Pith. sign in

REVIEW 3 major objections 7 minor 39 references

BrightVAE: Luminosity Enhancement in Underexposed Endoscopic Images

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read BrightVAE, an attention-augmented hierarchical VQ-VAE, reports top PSNR, SSIM, and LPIPS for enhancing low-light endoscopic images on Endo4IE.

desk verdict A credible incremental VQ-VAE extension with a strong-looking but under-validated SOTA claim; the loss ablation appears to use the test set, and the paper needs code and a validation split before the numbers can be trusted. read the letter →

arxiv 2411.14663 v1 pith:2LDALNDS submitted 2024-11-22 eess.IV cs.CV

classification eess.IVcs.CV
keywords low-lightimageenhancementluminosityendoscopicimaginghierarchicalVQ-VAEattentionmechanismSSIMLPIPSEndo4IEdataset
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

BrightVAE is an architecture for brightening underexposed endoscopic images, built on a hierarchical vector-quantized variational autoencoder (VQ-VAE). The paper claims that adding attention to the encoder and to the vector-quantization step, processing features through global and local receptive fields, and joining encoder to decoder with skip connections lets the model beat all compared methods on the Endo4IE benchmark. The reported margins are about 3.25 dB in PSNR and 0.045 in SSIM over the closest medical-imaging baseline, with a lower (better) LPIPS score as well. If these numbers hold, BrightVAE would give clinicians a clearer view of dark endoscopic scenes, and that visibility matters because diagnosis and treatment planning depend on what these images reveal.

What carries the argument

The load-bearing mechanism is the pair of attention-augmented modules inside a hierarchical VQ-VAE: the Attencoder and the Attenquant. The Attencoder appends a multi-head attention layer to a convolutional encoder so that the model can emphasize the image regions that matter most for brightness and detail. The Attenquant computes an attention score for each feature, normalizes the scores with softmax, and uses them to weight the squared Euclidean distances between encoder outputs and codebook embeddings before selecting the closest embedding, which biases quantization toward perceptually important features. These modules run in a dual-receptive-field design with global and local branches, and skip connections carry encoder feature maps directly to the decoder to preserve spatial detail. The total loss combines an MSE restoration term, a VQ latent-loss term with a stop-gradient operator, and an SSIM-based similarity term weighted 0.08.

What would settle it

Re-run the eight-way loss ablation from Section 4.D.2 with a held-out validation split, freeze the winning loss, retrain BrightVAE, and evaluate on the withheld 266-image Endo4IE test set; if a non-SSIM loss wins on validation or BrightVAE no longer leads every baseline, the central claim is not supported.

Watch

Extended reading notes

Core claim

The central claim is that BrightVAE surpasses all compared models in PSNR, SSIM, and LPIPS for low-light endoscopic image enhancement on the Endo4IE dataset. This is achieved by a hierarchical VQ-VAE whose encoder (Attencoder) and quantizer (Attenquant) are augmented with attention, and whose global and local receptive-field branches are coupled through skip connections to the decoder. The paper reports a PSNR gain of about 3.252 dB, an SSIM gain of 0.045, and an LPIPS reduction of about 0.014 over the nearest medical-imaging competitor, and gains of 3.337 dB PSNR, 0.03 SSIM, and 0.038 LPIPS over the plain hierarchical VQ-VAE that inspired it. Those numbers are presented as evidence that the attention modules and dual-receptive design preserve fine details while correcting uneven illumination.

Load-bearing premise

The results stand on the assumption that the SSIM similarity loss was chosen without looking at the Endo4IE test set; if the test set guided that choice, the reported ranking is not an independent comparison.

Editorial extensions

If this is right

  • On the Endo4IE benchmark, BrightVAE currently stands as the top-performing method across PSNR, SSIM, and LPIPS, assuming the reported test numbers are accurate.
  • The 3.34 dB gain over a plain hierarchical VQ-VAE indicates that the attention-augmented encoder and quantizer, not the backbone alone, produce most of the improvement.
  • The identified loss configuration, with reconstruction, latent, and SSIM similarity terms weighted 1.0, 0.25, and 0.08, gives a starting recipe for future endoscopic low-light enhancement methods.
  • After training, the method takes a single underexposed frame and produces an enhanced image without needing paired reference data at inference, so it is directly applicable to endoscopic video frames.

