CARDIE: clustering algorithm on relevant descriptors for image enhancement

Giulia Bonino; Luca Alberto Rizzo

REVIEW 5 major objections 6 minor 21 references

CARDIE: clustering algorithm on relevant descriptors for image enhancement

T0 review · 5 major / 6 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 unsupervised clustering by luminosity and hue, rather than by semantic content, yields the clusters that matter for image enhancement, and that oversampling those clusters measurably improves tone mapping and…

desk verdict A useful, low-tech clustering method for rebalancing image enhancement datasets, but the relevance metric rests on a shaky functional fit and the PSNR gains need error bars. read the letter →

arxiv 2509.06116 v1 pith:2QSAFCNY submitted 2025-09-07 cs.CV

Giulia Bonino , Luca Alberto Rizzo This is my paper · ORCID

classification cs.CV

keywords imageclusteringenhancementtonemappingdenoisingdatasetresamplingluminancedescriptorshueunsupervisedlearning

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 introduces CARDIE, an unsupervised algorithm that groups images by luminosity and hue descriptors rather than by semantic content, and argues that these photometric clusters are the relevant units for image enhancement. Its evidence is a new quantification method: fitting a two-parameter luminance-response curve to each input-output image pair and testing, with Kolmogorov-Smirnov tests, whether clusters undergo statistically different transformations. On the 5K tone-mapping dataset, CARDIE clusters differ exactly where the tone-mapping network behaves differently, while semantic clusters from scene recognition and manual labels are nearly indistinguishable under the same test. Oversampling CARDIE's minority clusters improves tone-mapping PSNR by 0.22 dB and denoising PSNR by 1.18 dB on full test sets, with minority-class gains of 0.60 dB and 2.73 dB. If correct, the paper shows that a cheap, label-free color-and-luminosity rebalancing of training data can beat semantic clustering and random oversampling for image enhancement.

What carries the argument

The central object is a boolean descriptor per image: one luminosity flag (low, average, or high, from the median against the 20th and 80th percentiles of a $K$-image average luminance) plus a set of dominant hue bins, each flagged when at least $1/l$ of the hue distribution falls in a $\Delta\theta$ window around a reference color. These descriptors reduce an image dataset to a small tabular dataset, which HDBSCAN clusters with the Jaccard distance; the minimum-cluster-size and variance-threshold hyperparameters are selected via silhouette score. The companion machinery is the quantification method: fit the two-parameter Naka-Rushton curve $L_{\mathrm{pp}} = L_{\mathrm{i}}^{\gamma}/(L_{\mathrm{i}}^{\gamma}+\mu^{\gamma})$ to 100 sampled pixels per image pair, then run two-sided Kolmogorov-Smirnov tests on the fitted $\gamma$ and $\mu$ distributions, and on local-variance changes $\Delta\bar{\sigma}$ for denoising, to form the cluster-difference indicator $I^A_{l,a}$. This pair of tools converts "are these clusters relevant?" into a concrete statistical comparison and makes the descriptors cheap enough to compute on any enhancement dataset.

What would settle it

Re-run the oversampling experiment on a tone-mapping network trained with CARDIE clusters selected by the two-parameter KS test, but on a dataset where the ground-truth operator is strongly local (e.g., local tone curves): if PSNR gains vanish or reverse, the relevance criterion is an artifact of the global curve family. Alternatively, replace the fitted curve with a per-pixel gain map and recompute the cluster-difference indicator; clusters that become indistinguishable under the richer model would show that the claimed relevance did not survive a more complete description of the operator.

Watch

Extended reading notes

Core claim

On its own terms, the paper discovers that the clusters that matter for image enhancement are defined by low-level photometric properties, not by objects or scenes. CARDIE computes a per-pixel average luminance over a subsample of the dataset, flags each image as dark, average, or bright by comparing its median luminance with the 20th and 80th percentiles of that average, and records which of five hue windows contain enough of the hue distribution; the resulting boolean vectors are clustered with HDBSCAN under the Jaccard distance. The relevance test fits $L_{\mathrm{pp}}=L_{\mathrm{i}}^{\gamma}/(L_{\mathrm{i}}^{\gamma}+\mu^{\gamma})$ to each input-output pair, obtains distributions of $\gamma$ and $\mu$ per cluster, and uses two-sided KS tests to build the cluster-difference indicator. The paper reports that CARDIE's clusters are statistically distinct exactly where the tone-mapping network applies different transformations—by luminosity level—whereas semantic clusters are not. It then shows that oversampling the minority CARDIE clusters three times improves tone mapping by 0.22 dB PSNR and denoising by 1.18 dB PSNR on the full test sets, with minority-class improvements of 0.60 dB and 2.73 dB, outperforming random oversampling and semantic-cluster oversampling.

