Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation

Chunyan Wang; Dong Zhang; Jinhui Tang

REVIEW 3 major objections 5 minor 88 references

Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a diffusion model which denoises low-light feature maps can make weakly supervised semantic segmentation work nearly as well in the dark as in normal light, with depth maps providing the structural cues that keep…

desk verdict Sensible diffusion-based KD idea with strong internal ablations, but the real-dataset SOTA claim is undercut by an unfair comparison and a 12.6-point unexplained gap between the same baseline in two tables. read the letter →

arxiv 2507.07578 v2 pith:RJOYNQTY submitted 2025-07-10 cs.CV

Chunyan Wang , Dong Zhang , Jinhui Tang This is my paper · ORCID

classification cs.CV

keywords weakly-supervisedsemanticsegmentationlow-lightdiffusionmodelknowledgedistillationdepthpriorfeaturedenoisingclassactivationmapsdarkPASCALVOC

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

The paper sets out to make weakly supervised semantic segmentation work under low-light conditions, where existing methods produce unreliable class activation maps and semantically ambiguous pseudo-labels. Its central claim is that dark feature maps can be treated as noise-corrupted versions of normal-light feature maps, denoised by a diffusion model, and then aligned to a normal-light teacher by knowledge distillation. Adding depth maps as illumination-invariant geometric priors sharpens object boundaries and structures. On a synthetically darkened PASCAL VOC benchmark, the method reports 57.1% mIoU versus 43.4% for the baseline, nearly closing the gap to the normal-light teacher's 59.7%, and it reports 46.3% mIoU on the real LIS test set. If correct, this would give a practical way to segment dark images using only image-level labels.

What carries the argument

The central mechanism is the diffusion-guided knowledge distillation loop: a diffusion model is trained to predict the noise added to normal-light features, and at distillation time the dark feature is used as the starting noisy sample, denoised over a few DDIM steps, and then aligned to the normal-light feature by a distance loss, turning cross-illumination alignment into a denoising problem. The second mechanism is depth-guided feature fusion: monocular depth maps are encoded by convolutional layers into scale and shift parameters for spatial feature transformation, and an attention map combines the original dark features with the depth-enhanced features so that consistent semantic regions are preserved while fine-grained structure is added.

What would settle it

Compute the distributional distance (for example, Fréchet distance) between dark features and noise-corrupted normal-light features at the same network layers; if the two distributions overlap poorly, then the 'dark features are noisy normal features' premise is false. A second test is to replace the diffusion denoiser with a generic denoiser not trained on normal-light features and check whether the distillation gain disappears.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the illumination gap between training and deployment can be bridged at the feature level without first enhancing the image. The DGKD module trains a diffusion model on normal-light features, then feeds the dark feature as the initial noisy input of the denoising chain; the denoised features are matched to the normal-light teacher with a distillation loss applied to intermediate features and to the predicted segmentation masks. The DGF2 module takes depth maps from a pretrained monocular depth estimator and uses spatial feature transformation layers plus an attention-guided fusion to fold geometric structure into the dark features. The authors test this on a synthetically darkened PASCAL VOC 2012 dataset and on the real LIS dataset, reporting state-of-the-art results and showing that the gains do not come from a stronger backbone.

Load-bearing premise

The load-bearing premise is that dark feature maps actually resemble noise-corrupted normal-light features of the kind the diffusion model was trained to clean; if they sit outside that noise distribution, the denoising step would distort rather than align them and the reported gains would not transfer.

Editorial extensions

If this is right

Under the reported results, a weakly supervised low-light segmentation model can nearly match its normal-light teacher: the gap on the synthetic benchmark shrinks from 16.3 to 2.6 mIoU points.
Training on synthetically darkened images transfers to a real low-light dataset (LIS), where the method outperforms enhance-then-segment pipelines that first brighten the image and then run a segmentor.
Distilling and denoising the predicted segmentation mask gives the largest single improvement in the ablation, larger than intermediate-feature distillation alone.
The diffusion model is used only during training, so the deployed segmentor keeps the same inference cost and parameter count as the baseline.

Reading between the lines

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

The same 'degraded features are noisy clean features' recipe could be tested on other degradations such as haze, blur, or JPEG artifacts, since none of the machinery is illumination-specific.
If the dark-feature distribution assumption holds, the deeper lesson is that cross-domain distillation benefits from removing degradation noise at the feature level before matching, rather than aligning raw domain features directly.
The method's gains depend on the quality of the pretrained depth estimator; replacing it with weaker monocular depth models reduces but does not eliminate the improvement, so a stress test would use no depth prior at all and measure the residual benefit.
The reported benchmarks share categories between training and evaluation; extending to unseen or open-vocabulary categories would require the normal-light teacher and the depth priors to cover those categories, which the paper does not show.