Reading between the lines

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

  • Because Endo4IE's low-light images are synthetic pairs produced by a pretrained generative model, the reported advantage should be rechecked on real endoscopic video before relying on it clinically.
  • The paper gives no inference-time cost or parameter count, so a deployment study measuring latency and memory on endoscopy hardware is a natural next step.
  • Since the SSIM loss won the ablation while a perceptual loss achieved the best secondary LPIPS value, combining SSIM with a perceptual term is a plausible variant the authors did not test.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The manuscript proposes BrightVAE, a hierarchical VQ-VAE architecture for low-light endoscopic image enhancement. The main architectural additions are an attention-based encoder (Attencoder), an attention-based quantization module (Attenquant), dual global/local receptive fields, and skip connections between encoder and decoder. The training loss combines an MSE reconstruction term, a VQ latent loss, and a similarity loss that is selected by an ablation over eight candidates. Experiments on the Endo4IE dataset compare BrightVAE with 14 baselines plus VQ-VAE and report that it achieves the highest PSNR, SSIM, and LPIPS, with claimed gains of about 3.25 dB PSNR over Mou et al. and 3.34 dB over VQ-VAE. The paper also includes component and loss-function ablations.

Significance. If the empirical claims hold, BrightVAE would be a useful new state of the art on the public Endo4IE benchmark, and the two attention-based modules are simple enough to be adapted by other researchers. The paper's strengths are that it targets a concrete medical imaging task, uses a public benchmark, compares against many recent methods, and includes ablations. The central SOTA claim is falsifiable on a public dataset. However, the contribution is primarily empirical, and the current evidence is incomplete: no code or pretrained weights are provided, the comparison protocol is underspecified, and no variance estimates are reported. In particular, the loss-function selection may have been performed on the test set, which would make the reported SOTA margins unverified. With additional details and a clean evaluation protocol, the work could be a solid benchmark contribution.

major comments (3)
  1. [Section 4.D.2 (Table 3)] The selection of LSimilarity is load-bearing for the paper's central claim, but the paper never states which data were used for this ablation. Section 4.B describes only a training set (690 pairs) and a test set (266 images); no validation split is mentioned. If the eight loss candidates were compared on the same Endo4IE test images subsequently used in Table 1, then the model has been chosen by optimizing the exact metrics on the test set, and the reported SOTA margins in Section 4.C are not an independent evaluation. Please state explicitly the split used for Table 3 and, if necessary, redo the loss selection on a held-out validation split before computing final test numbers.
  2. [Section 4.C (Table 1)] The protocol for comparing with the 14 baselines is not specified. It is unclear whether each baseline was retrained on the Endo4IE training split, used with publicly released checkpoints, or evaluated by the authors, and whether any pre/post-processing was applied. Additionally, Table 1 is embedded as an image rather than as a text table; the surrounding text gives only the improvements over two methods, not the full numerical values for all methods. Without the protocol and the full numeric table, the claim that BrightVAE 'surpasses all other models' cannot be checked or reproduced.
  3. [Section 4.C] No error bars, standard deviations, or multiple-seed results are reported, and no statistical test is applied. The claimed advantages over the nearest competitor are small in some metrics (0.045 SSIM and 0.014 LPIPS), so it is possible that the ranking is within run-to-run variation. Please report results over at least three training runs for the proposed method (and ideally for baselines) or otherwise characterize the variability, so the reader can judge whether the stated margins are meaningful.
minor comments (7)
  1. [Section 3.E] The equations for LRest and LLatent are missing from the text; only prose descriptions and placeholders appear. Please include the actual formulas.
  2. [Tables 1-3] Tables 1, 2, and 3 appear only as figures; the numeric values are not available to the reader. Please typeset them as real tables.
  3. [Sections 4.D.2 and 4.E] The paper uses both 'SSI loss' and 'SSIM loss' for the same quantity; choose one term and use it consistently.
  4. [Section 3.A] There is a redundant phrase: 'for local branch operations aimed at local branch.' Please rephrase.
  5. [Section 4.A] Specific hyperparameters (codebook size, attention heads, embedding dimension, learning-rate schedule values) are not given; these details are needed to reproduce the architecture.
  6. [Related Work and Section 4.C] Reference [15] is mentioned in Related Work but does not appear among the methods compared in Section 4.C; please clarify its status.
  7. [Abstract and Conclusion] The phrase 'significant advancements' is not supported by statistical testing; please either soften it or add significance tests.