Load-bearing premise

The argument rests on a single premise: that the two-parameter luminance-response curve $L_{\mathrm{pp}}=L_{\mathrm{i}}^{\gamma}/(L_{\mathrm{i}}^{\gamma}+\mu^{\gamma})$ summarizes what a tone-mapping operator does to an image, so that statistically different fitted parameters between clusters indicate genuinely different treatment; local, content-dependent, or clipped adjustments outside this curve family would break that link.

Editorial extensions

If this is right

A practical corollary is that image-enhancement training sets can be audited and rebalanced without any labels or pretrained semantic networks; the paper shows this improves PSNR for both tone mapping and denoising.
The cluster-difference indicator can be used on any input-output enhancement dataset to find which splits of the data correspond to genuinely different operator behavior, before any training happens.
Because the descriptors are luminosity and hue only, the approach transfers to any dataset where enhancement effects are global and photometric, and the additional raw-pipeline results indicate it also works for classical processing pipelines.
The resampling procedure is architecture-agnostic: it changes only the training set, so it can be combined with any tone-mapping or denoising network, not only the two used in the paper.

Reading between the lines

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

Editorial extension: for operators with strong local or content-dependent effects, such as dehazing or local tone mapping, the two-parameter curve used in the relevance test may be too rigid; a richer per-pixel gain model could reveal that CARDIE's clusters are not the relevant ones for those operators.
A natural next experiment, not run in the paper, is to make the hue bins and luminosity thresholds adaptive per dataset and check whether the PSNR gains scale; the paper fixes these by stability rather than optimizing them.
Because resampling only changes training-data composition, combining CARDIE oversampling with architectural or augmentation improvements could compound the reported gains, and the same descriptors could be used to detect distribution shift in incoming image streams.

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.

Desk Editor's Note

read the letter

Let me give you the short version. This paper proposes CARDIE, an unsupervised clustering method that labels images by low-level luminance and hue descriptors and clusters them with HDBSCAN. The new part is combining these simple descriptors with a Naka-Rushton-based relevance metric to guide dataset resampling for image enhancement. It's a genuinely useful idea, and the authors compare it against semantic clustering (ResNet Places365 and MIT5K labels) and release their code.

What the paper does well: the pipeline is refreshingly low-tech, the comparisons are sensible, and the downstream resampling experiments on tone mapping (HDRNet) and denoising (NAFNet) are the right way to test practical value. The HDR+ extension in the appendix is a good robustness check. If the PSNR gains are real, oversampling CARDIE minority clusters is a cheap win for practitioners.

Now the soft spots, in rough order of concern. First, the headline PSNR numbers (Tables 5-6) come from a single training run per condition, with no error bars or seed variation. A 0.22 dB full-set gain and a 1.18 dB denoising gain could easily move within run-to-run noise for these models. Second, the cluster relevance indicator (Eq. 7-8) runs many pairwise KS tests at alpha=0.05 without multiple-testing correction, and the OR-logic (gamma or mu) makes differences easier to flag. This likely overstates how distinctive the clusters are. Third, the synthetic denoising dataset is built by injecting extra noise into CARDIE-defined minority classes; that makes the oversampling benefit partly self-fulfilling, though it's a fair test of rebalancing if the test set has the same distribution. Fourth, and this is the load-bearing assumption, Eq. 4 fits a two-parameter global Naka-Rushton curve to each image under a tone mapper. For a local operator like HDRNet, that functional form is misspecified, and the paper shows only a few fits with no residual analysis across the roughly 5000 images. If gamma and mu are poorly identified, the KS differences between CARDIE clusters could reflect fitting artifacts rather than genuine differences in enhancement. There's also an inherent circularity: CARDIE clusters by luminosity, and a global tone mapper's fitted parameters depend on luminosity, so separation is partly baked in.

