient, and Charbonnier) are literally the loss functions from Section 3.1.7, so those columns are not independent confirmations of the model's quality. The external SISR comparison in Table 3 is weakened by the caption's admission that the baseline results were 'obtained totally experimentally, by building the model again and training it manually' and 'may vary according to different sessions and projects'; that is a validity/reproducibility flaw rather than a circular-derivation flaw, so I do not count it as a circular step. There are no load-bearing self-citations or imported uniqueness theorems. An overall score of 4 reflects the metric-overlap in a central comparison while recognizing that the independent PSNR and SSIM columns keep the core claim from being fully reduced by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper is an empirical engineering report, so the ledger is dominated by hand-set hyperparameters and domain assumptions rather than mathematical axioms. The only formal background is the non-local block formulation from Wang et al. and standard GAN/backpropagation machinery. No invented entities are introduced; '3D non-local blocks' are taken from [5]. The main epistemic burden falls on the assumptions that synthetic degradation mimics real video, that per-frame metrics capture temporal quality, and that the baseline comparisons are fair.

free parameters (5)
  • Leaky ReLU negative slope = 0.1
    Set by hand in Section 1.1 and Eq. (4) for all residual blocks; affects gradient flow through the generator.
  • Ricker wavelet width = 0.55
    Hand-chosen in Section 3.1.7 for the custom Ricker edge loss kernel; determines which high-frequency scales are penalized.
  • Training patch sizes = 16x16 and 32x32
    Selected in Sections 3.1.4 and 3.1.5 for 2x and 4x upscaling; controls the local/global trade-off and memory usage.
  • Gradient clipping saturation value
    Mentioned in Section 3.2 as a constant to prevent gradient explosion, but the value is not reported; the clipped magnitude affects training stability.
  • Per-loss weights for the 8 losses
    Section 3.1.7 defines 8 content, perceptual, and edge losses but never gives the weighting coefficients used to combine them; the balance determines the reported metric trade-offs.
assumptions (5)
  • domain assumption 3D non-local blocks as defined by Wang et al. (2018) capture long-range spatio-temporal dependencies when applied to video tensors.
    Used throughout Section 2.1 as the core temporal mechanism; the paper cites [5] rather than deriving or ablating the operation.
  • domain assumption Synthetic degradations (bicubic/bilinear plus blur, noise, JPEG, etc.) approximate real-world low-resolution video conditions.
    Adopted in Section 3.1.2 to justify training data; the Discussion concedes models 'were trained on synthetic downsampling, which may not fully represent real-world low-resolution videos.'
  • domain assumption Per-frame metrics (PSNR, SSIM, LPIPS) and visual inspection are sufficient to substantiate temporal coherence.
    The paper claims improved temporal coherence in Sections 1 and 4 but reports no temporal metric; this assumption is load-bearing for the coherence claim.
  • domain assumption U-Net discriminator training converges to a useful non-trivial equilibrium that improves the generator.
    The discriminator design in Section 2.1 is justified by frequency-band and structural arguments, not by analysis or ablation.
  • standard math Standard stochastic gradient descent, backpropagation, and adversarial training behave as expected for this architecture.
    All training claims in Section 3.2 rely on the standard machinery without adaptation or formal guarantees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Super-Resolution Generative Adversarial Networks based Video Enhancement." pith.science (2026). https://pith.science/paper/T6X2JUJQ

@misc{pith2026250510589,
  author       = {Pith},
  title        = {Pith review of: Super-Resolution Generative Adversarial Networks based Video Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T6X2JUJQ}},
  note         = {Machine review of arXiv:2505.10589}
}
read the original abstract

This study introduces an enhanced approach to video super-resolution by extending ordinary Single-Image Super-Resolution (SISR) Super-Resolution Generative Adversarial Network (SRGAN) structure to handle spatio-temporal data. While SRGAN has proven effective for single-image enhancement, its design does not account for the temporal continuity required in video processing. To address this, a modified framework that incorporates 3D Non-Local Blocks is proposed, which is enabling the model to capture relationships across both spatial and temporal dimensions. An experimental training pipeline is developed, based on patch-wise learning and advanced data degradation techniques, to simulate real-world video conditions and learn from both local and global structures and details. This helps the model generalize better and maintain stability across varying video content while maintaining the general structure besides the pixel-wise correctness. Two model variants-one larger and one more lightweight-are presented to explore the trade-offs between performance and efficiency. The results demonstrate improved temporal coherence, sharper textures, and fewer visual artifacts compared to traditional single-image methods. This work contributes to the development of practical, learning-based solutions for video enhancement tasks, with potential applications in streaming, gaming, and digital restoration.