Circularity Check

1 steps flagged · score 6.0 of 10

Loss-function selection in §4.D.2 appears to use the Endo4IE test set to pick LSimilarity, making the §4.C SOTA metrics partly fitted rather than independently measured.

  1. fitted input called prediction [Section 4.B (Dataset and Metrics), Section 4.D.2 (Loss-function ablation), Section 4.C (Comparison with State-of-the-art), Tables 1 and 3]
    "The Endo4IE dataset is structured into two subsets: a training set containing 690 image pairs and a testing set comprising 266 images. ... Our analysis of various loss functions for enhancing dimly lit endoscopic images reveals that each function offers unique benefits. Still, the SSI loss function stands out as the most effective overall. Table 3 shows that SSI achieves the highest metric scores in SSIM at 0.879 and PSNR at 30.576 ... making it the preferred choice ..."

    The paper defines only a training split (690 image pairs) and a testing split (266 images); no validation split is ever introduced. In the ablation of §4.D.2, the LSimilarity loss is selected as the SSIM loss because it gave the highest PSNR (30.576) and SSIM (0.879) on the Endo4IE data. The same Endo4IE test set is then used in §4.C to report BrightVAE's state-of-the-art PSNR/SSIM/LPIPS margins. Choosing the loss to maximize the exact metrics on the exact test set, and then presenting those test-set numbers as an independent comparison, is a fitted-input-called-prediction loop: the reported advantage over Mou et al. and VQ-VAE is partly produced by the selection procedure rather than measured against a fixed model.

full rationale

The paper is empirical rather than derivational, and its architecture-level claims (Attencoder, Attenquant, dual receptive fields, skip connections) are evaluated by ablation against a VQ-VAE baseline; those ablations are not circular. The self-citations to refs. [15] and [22] are not load-bearing: VQLIE is used as a comparison baseline and the shadow-revealing method appears only in the related-work survey. The one substantive circular component is the selection of LSimilarity in §4.D.2: because the paper describes only train and test splits and does not state that the loss ablation was performed on a held-out set, the reported test-set SOTA in §4.C is statistically contaminated by the loss choice. That makes the headline claim only partially independent evidence, although the architecture itself is not derived from the benchmark metrics.

Assumptions & free parameters 3 free parameters · 5 assumptions · 2 invented entities

The central claim rests on the validity of the Endo4IE synthetic benchmark, the adequacy of PSNR/SSIM/LPIPS as clinical proxies, a clean validation protocol for choosing the loss, and comparable baseline numbers. None of these are established by the paper itself; the loss weights are free parameters chosen by hand.