That said, none of this is fatal. The resampling result is a concrete downstream outcome independent of the KS metric, and the method is simple enough to verify quickly. I'd send it to peer review with a request for: multiple seeds with mean and std, proper multiple-testing handling, residual diagnostics for Eq. 4, and a less contrived noise experiment (or a clear justification for the construction). The paper targets researchers in image enhancement and dataset analysis; after revision it could be a solid practical contribution.

Referee Report

5 major / 6 minor

Summary. The manuscript introduces CARDIE, an unsupervised clustering algorithm that maps each image to binary descriptors for luminosity level and dominant hue intervals, then clusters the resulting descriptor vectors with HDBSCAN. To evaluate cluster usefulness, the authors propose fitting the Naka-Rushton curve of Eq. (4) to each input/output image pair and testing, via pairwise KS tests, whether the fitted gamma and mu distributions differ across clusters (Eqs. 6-7); for denoising they use the change in average local variance (Eqs. 5 and 8). They compare CARDIE with semantic clustering baselines derived from MIT5K manual labels and ResNet50-Places365 attributes on the MIT5K (D5K) and HDR+ datasets, and report that oversampling CARDIE minority clusters improves PSNR for HDRNet tone mapping (+0.22 dB full, +0.60 dB minority) and NAFNet denoising (+1.18 dB full, +2.73 dB minority) relative to baseline.

Significance. If the claims hold, the paper offers a simple, training-free alternative to semantic clustering for image-enhancement dataset analysis and rebalancing, with a public implementation as a reproducibility asset. The descriptor construction is transparent and computationally light, and the comparison with two semantic baselines is a useful step for the community. The reported PSNR gains are practically relevant for tone mapping and denoising pipelines. However, the evidence rests on an unvalidated global functional form, on multiple-testing issues, and on a denoising experiment with an injected-noise confound, so the significance is conditional on the requested revisions.

major comments (5)

[§2.2, Eq. (4), and §3.2] The cluster-relevance comparison rests entirely on the assumption that each tone mapping operator can be summarized by the two-parameter Naka-Rushton curve L^pp = L^i^gamma / (L^i^gamma + mu^gamma). For a local operator such as HDRNet, output luminance at a pixel depends on local context, so this global functional form is not the data-generating process. The paper reports only visual fit examples in Fig. 2 and provides no per-image residuals, R^2 values, or parameter-identifiability statistics over the roughly 5,000 images. If the fits are systematically poor, or if gamma and mu are unstable, the KS differences between CARDIE clusters in Fig. 5 can reflect fitting artifacts rather than enhancement-relevant differences. The authors should provide quantitative goodness-of-fit diagnostics and a robustness check with an alternative functional form or with a direct comparison of input/output luminance histograms.
[§2.1-§2.2, cluster difference indicator Eq. (7)] CARDIE's descriptors are exactly the luminosity level and hue that determine the fitted gamma and mu in Eq. (4), so the cluster difference indicator I^T_l,a is biased toward finding separation for CARDIE. For example, low-luminosity clusters will by construction produce different mu from high-luminosity clusters for any tone mapper that shifts mean luminance. This construction bias does not by itself invalidate the resampling results, but it weakens the comparative claim that CARDIE is more relevant than semantic clustering, because the relevance metric shares the same variables that CARDIE clusters on. The authors should either use a metric not defined through the clustered variables, such as direct perceptual or pixel-wise error of the enhancement output, or explicitly justify why this construction bias does not drive the Sec. 3 conclusions.
[§2.2, Eqs. (6)-(8), Figs. 5-6] Pairwise KS tests are performed for every cluster pair and a p-value threshold of 0.05 is used without any multiple-testing correction. With 12 clusters on D5K there are 66 pairwise comparisons, so several false 'statistically different' results are expected under the null hypothesis. The cluster difference indicators in Figs. 5-6 and the conclusions drawn from them should be recomputed with adjusted p-values, for example using Benjamini-Hochberg correction, or the authors should justify the uncorrected alpha level and report the number of tests performed.
[§3.1 and §4, Tables 5-6] The denoising experiment is confounded by the deliberate injection of approximately twice the noise into the two least represented classes of D5K before the oversampling experiment. Because the same classes are then oversampled by CARDIE, the +1.18 dB full-dataset and +2.73 dB minority-class gains in Tables 5 and 6 can be explained by the extra noise added to those classes rather than by CARDIE's cluster relevance. A control experiment with uniform noise, or with the same noise-injection protocol applied to random and ResNet oversampling, is required to attribute the gain to the clustering method.
[§4, Tables 5-6] The PSNR differences are reported from what appears to be a single training run per condition, with no standard deviations, error bars, or repeated-seed results. The claimed improvements (0.22 dB and 1.18 dB) are small relative to typical run-to-run variation in trained HDRNet and NAFNet models. At minimum, the authors should report mean and standard deviation over three or more seeds, and ideally a paired significance test across runs.