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.

Desk Editor's Note

read the letter

Colleague,

The thing to know about this paper: the method itself is reasonable and the internal ablations are the strongest part, but the real-dataset comparison in Table 4 is not apples-to-apples, and there is a 12.6-point gap between the SSSS baseline in Table 1 and the same-named baseline in Table 3 that the paper never explains.

What's new: this is the first systematic take on weakly supervised low-light segmentation with image-level labels. The DGKD idea — treat dark features as noisy normal-light features, train a diffusion model on teacher features, then denoise student features before KD — is a sensible adaptation of DiffKD to a new domain. DGF2 with depth is standard SFT-style fusion, but the attention-gated combination is a small twist. Code is released.

What's good: the ablations are clean and progressive. Each module adds a solid chunk: +11.8 for DGKD, +1.9 more for DGF2 on the synthetic benchmark. The comparison against alternative KD losses and depth estimators is the right way to justify design choices. On synthetic dark PASCAL VOC, where everyone is retrained on the same dark data, the method's advantage is credible.

Soft spots, in order of severity:

First, Table 4. The method is trained on synthetically darkened PASCAL VOC, while all baselines are normal-light PASCAL-trained models that receive the enhanced LIS image only at test time. That is not a fair test of whether the DGKD/DGF2 modules help on real low-light data; it mostly measures that training on dark data helps. Without a dark-trained SSSS baseline, the 46.3% "state of the art" on LIS does not support the claim. The comparison needs all methods to be trained on the same dark-domain data.

Second, the baseline inconsistency: Table 1 reports SSSS on dark PASCAL at 43.4%, Table 3 reports the same method at 30.8%. The paper says both were retrained on the same synthetic low-light dataset. A 12.6-point gap suggests different training sets or protocols (e.g., SBD augmentation or PAMR usage), and the text does not say which. Until that is clarified, the headline +13.7 and the 10.1-point SOTA margin are hard to interpret.

Third, reproducibility details: no variance across runs, no diffusion model architecture specification, and no low-light synthesis parameters in the paper. The code helps, but these details should be in the main text or supplement.

Bottom line: the central idea is worth pursuing and the internal evidence is decent, but the paper overstates the real-dataset result. The paper is for people working on WSSS, low-light perception, and knowledge distillation. It deserves a serious referee — the right call is to send it out with a request to tighten the evaluation protocol, not to desk-reject it.

Referee Report

3 major / 5 minor

Summary. The paper proposes DGKD-WLSS, a framework for weakly-supervised semantic segmentation in low-light conditions. It combines two modules: Diffusion-Guided Knowledge Distillation (DGKD), which trains a diffusion model on normal-light teacher features and then denoises low-light student features before applying knowledge distillation, and Depth-Guided Feature Fusion (DGF2), which injects depth priors extracted by Depth Anything into the student backbone. The method is trained on synthetically darkened PASCAL VOC 2012 with image-level labels and evaluated on the synthetic dark PASCAL validation set and the real LIS test set. The authors report 57.1% mIoU on dark PASCAL (vs. 30.8–47.0% for adapted normal-light WSSS methods) and 46.3% mIoU on LIS (vs. 34.5–40.2% for enhancement-based baselines), and claim state-of-the-art performance in weakly supervised low-light segmentation.

Significance. If the results hold, this is a potentially useful contribution: it is one of the first systematic studies of WSSS under low light, and the idea of treating low-light features as noisy variants of normal-light features, then removing that noise with a diffusion model, is interesting and plausibly effective. The ablation study (Table 2) shows consistent improvements from each module, and the supplementary material adds experiments with a ViT backbone, a two-stage pipeline, hyperparameter sensitivity, and Cityscapes→NightCity transfer, which are commendable. The source code is said to be released. However, the significance is substantially tempered by evaluation-fairness issues in the real-dataset comparison and by an unexplained inconsistency in the baseline used for the headline gains.

major comments (3)