free parameters (3)
  • lambda_Rest = 1.0
    Weight of MSE restoration loss in total loss, chosen by hand (Section 4.E).
  • lambda_Latent = 0.25
    Weight of VQ-VAE latent loss, set as in [23] with beta balancing; chosen by hand (Section 4.E).
  • lambda_Similarity = 0.08
    Weight of similarity loss; chosen by hand alongside the selection of SSIM loss among eight alternatives (Sections 4.E and 4.D.2).
assumptions (5)
  • domain assumption Endo4IE synthetic low-light pairs generated by a pre-trained generative model are representative of real underexposed endoscopic images.
    Section 4.B states the synthetic images were generated using a sophisticated pre-trained generative model. The clinical relevance of the central claim inherits this premise.
  • domain assumption The ablation in Section 4.D.2 used a validation split that does not leak test-set information.
    The paper does not describe a validation split; the choice of SSIM loss based on PSNR/SSIM/LPIPS on the benchmark is only methodologically sound if no test leakage occurred. This is not stated.
  • domain assumption PSNR, SSIM, and LPIPS are adequate proxies for diagnostic quality in endoscopy.
    Used throughout as the sole evaluation; no physician study or task-based evaluation is reported.
  • domain assumption Baseline results in Table 1 are produced under comparable training and evaluation conditions, such as the same dataset splits, preprocessing, and input sizes.
    The paper does not specify whether baselines were retrained or numbers were copied from original papers.
  • standard math The hierarchical VQ-VAE formulation of [23] is assumed valid as the base architecture.
    Section 3 builds on [23] without re-derivation; acceptable as background, not a new claim.
invented entities (2)
  • Attencoder
    purpose: Encoder with multi-head attention to selectively amplify relevant features.
    Proposed architectural module; only internal ablation on Endo4IE demonstrates its benefit; no external validation or formal analysis.
  • Attenquant
    purpose: Vector quantization with attention-weighted distances to bias embedding selection.
    Proposed architectural module; effectiveness shown only via self-reported ablation; no independent test.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BrightVAE: Luminosity Enhancement in Underexposed Endoscopic Images." pith.science (2026). https://pith.science/paper/2LDALNDS

@misc{pith2026241114663,
  author       = {Pith},
  title        = {Pith review of: BrightVAE: Luminosity Enhancement in Underexposed Endoscopic Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LDALNDS}},
  note         = {Machine review of arXiv:2411.14663}
}
read the original abstract

The enhancement of image luminosity is especially critical in endoscopic images. Underexposed endoscopic images often suffer from reduced contrast and uneven brightness, significantly impacting diagnostic accuracy and treatment planning. Internal body imaging is challenging due to uneven lighting and shadowy regions. Enhancing such images is essential since precise image interpretation is crucial for patient outcomes. In this paper, we introduce BrightVAE, an architecture based on the hierarchical Vector Quantized Variational Autoencoder (hierarchical VQ-VAE) tailored explicitly for enhancing luminosity in low-light endoscopic images. Our architecture is meticulously designed to tackle the unique challenges inherent in endoscopic imaging, such as significant variations in illumination and obscured details due to poor lighting conditions. The proposed model emphasizes advanced feature extraction from three distinct viewpoints-incorporating various receptive fields, skip connections, and feature attentions to robustly enhance image quality and support more accurate medical diagnoses. Through rigorous experimental analysis, we demonstrate the effectiveness of these techniques in enhancing low-light endoscopic images. To evaluate the performance of our architecture, we employ three widely recognized metrics-SSIM, PSNR, and LPIPS-specifically on Endo4IE dataset, which consists of endoscopic images. We evaluated our method using the Endo4IE dataset, which consists exclusively of endoscopic images, and showed significant advancements over the state-of-the-art methods for enhancing luminosity in endoscopic imaging.

Figures

Figures reproduced from arXiv: 2411.14663 by the authors.

Figure 1
Figure 1. Endoscopic Image Comparison: Low-light endoscopic images with corresponding normal-light (ground truth) images, highlighting challenges like reduced visibility and contrast loss. Ground truth images illustrate the need for luminosity enhancement to restore clarity and diagnostic accuracy. Many general image enhancement methods fail with low-light images, such as those from endoscopic procedures. These methods may en… view at source ↗
Figure 2
Figure 2. Block diagram of BrightVAE, featuring dual processing paths. The top pathway displays the Global Attencoder, [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Diagram of the proposed Attencoder architecture. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Diagram of the proposed Attenquant architecture. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Comparative display of endoscopic images of Endo4IE [3]: the first row shows original low [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Sequential comparison of endoscopic image enhancements [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 37 canonical work pages

  1. [1]

    Bai et al

    Medical Methods: Several contributions have been made to improve image quality under poor lighting and exposure conditions in medical imaging enhancement. Bai et al. [6] introduce EndoUIC, a promptable diffusion transformer capable of adaptively correcting illumination in capsule endoscopy, significantly improving image quality across varying exposure lev...