minor comments (6)

[Author affiliation and Fig. 10 caption] There are typos that should be corrected: 'Nice Reseach Center' should be 'Nice Research Center', and 'ground thruth' in the Fig. 10 caption should be 'ground truth'.
[§2.2, Eq. (3)] Equation (3) mixes image-level and pixel-level notation: L(x^pp_j, p) = A(L(x^i_j), p) is hard to parse. Please clarify that A is applied to the luminance channel and what p indexes in the operator argument.
[§4, oversampling protocol] The text does not specify whether the oversampling for the denoising experiment uses cluster labels computed on D5K or on D_noisy; Tables 2 and 3 show different cluster structures, so this choice matters for reproducibility. Please state explicitly which cluster labels are used for each task.
[Eq. (9)] The noise model x_noisy ~ N(x_gt, a_j x_gt + b_j) does not state whether pixel values are in [0,1] or [0,255]; the reported ranges for a_j and b_j are otherwise ambiguous.
[§2.2, sensitivity to P] The claim that the results are 'not too sensitive' to P=100 is not accompanied by any sensitivity analysis. Please add a short experiment varying P or state the range of values tested.
[Figs. 5 and 6] The figure captions should explicitly define the color code (green = I_l,a = 1, red = I_l,a = 0) rather than leaving it to the main text.

Circularity Check

1 steps flagged · score 4.0 of 10

The Sec. 3 cluster-relevance result is partly self-confirming because Eq. 4's μ parameter is defined to capture luminosity-level differences, the same axis CARDIE clusters on; the resampling experiments in Sec. 4 provide independent, non-circular support.

self definitional [Sec. 2.1.1 (Algorithm 1), Sec. 2.2 (Eq. 4), Sec. 3.2 (Eq. 7)]
"Eq.4 represents a modified gamma compression, 17 with the additional global parameter μ that aims to capture the difference in tone mapping for images at different luminosity levels. ... asking if clusters are relevant for the operators A is in fact equivalent to asking for which combinations of cluster indexes l and a the following statements are true: ... is M A l statistically different to M A a? ... and verifying if those combinations reflect the expected dynamics of A (e.g. images with different luminosity levels are in different clusters for tone mapping operators)."

CARDIE's primary descriptor is the low/average/high luminosity flag computed from per-image median luminance, so its clusters are separated along the luminosity axis. Eq. 4's fitted parameter μ is introduced specifically to capture 'the difference in tone mapping for images at different luminosity levels,' and Eq. 7 declares clusters relevant precisely when the fitted μ/γ distributions differ across clusters. Consequently, the KS-based demonstration that CARDIE clusters are 'relevant' is largely a restatement of the construction: a clustering that separates luminosity levels will, by the assumed model family, produce statistically distinct μ distributions for any global or approximately global tone mapper. The Sec.

full rationale

The paper's core practical result—that oversampling CARDIE minority clusters improves tone mapping and denoising PSNR—is anchored to actual trained models (HDRNet and NAFNet) on held-out test data, so it is not circular. The circularity concern is confined to the Sec. 3 relevance metric: CARDIE clusters by luminosity, while Eq. 4's μ parameter is defined to capture luminosity-dependent tone mapping, so the statistical separation of fitted γ/μ distributions between CARDIE clusters is expected from the definitions and does not independently establish that the clusters matter for enhancement quality. However, the paper does not rely solely on this metric; the resampling experiments in Tables 5 and 6 provide an external, task-level validation. No self-citation chains, imported uniqueness theorems, or renamed known results were found. Overall, the paper has one partly self-confirming evaluation step but retains independent empirical content, warranting a moderate score rather than a high one.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The method depends on several hand-chosen constants and domain assumptions. The most important is that Eq. 4 provides a faithful summary of tone mapping for relevance testing. The synthetic noise construction and single-run training further limit the independence of the resampling evidence.