Figures

Figures reproduced from arXiv: 2505.10589 by the authors.

Figure 1
Figure 1. Basic overview of a GAN architecture [13] Fig. (2) shown below also reveals the steps and the approach that should be taken in a healthy SRGAN environment to be able to build a reliable generator with trustworthy results that are desirably similar to the ground truths in the end [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Basic overview of an SRGAN archi￾tecture [13] As it is already said, the general architecture of an SRGAN model is being discussed in "Super-Resolution Generative Adversarial Net￾works (SRGAN)", introduced by Ledig et al. in 2017 [1], for the first time, and revolutionized image enhancement technologies bu upscal￾ing low-resolution images into perceptually realistic high-resolution outputs. SRGAN’s generator-discrim… view at source ↗
Figure 5
Figure 5. a. Leaky ReLU compared to b. (or￾dinary) ReLU [18] product as the main operation) to capture inter￾frame relationships, moving beyond frame-by￾frame processing. Non-local blocks are simply units that computes weighted mean of all pix￾els in an image batch. Their generic formula is given below in Eq. (3) [19] yi = 1/C(x) X ∀j f (xi , xj)g(xj) (3) The spacetime view (or, diagram) of a non￾local block is illustrated be… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Residual Block by 1x1 upgrades can still be made on residual blocks by stacking them consecutively and let them share information among each other. Each output of those blocks are stored and shared directly with the any other residual block waiting on the flow way. The…
Figure 6
Figure 6. Figure 6: Spacetime View of Non-Local Block [5] It is shown that they are more accurate and better in terms of computational efficiency than basic convolutional 3D blocks. Additionally, speaking for both sequence length and (height, width) pair, formulation of non-local blocks a…
Figure 7
Figure 7. Figure 7: Architecture of RRDB based model (has 27M many of parameters) [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Architecture of Residual Block based model (has 5M many of parameters) [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: a. Residual Block and b. Residual in Residual Dense Block modules designed for this project. • i: Index of sequence (in time domain). • j: Index that enumerates all possible posi￾tions. • f (·): A pairwise function f computes a scalar (representing relationship such as…
Figure 10
Figure 10. Figure 10: U-Net architecture used as discriminator [ [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: a. Moore Neighborhood compared to b. Von Neumann Neighborhood [46] the objects or the shapes in the both im￾ages. • Edge-Detection (Aware) Losses: Edge detection losses are defined to evaluate the generator according to the precision of the edges of the upscaled image…
Figure 12
Figure 12. Figure 12: Evaluation of Laplacian Pyramid [49] •Gradient Loss: Gradietn loss mea￾sures the difference (e.g., L2) between first￾order gradients (e.g., computed via finite differences or Sobel filters) of two batches, focusing on edge and texture transitions. It is used to penali…
Figure 13
Figure 13. Figure 13: Some results obtained from the model built (Residual-based) from BVI-HOMTex dataset. [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: From left to right, low-resolution image, result obtained using Residual-based model, result obtained using Real-ESRGAN, and high-resolution image. Real-ESRGAN) and BVI-HOMTex (0.924 vs. 0.875) suggests better handling of motion and texture preservation, a critical ad…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 43 canonical work pages

  1. [1]

    Photo-realistic sin- gle image super-resolution using a gener- ative adversarial network

    Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunning- ham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic sin- gle image super-resolution using a gener- ative adversarial network. In2017 IEEE Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 105–114, 2017

    work page 2017

  2. [2]