  2. [2]

    LLFlow method by Wang et al

    General Methods: Recent advancements in image enhancement technology have effectively combined sophisticated AI techniques with traditional methods, achieving significant improvements in image quality. LLFlow method by Wang et al. [13] uses normalizing flows to model light distribution effectively. Cui et al. [14] introduced the Innovative Adaptive Transf...

  3. [3]

    General”, and M refers to “Medical

    The third section of the loss function, LSimilarity, calculates the similarity between the output and ground truth images. Eight different loss functions were evaluated in subsection 4.D.2 of the ablation study to identify the appropriate loss function for assessing similarity. Each loss component is weighted appropriately in the total loss function to en...

  4. [4]

    Image Restoration Loss (LRest): This loss function is crucial for ensuring that the reconstructed image closely matches the original in terms of pixel values. It is calculated using the Mean Squared Error (MSE), a common measure in image processing that quantifies the difference between the enhanced and original images: Here, yi represents the pixel value...

  5. [5]

    Latent Space Optimization Loss (L Latent): Following the approach used in [23], we apply a latent loss to ensure that the encoded features closely match the nearest vectors in the embedding space, promoting consistency and stability in the encoding process. The formula for this loss is: where Ze(x) denotes the encoder output, e is the closest embedding ve...

  6. [6]

    EndoUIC: Promptable Diffusion Transformer for Unified Illumination Correction in Capsule Endoscopy

    L. Bai, Q. Tan, T. Chen, W. J. Nah, Y. Li, Z. He, S. Yuan, et al., “EndoUIC: Promptable Diffusion Transformer for Unified Illumination Correction in Capsule Endoscopy,” arXiv preprint arXiv:2406.13705, 2024. 17

  7. [7]

    Replacing the baseline encoder with our Attencoder and the quantization module with our Attenquant module significantly improves model efficiency

    Analysis of Components and Structure: This section examines the impact of modifying components in the baseline network. Replacing the baseline encoder with our Attencoder and the quantization module with our Attenquant module significantly improves model efficiency. Table II highlights the performance benefits of connecting the Attencoder directly to the ...

  8. [8]

    Assessing the Impact of Various Loss Functions on Model Performance: In this section, we evaluate various loss functions to optimize our model’s performance in determining the similarity between reconstructed images and ground truth (LSimilarity) for endoscopic low-light enhancement. This analysis identifies metrics that best capture image fidelity and ac...

Show all 39 references
  1. [9]

    Multiscale structural-aware exposure correction for endoscopic imaging,

    A. Garc´ıa -Vega, R. Espinosa, L. Ram´ırez -Guzm´an, T. Bazin, L. Falc´on -Morales, G. Ochoa -Ruiz, D. Lamarque, and C. Daul, “Multiscale structural-aware exposure correction for endoscopic imaging,” in IEEE 20th International Symposium on Biomedical Imaging (ISBI), pp. 1 -5, 2023

  2. [10]

    Adaptive image enhancement based on guide image and fraction- power transformation for wireless capsule endoscopy,

    M. Long, Z. Li, X. Xie, G. Li, and Z. Wang, “Adaptive image enhancement based on guide image and fraction- power transformation for wireless capsule endoscopy,” IEEE Transactions on Biomedical Circuits and Systems, vol. 12, no. 5, pp. 993 –1003, 2018

  3. [11]

    A novel hybrid endoscopic dataset for evaluating machine learning-based photometric image enhancement models,

    A. Garc´ıa-Vega, R. Espinosa, G. Ochoa-Ruiz, T. Bazin, L. Falc´on-Morales, D. Lamarque, and C. Daul, “A novel hybrid endoscopic dataset for evaluating machine learning-based photometric image enhancement models,” in Mexican International Conference on Artificial Intelligence, ...