free parameters (9)

K = 100
Number of images subsampled to compute the average image used for luminosity thresholds (Alg. 1, Sec. 2.1.1). Called a free parameter of CARDIE.
Luminosity percentiles = 20th and 80th
Thresholds for low, average, and high luminosity flags chosen by hand in Alg. 1.
Dominant color set C = {0, pi/3, 2pi/3, 4pi/3, 5pi/3}
Chosen hue centers (red, yellow, green, blue, magenta) for dominant color flags, stated in Sec. 3.
Hue window delta_theta = pi/6
Interval around each dominant color center used in Eq. 2, fixed in Sec. 3.
HDBSCAN min_cluster_size m_min = not reported
Hyperparameter selected by grid search over approximately 200 values using silhouette score, Sec. 2.1.2. Exact chosen values are not reported.
Descriptor variance threshold sigma_d = not reported
Hyperparameter excluding low-variance descriptors, selected by grid search together with m_min, Sec. 2.1.2.
Pixel subsample P = 100
Number of pixels used to fit Eq. 4 per image to avoid overfitting, Sec. 2.2.
Oversampling factor n_overs = 3
Replication factor for minority classes in the resampling experiments, Sec. 4.
Noise model bounds a_j and b_j = a in [0.3,0.5], b in [0.1,0.2]; high-noise classes a in [0.6,1.0], b in [0.2,0.4]
Parameters of the synthetic noise used to build D_noisy, Sec. 3.1.

assumptions (5)

domain assumption The two-parameter Naka-Rushton curve (Eq. 4) adequately describes tone mapping luminance changes for every image pair.
The entire cluster relevance metric is built on differences in fitted gamma and mu; if the curve family is misspecified, relevance conclusions do not follow.
domain assumption KS p-value < 0.05 without multiple-testing correction indicates meaningful cluster differences.
Pairwise tests across many cluster pairs at alpha = 0.05 inflate the false positive rate; the paper does not correct for multiple testing.
domain assumption Luminance percentiles computed from a K-image average are stable and dataset-representative.
Thresholds are based on K = 100 subsample and define all luminosity labels; no stability analysis is given for different K.
domain assumption Change in average local variance over 16x16 windows quantifies the impact of denoising.
Eq. 5 defines delta_sigma as the denoising effect; no justification is given that this captures perceptually or PSNR-relevant denoising quality.
domain assumption Silhouette score on the same dataset used for evaluation selects good clusterings.
Hyperparameter search optimizes silhouette on D, and the same D is used for relevance tests and to define minority classes; this risks overfitting the cluster labels to the evaluation data.

how reviews work

0 comments

Cite this review

Pith. "Pith review of CARDIE: clustering algorithm on relevant descriptors for image enhancement." pith.science (2026). https://pith.science/paper/2QSAFCNY

@misc{pith2026250906116,
  author       = {Pith},
  title        = {Pith review of: CARDIE: clustering algorithm on relevant descriptors for image enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2QSAFCNY}},
  note         = {Machine review of arXiv:2509.06116}
}

read the original abstract

Automatic image clustering is a cornerstone of computer vision, yet its application to image enhancement remains limited, primarily due to the difficulty of defining clusters that are meaningful for this specific task. To address this issue, we introduce CARDIE, an unsupervised algorithm that clusters images based on their color and luminosity content. In addition, we introduce a method to quantify the impact of image enhancement algorithms on luminance distribution and local variance. Using this method, we demonstrate that CARDIE produces clusters more relevant to image enhancement than those derived from semantic image attributes. Furthermore, we demonstrate that CARDIE clusters can be leveraged to resample image enhancement datasets, leading to improved performance for tone mapping and denoising algorithms. To encourage adoption and ensure reproducibility, we publicly release CARDIE code on our GitHub.