    Xintao Wang, Kelvin C. K. Chan, Ke Yu, Chao Dong, and Chen Change Loy. Edvr: Video restoration with enhanced de- formable convolutional networks, 2019

    work page 2019

  3. [3]

    Enhanc- ing space-time video super-resolution via spatial-temporal feature interaction, 2024

    Zijie Yue and Miaojing Shi. Enhanc- ing space-time video super-resolution via spatial-temporal feature interaction, 2024

    work page 2024

  4. [4]

    Image super- resolution using deep convolutional net- works, 2015

    Chao Dong, Chen Change Loy, Kaim- ing He, and Xiaoou Tang. Image super- resolution using deep convolutional net- works, 2015

    work page 2015

  5. [5]

    Non-local neural networks, 2018

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks, 2018

    work page 2018

  6. [6]

    Deep residual learning for image recognition, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition, 2015

    2015

  7. [7]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need, 2023

    2023

  8. [8]

    Perceptual losses for real-time style transfer and super-resolution, 2016

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution, 2016

    2016

Show all 49 references
  1. [9]

    Gradient variance loss for structure- enhanced image super-resolution, 2022

    Lusine Abrahamyan, Anh Minh Truong, Wilfried Philips, and Nikos Deligian- nis. Gradient variance loss for structure- enhanced image super-resolution, 2022

    2022

  2. [10]

    Deep learning for sin- gle image super-resolution: A brief re- view.IEEE Transactions on Multimedia, 21(12):3106–3121, December 2019

    Wenming Yang, Xuechen Zhang, Yapeng Tian, Wei Wang, Jing-Hao Xue, and Qingmin Liao. Deep learning for sin- gle image super-resolution: A brief re- view.IEEE Transactions on Multimedia, 21(12):3106–3121, December 2019

    2019

  3. [11]

    Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio

    Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial net- works, 2014

    2014

  4. [12]

    Generative adversarial networks for image super-resolution: A survey, 2024

    Chunwei Tian, Xuanyu Zhang, Qi Zhu, Bob Zhang, and Jerry Chun-Wei Lin. Generative adversarial networks for image super-resolution: A survey, 2024

    2024

  5. [13]

    Generative adversarialnetworksforsyntheticdatagen- eration: A comparative study, 12 2021

    Claire Little, Mark Elliot, Richard All- mendinger, and Sahel Samani. Generative adversarialnetworksforsyntheticdatagen- eration: A comparative study, 12 2021

    2021

  6. [14]

    Laploss: Laplacianpyramid-basedmultiscalelossfor image translation, 2025

    Krish Didwania, Ishaan Gakhar, Prakhar Arya, and Sanskriti Labroo. Laploss: Laplacianpyramid-basedmultiscalelossfor image translation, 2025

    2025

  7. [15]

    Weinberger

    Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kilian Q. Weinberger. Densely connected convolutional networks, 2018

    2018

  8. [16]

    Esrgan: Enhanced super-resolution generative ad- versarial networks, 2018

    Xintao Wang, Ke Yu, Shixiang Wu, Jinjin Gu, Yihao Liu, Chao Dong, Chen Change Loy, Yu Qiao, and Xiaoou Tang. Esrgan: Enhanced super-resolution generative ad- versarial networks, 2018

    2018

  9. [17]

    Empirical evaluation of rectified ac- tivations in convolutional network, 2015

    Bing Xu, Naiyan Wang, Tianqi Chen, and Mu Li. Empirical evaluation of rectified ac- tivations in convolutional network, 2015

    2015

  10. [18]

    Residual network improves the prediction accuracy of genomic selection.Animal Ge- netics, 55:n/a–n/a, 05 2024

    Huaxuan Wu, Bingxi Gao, Rong Zhang, Zehang Huang, Zongjun Yin, Xiaoxiang Hu, Cai-Xia Yang, and Zhi-Qiang Du. Residual network improves the prediction accuracy of genomic selection.Animal Ge- netics, 55:n/a–n/a, 05 2024

    2024

  11. [19]