  4. [12]

    Deep Pyramid Network for Low -light Endoscopic Image Enhancement,

    G. Yue, J. Gao, R. Cong, T. Zhou, L. Li, and T. Wang, “Deep Pyramid Network for Low -light Endoscopic Image Enhancement,” IEEE Transactions on Circuits and Systems for Video Technology, Oct. 9, 2023

  5. [13]

    Global and Local Enhancement of Low- light Endoscopic Images,

    E. Mou, H. Wang, M. Yang, E. Cao, Y. Chen, C. Ran, and Y. Pang, “Global and Local Enhancement of Low- light Endoscopic Images,” Preprints, no. 2023111954, Nov. 2023. Available: https://doi.org/10.20944/preprints202311.1954.v1

  6. [14]

    You only need 90k parameters to adapt light: a light weight transformer for image enhancement and exposure correction,

    Z. Cui, K. Li, L. Gu, S. Su, P. Gao, Z. Jiang, Y. Qiao, and T. Harada, “You only need 90k parameters to adapt light: a light weight transformer for image enhancement and exposure correction,” arXiv preprint arXiv:2205.14871, 2022

  7. [15]

    A dynamic histogram equalization for image contrast enhancement,

    M. Abdullah -Al-Wadud, M. H. Kabir, M. A. A. Dewan, and O. Chae, “A dynamic histogram equalization for image contrast enhancement,” IEEE Transactions on Consumer Electronics, vol. 53, no. 2, pp. 593 –600, 2007

  8. [16]

    Image enhancement based on equal area dualistic sub -image histogram equalization method,

    Y. Wang, Q. Chen, and B. Zhang, “Image enhancement based on equal area dualistic sub -image histogram equalization method,” IEEE Transactions on Consumer Electronics, vol. 45, no. 1, pp. 68 –75, 1999

  9. [17]

    Enhancement of low exposure images via recursive histogram equalization algorithms,

    K. Singh, R. Kapoor, and S. K. Sinha, “Enhancement of low exposure images via recursive histogram equalization algorithms,” Optik, vol. 126, no. 20, pp. 2619 –2625, 2015

  10. [18]

    A fusion -based enhancing method for weakly illuminated images,

    X. Fu, D. Zeng, Y. Huang, Y. Liao, X. Ding, and J. Paisley, “A fusion -based enhancing method for weakly illuminated images,” Signal Processing, vol. 129, pp. 82 –96, 2016

  11. [19]

    Structure -revealing low-light image enhancement via robust retinex model,

    M. Li, J. Liu, W. Yang, X. Sun, and Z. Guo, “Structure -revealing low-light image enhancement via robust retinex model,” IEEE Transactions on Image Processing, vol. 27, no. 6, pp. 2828 –2841, 2018

  12. [20]

    Structure and illumination constrained GAN for medical image enhancement,

    Y. Ma, J. Liu, Y. Liu, H. Fu, Y. Hu, J. Cheng, H. Qi, Y. Wu, J. Zhang, and Y. Zhao, “Structure and illumination constrained GAN for medical image enhancement,” IEEE Transactions on Medical Imaging, vol. 40, no. 12, pp. 3955 –3967, 2021

  13. [21]

    Low-light image enhancement with normalizing flow,

    Y. Wang, R. Wan, W. Yang, H. Li, L.-P. Chau, and A. Kot, “Low-light image enhancement with normalizing flow,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 3, pp. 2604 –2612, 2022

  14. [22]

    These diverse approaches underline a dynamic evolution of learning-based methods in image processing

    employs a VQ-VAE architecture to enhance low-light images, adding another layer of sophistication. These diverse approaches underline a dynamic evolution of learning-based methods in image processing. By leveraging deep learning, these methods not only enhance the capabilities...