Figures

Figures reproduced from arXiv: 2509.06116 by the authors.

**Figure 2.** Examples of luminance distributions fits for classical (Reinhard), deep-learning (HDRNET) [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Γ AC l and MAC l distributions at different luminosity levels each image couple x i j , x pp j the change of the local pixel variance due to the denoising algorithm ∆¯σ A j = |σ¯ (xpp = A(xj )) − σ¯ (xj )| (5) where σ¯ is the average of the local variances computed on 16x16 windows for each image. As discussed in Sec.1, one of the main goals of our analysis is to test the relevance of CARDIE for IE algorithms, i.e… view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: I AT l,a for images with different luminosity levels for AC on D5K pair x i j , x pp j a series of descriptors (dj,1, · · · , dj,k), a cluster label c l j and two fitting parameters [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

**Figure 5.** Figure 5: I T l,a for clusters obtained by ResNet50 (left) and by MIT5K (right) for D5K 4 Experimental Results on Resampling In this section, we present the experimental results obtained by applying the oversampling to the two Image Enhancement tasks: HDRNet for Tone Mapping and…

**Figure 6.** Figure 6: Il,a for clusters obtained by ResNet50 (left) and by MIT5K (right) for Dnoisy [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

**Figure 7.** Figure 7: Example of visual results for the Tone Mapping task using HDRNet and [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

**Figure 8.** Figure 8: Example of visual results for the Denoising task using NAFNet and [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

**Figure 9.** Figure 9: Sample of images clustered by CARDIE for D5K (left) ResNet50-places365 for D5K (right, ground truth images are displayed for clarity) Appendix A: Clusters’ samples for CARDIE Fig.9 show samples of images belonging to different clusters obtained by applying respectively…

**Figure 10.** Figure 10: Sample of images clustered by CARDIE for D5K (left) ResNet50-places365 for HDRPLUS (ground thruth images are displayed for clarity) [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

**Figure 11.** Figure 11: I T l,a for clusters obtained by ResNet50 (left) and by CARDIE (right) for HDR+ Appendix C: L and H distributions Throughout all the analysis conducted in this paper we computed the luminance and hue distributions for each image in terms of its RGB content. 22 [PITH…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

[1]

L. Chen, X. Chu, X. Zhang, et al. , ``Simple baselines for image restoration,'' (2022)

work page 2022
[2]

L. Zhao, A. Abdelhamed, and M. Brown, ``Learning tone curves for local image enhancement,'' IEEE Access 10 , 1--1 (2022)

work page 2022
[3]

Deep Bilateral Learning for Real-Time Image Enhancement

M. Gharbi, J. Chen, J. T. Barron, et al. , ``Deep bilateral learning for real-time image enhancement,'' CoRR abs/1707.02880 (2017)

work page Pith review arXiv 2017
[4]

M. A. Hashmani, S. M. Jameel, H. Alhussain, et al. , ``Accuracy performance degradation in image classification models due to concept drift,'' International Journal of Advanced Computer Science and Applications 10 (5) (2019)

work page 2019
[5]

Khan and N

A. Khan and N. Malim, ``Comparative studies on resampling techniques in machine learning and deep learning models for drug-target interaction prediction,'' Molecules (Basel, Switzerland) 28 (2023)

work page 2023
[6]

B. Zhou, A. Lapedriza, A. Khosla, et al. , ``Places: A 10 million image database for scene recognition,'' IEEE Transactions on Pattern Analysis and Machine Intelligence (2017)

work page 2017
[7]

B. Zhou, H. Zhao, X. Puig, et al. , ``Semantic understanding of scenes through the ade20k dataset,'' (2018)

work page 2018
[8]

T.-Y. Lin, M. Maire, S. Belongie, et al. , ``Microsoft coco: Common objects in context,'' (2015)

work page 2015

Show all 21 references

[9]

J. Xiao, J. Hays, K. A. Ehinger, et al. , ``Sun database: Large-scale scene recognition from abbey to zoo,'' in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition , 3485--3492 (2010)

2010
[10]

W. V. Gansbeke, S. Vandenhende, S. Georgoulis, et al. , ``Learning to classify images without labels,'' CoRR abs/2005.12320 (2020)