[Section 4.4, Table 4] The comparison on the LIS dataset is not apples-to-apples. DGKD-WLSS is trained on synthetically darkened PASCAL VOC and evaluated directly on dark LIS test images, whereas every baseline is trained on normal-light PASCAL VOC and receives either the raw or enhanced LIS image only at test time. No baseline is trained on dark-domain data, so the reported 46.3% mIoU cannot be separated from the advantage of having been trained on dark images at all. In addition, Section 4.2 states that for LIS experiments the teacher is trained on a normal-light LIS-augmented dataset; if that teacher is used in the Table 4 evaluation, DGKD-WLSS also has access to LIS training pairs that the comparison methods do not. The paper must add a dark-trained baseline (e.g., SSSS trained on synthetic dark PASCAL) and state explicitly what teacher data is used for the Table 4 results.
[Table 1 vs. Table 3] The baseline SSSS/WideResNet38 is reported inconsistently across the two tables. Table 1 gives the 'stu' baseline at 43.4% mIoU on the synthetic dark PASCAL VOC val set, while Table 3 reports SSSS retrained on the same synthetic low-light dataset at only 30.8% mIoU. This 12.6-point discrepancy is not explained. If the Table 3 protocol is what was used for the state-of-the-art comparison, then the '+13.7' gain over the Table 1 baseline is misleading; if the Table 1 protocol is the intended one, then the comparison in Table 3 is against an unfairly weak baseline. The authors must clarify the training protocols (e.g., whether PAMR, SBD augmentation, or different hyperparameters were used) and report a single consistent baseline.
[Section 3.2, Eqs. (4)–(7)] The diffusion model component is underspecified and its central assumption is unvalidated. The manuscript does not specify the architecture of the noise-prediction network Φθ, the exact training distribution for the diffusion loss (which layers and which pseudo-masks), the noise schedule, or the number of denoising steps used in inference beyond the supplementary mention of T=5. More importantly, using dark features as the initial noisy input for iterative denoising presumes that those features lie on the forward-noise manifold of normal-light features. No evidence is provided that this holds; if it does not, the 'denoised' features could be arbitrarily distorted and the reported gains would not transfer to new low-light domains. The authors should provide the missing architectural/training details and include a diagnostic experiment (e.g., comparing dark-feature initialization with Gaussian-noise initialization, or analyzing the feature-space distribution overlap).

minor comments (5)

[Throughout] There are numerous typos and spacing errors: 'Gussian' in Section 3.1, 'denosing' in Section 2.4, 'sythetic' in the Table 2 caption, 'consist' in Section 2.4, and inconsistent spacing around citations and equations. A careful proofreading pass is needed.
[Section 3.4, Eq. (14)] The total loss sums over m=3 levels, but Section 4.2 says distillation is applied to 'Layer 3 and Layer 6' and the predicted segmentation maps. The correspondence between the three terms (two feature layers plus the mask) should be stated explicitly.
[Section 4.2 / Table 4] The caption of Table 4 and the surrounding text should state clearly whether the DGKD-WLSS teacher used for the LIS evaluation is the LIS-augmented teacher described in Section 4.2. As written, the text says only that the method was 'trained on synthetically dark train set of PASCAL VOC', which is ambiguous about the teacher's data.
[All tables] No standard deviations or multiple-seed results are reported. Given that the DGF2 module adds only 1.9 mIoU over DGKD alone, error bars (or at least results over two or three seeds) would help determine whether that gain is significant.
[Section 2.2 / Section 4.1] The low-light synthesis pipeline is cited as [5,13] but never described. Since the entire training set is created by this pipeline, at least a short description of the synthesis process (e.g., gamma adjustment, noise model, whether paired data is used) is needed for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DGKD-WLSS is an empirical framework trained with standard losses and evaluated on external benchmarks; no reported quantity reduces by construction to its inputs.

full rationale

The paper does not derive its headline mIoU from a fitted quantity. The DGKD module trains a diffusion model on normal-light teacher features (Eq. 5) and then uses dark student features as the initial noisy input for denoising, with a knowledge-distillation loss (Eq. 7) between the denoised dark features and the normal-light features. This is a training objective, not a definition of the reported accuracy; the student is optimized to match the teacher, and the downstream segmentation mIoU is measured on held-out ground truth. The premise that low-light features are noisy variants of normal-light features is an explicit assumption ('Inspired by DiffKD [28], we assume that low-light features are essentially noisy variants of normal-light features') and is imported from an external prior work, not from a self-citation chain. Self-citations in the reference list (e.g., WS-FCN [56], Causal Intervention [70]) are used as comparison or related work, not as load-bearing justification for the method's validity. The diffusion model is trained on the teacher's own features rather than on the student's output, so the denoising target is not constructed from the student's predictions. The Tables 1 vs 3 baseline discrepancy (43.4 vs 30.8 mIoU for SSSS retrained on the same synthetic dark data) is a potential evaluation-protocol inconsistency, but it concerns experimental fairness rather than a derivation that reduces to its own inputs. No equation in the paper is equivalent by construction to the reported improvement; therefore no circular step is established.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central framework introduces two trainable modules (DGKD and DGF2) and three tuned hyperparameters (lambda, T, m), plus artifacts from the low-light synthesis pipeline. The main stated assumptions are the noise-variant view of low-light features, the reliability of off-the-shelf depth maps, and the transferability of synthetic darkness to real low-light images. No new physical or ontological entities are postulated.