  15. [23]

    Revealing Shadows: Low-Light Image Enhancement Using Self- Calibrated Illumination,

    F. Koohestani, N. Karimi, and S. Samavi, “Revealing Shadows: Low-Light Image Enhancement Using Self- Calibrated Illumination,” in Proceedings of the 32nd International Conference on Electrical Engineering (ICEE), IEEE, pp. 1 -7, May 14, 2024

  16. [24]

    Beyond brightening low-light images,

    Y. Zhang, X. Guo, J. Ma, W. Liu, J. Zhang, “Beyond brightening low-light images,” International Journal of Computer Vision, vol. 129, pp. 1013 –1037, 2021

  17. [25]

    EFINet: Restoration for low-light images via enhancement-fusion iterative network,

    C. Liu, F. Wu, X. Wang, “EFINet: Restoration for low-light images via enhancement-fusion iterative network,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 32, no. 12, pp. 8486 –8499, 2022

  18. [26]

    Low -light image enhancement via feature restoration,

    Y. Yang, Y. Zhang, and X. Guo, “Low -light image enhancement via feature restoration,” in ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2440-2444, 2022

  19. [27]

    A fast and lightweight network for low -light image enhancement,

    Y. Zhang, X. Di, J. Wu, R. Fu, Y. Li, Y. Wang, Y. Xu, G. Yang, C. Wang, “A fast and lightweight network for low -light image enhancement,” arXiv preprint arXiv:2304.02978, 2023

  20. [28]

    Learning multiscale photo exposure correction,

    M. Afifi, K. G. Derpanis, B. Ommer, and M. S. Brown, “Learning multiscale photo exposure correction,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9157 -9167, 2021

  21. [29]

    Multi -branch and progressive network for low -light image enhancement,

    K. Zhang, C. Yuan, J. Li, X. Gao, and M. Li, “Multi -branch and progressive network for low -light image enhancement,” IEEE Transactions on Image Processing, vol. 32, pp. 2295 -2308, 2023

  22. [30]

    Preserving Details in Darkness: A VQ - VAE Based Approach with SSIM Loss for Low-Light Image Enhancement,

    F. Koohestani, Z. Nabizadeh Shahr Babak, N. Karimi, S. Samavi, “Preserving Details in Darkness: A VQ - VAE Based Approach with SSIM Loss for Low-Light Image Enhancement,” in Proceedings of the World AI IoT Congress (AIoT), IEEE, pp. 342 -348, May 29, 202 4. 18

  23. [31]

    Generating diverse structure for image inpainting with hierarchical VQ-VAE,

    J. Peng, D. Liu, S. Xu, H. Li, “Generating diverse structure for image inpainting with hierarchical VQ-VAE,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10775– 10784, 2021

  24. [32]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600 –612, 2004

  25. [33]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 586 –595, 2018

  26. [34]

    Optimizing the Dice Score and Jaccard Index for Medical Image Segmentation: Theory and Practice,

    J. Bertels, T. Eelbode, M. Berman, D. Vandermeulen, F. Maes, R. Bisschops, and M. B. Blaschko, “Optimizing the Dice Score and Jaccard Index for Medical Image Segmentation: Theory and Practice,” in Proc. 22nd Int. Conf. Medical Image Computing and Computer Assisted Intervention...

  27. [35]

    An algorithm for total variation minimization and applications,

    A. Chambolle, “An algorithm for total variation minimization and applications,” Journal of Mathematical Imaging and Vision, vol. 20, pp. 89 -97, 2004

  28. [36]

    Modern information retrieval: A brief overview,

    A. Singhal, “Modern information retrieval: A brief overview,” IEEE Data Eng. Bull., vol. 24, no. 4, pp. 35-43, 2001

  29. [37]

    Comparing kullback -leibler divergence and mean squared error loss in knowledge distillation,

    T. Kim, J. Oh, N. Kim, S. Cho, and S. -Y. Yun, “Comparing kullback -leibler divergence and mean squared error loss in knowledge distillation,” arXiv preprint arXiv:2105.08919, 2021

  30. [38]

    Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,

    W. Xue, L. Zhang, X. Mou, and A. C. Bovik, “Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,” IEEE Transactions on Image Processing, vol. 23, no. 2, pp. 684 -695, 2013

  31. [39]

    Intrinsic decompositions for image editing,

    N. Bonneel, B. Kovacs, S. Paris, and K. Bala, “Intrinsic decompositions for image editing,” in Computer Graphics Forum, vol. 36, no. 2, pp. 593 -609, 2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.