    Uni- fying nonlocal blocks for neural networks, 2021

    Lei Zhu, Qi She, Duo Li, Yanye Lu, Xuejing Kang, Jie Hu, and Changhu Wang. Uni- fying nonlocal blocks for neural networks, 2021

    2021

  12. [20]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep fea- tures as a perceptual metric, 2018

    2018

  13. [21]

    Understanding ssim, 2020

    Jim Nilsson and Tomas Akenine-Möller. Understanding ssim, 2020

    2020

  14. [22]

    Bvi- dvc: A training database for deep video compression.IEEE Transactions on Mul- timedia, 24:3847–3858, 2021

    Di Ma, Fan Zhang, and David R Bull. Bvi- dvc: A training database for deep video compression.IEEE Transactions on Mul- timedia, 24:3847–3858, 2021

    2021

  15. [23]

    Bvi-aom: A new training dataset for deep video compression optimization.arXiv preprint arXiv:2408.03265, 2024

    Jakub Nawała, Yuxuan Jiang, Fan Zhang, Xiaoqing Zhu, Joel Sole, and David Bull. Bvi-aom: A new training dataset for deep video compression optimization.arXiv preprint arXiv:2408.03265, 2024

    2024 arXiv

  16. [24]

    Deep learning tech- niques for super-resolution in video games, 2020

    Alexander Watson. Deep learning tech- niques for super-resolution in video games, 2020

    2020

  17. [25]

    Extrapolation, interpola- tion, and smoothing of stationary time se- ries with engineering applications, 1949

    Norbert Wiener. Extrapolation, interpola- tion, and smoothing of stationary time se- ries with engineering applications, 1949

    1949

  18. [26]

    C. E. Duchon. Lanczos filtering in one and two dimensions, 1979

    1979

  19. [27]

    R. G. Keys. Cubic convolution interpola- tion for digital image processing, 1981

    1981

  20. [28]

    Interpolation and sharpening forimageupsampling

    Xun Wang. Interpolation and sharpening forimageupsampling. In2022 2nd Interna- tional Conference on Computer Graphics, Image and Virtualization (ICCGIV), pages 73–77, 2022

    2022

  21. [29]

    Freeman, T.R

    W.T. Freeman, T.R. Jones, and E.C. Pasztor. Example-based super-resolution. IEEE Computer Graphics and Applica- tions, 22(2):56–65, 2002

    2002

  22. [30]

    L. R. Rabiner. A tutorial on hidden markov models and selected applications in speech recognition, 1989

    1989

  23. [31]

    Yang, Wright J., Huang T

    J. Yang, Wright J., Huang T. S., and Ma Y. Image super-resolution via sparse represen- tation, 2010

    2010

  24. [32]

    Y., and Xiong Y

    Chang H., Yeung D. Y., and Xiong Y. Super-resolution through neighbor embed- ding, 1981

    1981

  25. [33]

    Tipping M. E. and Bishop C. M. Bayesian image super-resolution, 2003

    2003

  26. [34]

    Super- resolution from a single image, 2009

    Glasner D., Bagon S., and Irani M. Super- resolution from a single image, 2009

    2009

  27. [35]

    Y. Jo, S. Wug Oh, J. Kang, and S. Joo Kim. Deep video super-resolution network using dynamic upsampling filters without explicit motion compensation, 2018

    2018

  28. [36]

    Kamrul Hasan, Shid- harthoRoy, Md.AshrafulAlam, EklasHos- sain, and Mohiuddin Ahmad

    Tarun Ghosh, Md. Kamrul Hasan, Shid- harthoRoy, Md.AshrafulAlam, EklasHos- sain, and Mohiuddin Ahmad. Multi-class probabilistic atlas-based whole heart seg- mentation method in cardiac ct and mri. IEEE Access, PP:1–1, 05 2021

    2021

  29. [37]

    U-net: Convolutional net- works for biomedical image segmentation, 2015

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional net- works for biomedical image segmentation, 2015

    2015

  30. [38]

    A u-net based discriminator for generative adversarial networks, 06 2020

    Edgar Schonfeld, Bernt Schiele, and Anna Khoreva. A u-net based discriminator for generative adversarial networks, 06 2020

    2020

  31. [39]

    Frequency-domain data augmenta- tion of vibration data for fault diagno- sis using deep neural networks

    Minseon Gwak, Seunghyun Ryu, Yong- beom Park, Hyeon-Woo Na, and PooGyeon Park. Frequency-domain data augmenta- tion of vibration data for fault diagno- sis using deep neural networks. In2022 22nd International Conference on Control, Automation and Systems (ICCAS), pages 1588–1...