2020 arXiv
[11]

Caron, I

M. Caron, I. Misra, J. Mairal, et al. , ``Unsupervised learning of visual features by contrasting cluster assignments,'' (2021)

2021
[12]

C. Niu, H. Shan, and G. Wang, ``Spice: Semantic pseudo-labeling for image clustering,'' IEEE Transactions on Image Processing 31 , 7264–7278 (2022)

2022
[13]

R. J. G. B. Campello, D. Moulavi, A. Zimek, et al. , ``Hierarchical density estimates for data clustering, visualization, and outlier detection,'' ACM Trans. Knowl. Discov. Data 10 (2015)

2015
[14]

V. V. Baligodugula and F. Amsaad, ``Unsupervised learning: Comparative analysis of clustering techniques on high-dimensional data,'' (2025)

2025
[15]

P. J. Rousseeuw, ``Silhouettes: A graphical aid to the interpretation and validation of cluster analysis,'' Journal of Computational and Applied Mathematics 20 , 53--65 (1987)

1987
[16]

R. W. Naka KI, ``S-potentials from luminosity units in the retina of fish (cyprinidae),'' J Physiol 3 , 587--99 (1966)

1966
[17]

Poynton, Digital Video and HDTV Algorithms and Interfaces , Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1 ed

C. Poynton, Digital Video and HDTV Algorithms and Interfaces , Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1 ed. (2003)

2003
[18]

Gharbi, J

M. Gharbi, J. Chen, J. T. Barron, et al. , ``Deep bilateral learning for real-time image enhancement,'' ACM Transactions on Graphics (TOG) 36 (4), 118 (2017)

2017
[19]

Bychkovsky, S

V. Bychkovsky, S. Paris, E. Chan, et al. , ``Learning photographic global tonal adjustment with a database of input / output image pairs,'' in The Twenty-Fourth IEEE Conference on Computer Vision and Pattern Recognition , (2011)

2011
[20]

A. Foi, M. Trimeche, V. Katkovnik, et al. , ``Practical poissonian-gaussian noise modeling and fitting for single-image raw-data,'' IEEE Transactions on Image Processing 17 (10), 1737--1754 (2008)

2008
[21]

S. W. Hasinoff, D. Sharlet, R. Geiss, et al. , ``Burst photography for high dynamic range and low-light imaging on mobile cameras,'' ACM Transactions on Graphics (Proc. SIGGRAPH Asia) 35 (6) (2016)

2016

Pith tools

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

[1] [1]

L. Chen, X. Chu, X. Zhang, et al. , ``Simple baselines for image restoration,'' (2022)

work page 2022

[2] [2]

L. Zhao, A. Abdelhamed, and M. Brown, ``Learning tone curves for local image enhancement,'' IEEE Access 10 , 1--1 (2022)

work page 2022

[3] [3]

Deep Bilateral Learning for Real-Time Image Enhancement

M. Gharbi, J. Chen, J. T. Barron, et al. , ``Deep bilateral learning for real-time image enhancement,'' CoRR abs/1707.02880 (2017)

work page Pith review arXiv 2017

[4] [4]

M. A. Hashmani, S. M. Jameel, H. Alhussain, et al. , ``Accuracy performance degradation in image classification models due to concept drift,'' International Journal of Advanced Computer Science and Applications 10 (5) (2019)

work page 2019

[5] [5]

Khan and N

A. Khan and N. Malim, ``Comparative studies on resampling techniques in machine learning and deep learning models for drug-target interaction prediction,'' Molecules (Basel, Switzerland) 28 (2023)

work page 2023

[6] [6]

B. Zhou, A. Lapedriza, A. Khosla, et al. , ``Places: A 10 million image database for scene recognition,'' IEEE Transactions on Pattern Analysis and Machine Intelligence (2017)

work page 2017

[7] [7]

B. Zhou, H. Zhao, X. Puig, et al. , ``Semantic understanding of scenes through the ade20k dataset,'' (2018)

work page 2018

[8] [8]

T.-Y. Lin, M. Maire, S. Belongie, et al. , ``Microsoft coco: Common objects in context,'' (2015)

work page 2015