    2022

  32. [40]

    Wavelet elm-ae based data augmentation and deep learning for efficient emotion recognition using eeg recordings.IEEE Access, 10:72171–72181, 2022

    Berna Ari, Kamran Siddique, Ömer Faruk Alçin, Muzaffer Aslan, Abdulkadir Şengür, and Raja Majid Mehmood. Wavelet elm-ae based data augmentation and deep learning for efficient emotion recognition using eeg recordings.IEEE Access, 10:72171–72181, 2022

    2022

  33. [41]

    Rethinking data augmentation for image super-resolution: A comprehensive analysis and a new strategy, 2020

    Jaejun Yoo, Namhyuk Ahn, and Kyung-Ah Sohn. Rethinking data augmentation for image super-resolution: A comprehensive analysis and a new strategy, 2020

    2020

  34. [42]

    Efficient blind super-resolution imaging via adaptive degradation-aware estimation.Knowledge- Based Systems, 297:111973, 2024

    HaoranYang, QileiLi, BinMeng, Gwanggil Jeon, Kai Liu, and Xiaomin Yang. Efficient blind super-resolution imaging via adaptive degradation-aware estimation.Knowledge- Based Systems, 297:111973, 2024

    2024

  35. [43]

    Sijin Ren and Cheryl Q. Li. Robustness of transfer learning to image degradation.Ex- pert Systems with Applications, 187:115877, 2022

    2022

  36. [44]

    Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan

    Dan Hendrycks, Norman Mu, Ekin D. Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan. Augmix: A sim- ple data processing method to improve ro- bustness and uncertainty, 2020

    2020

  37. [45]

    Ct-scan de- noising using a charbonnier loss genera- tive adversarial network.IEEE Access, 9:84093–84109, 2021

    Binit Gajera, Siddhant Raj Kapil, Dorsa Ziaei, Jayalakshmi Mangalagiri, Eliot Siegel, and David Chapman. Ct-scan de- noising using a charbonnier loss genera- tive adversarial network.IEEE Access, 9:84093–84109, 2021

    2021

  38. [46]

    Spa- tio temporal forest fire spread modeling us- ing cellular automata honey bee foraging and gis.Bulletin of Environment, Phar- macology and Life Sciences, 3:201–214, 12 2013

    Hassan Gazmeh, Ali Alesheikh, Moham- mad Karimi, and Alireza Chehreghan. Spa- tio temporal forest fire spread modeling us- ing cellular automata honey bee foraging and gis.Bulletin of Environment, Phar- macology and Life Sciences, 3:201–214, 12 2013

    2013

  39. [47]

    Single image deblurring based on auxiliary sobel loss function

    Jun-Pyo Hong, Sung-Jin Cho, Jeongmin Lee, Seo-Won Ji, and Sung-Jea Ko. Single image deblurring based on auxiliary sobel loss function. In2020 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia), pages 1–3, 2020

    2020

  40. [48]

    Wavelet transforma- tions and its applications in digital image processing, 10 2023

    A.H.M Barbhuiya. Wavelet transforma- tions and its applications in digital image processing, 10 2023

    2023

  41. [49]

    Design of fir filters for fast multiscale directional filter banks.Interna- tional Journal of u- and e-Service, Science and Technology, 7, 10 2014

    JEBAMALAR LEAVLINE EPIPHANY and s Sutha. Design of fir filters for fast multiscale directional filter banks.Interna- tional Journal of u- and e-Service, Science and Technology, 7, 10 2014

    2014

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.