free parameters (4)

lambda (depth fusion weight) = 0.5
Hyperparameter in Eq. (12) balancing consistent and complementary attention regions; ablated in Table 8 with best value 0.5.
T (diffusion timesteps) = 5
Number of denoising steps during training; ablated in Table 9 showing T=5 and T=6 nearly identical, chosen as 5.
m (number of distillation/diffusion levels) = 3
Number of feature levels included in Eq. (14) for diffusion and KD losses; set to 3 in experiments.
low-light synthesis parameters
The synthetic darkening pipeline from refs [5,13] introduces gamma and noise settings that are not specified in the paper; these parameters define the training distribution and are treated as fixed inputs.

assumptions (3)

domain assumption Low-light features are noisy variants of normal-light features
Stated at the start of Sec. 3.2: low-light features are essentially noisy variants of normal-light features due to illumination degradation. This is the conceptual basis for applying a diffusion denoiser trained on normal-light features.
domain assumption Depth maps from Depth Anything provide illumination-invariant geometric priors
Sec. 3.3 states depth maps serve as additional visual priors despite lighting variations; the paper relies on the pretrained Depth Anything model to produce accurate depth for dark images without fine-tuning.
domain assumption Synthetically darkened PASCAL VOC is a valid proxy for real low-light data
Sec. 4.1 uses a synthesis pipeline to create the training set and then evaluates on real LIS; this assumes distributional transfer from synthetic to real darkness.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation." pith.science (2026). https://pith.science/paper/RJOYNQTY

@misc{pith2026250707578,
  author       = {Pith},
  title        = {Pith review of: Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RJOYNQTY}},
  note         = {Machine review of arXiv:2507.07578}
}

read the original abstract

Weakly-supervised semantic segmentation aims to assign category labels to each pixel using weak annotations, significantly reducing manual annotation costs. Although existing methods have achieved remarkable progress in well-lit scenarios, their performance significantly degrades in low-light environments due to two fundamental limitations: severe image quality degradation (e.g., low contrast, noise, and color distortion) and the inherent constraints of weak supervision. These factors collectively lead to unreliable class activation maps and semantically ambiguous pseudo-labels, ultimately compromising the model's ability to learn discriminative feature representations. To address these problems, we propose Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-light Semantic Segmentation (DGKD-WLSS), a novel framework that synergistically combines Diffusion-Guided Knowledge Distillation (DGKD) with Depth-Guided Feature Fusion (DGF2). DGKD aligns normal-light and low-light features via diffusion-based denoising and knowledge distillation, while DGF2 integrates depth maps as illumination-invariant geometric priors to enhance structural feature learning. Extensive experiments demonstrate the effectiveness of DGKD-WLSS, which achieves state-of-the-art performance in weakly supervised semantic segmentation tasks under low-light conditions. The source codes have been released at:https://github.com/ChunyanWang1/DGKD-WLSS.

Figures

Figures reproduced from arXiv: 2507.07578 by the authors.

**Figure 2.** Comparisons of frameworks for weakly supervised [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** 3.1 Preliminaries A diffusion model is a form of generative model which has shown its impressive ability in a series of generative tasks [7, 12]. It comprises a forward process that adds noise to a sample and a reverse process that removes noises [25]. Concretely, given the sample data 𝒛0 ∈ R 𝐶×𝐻 ×𝑊 (where 𝐻 and 𝑊 are the height and width of the image spatial size, 𝐶 denotes the channel size), the goal is to model t… view at source ↗

Figures from the paper (3 more)

**Figure 3.** Figure 3: Our proposed DGKD-WLSS framework addresses the challenges of semantic confusion and low-confidence CAMs in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]

**Figure 4.** Figure 4: The feature visualization comparison (with vs. with [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Visualizations of segmentation masks generated by various methods, including the baseline, +DGKD, +All ( [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

88 extracted references · 69 canonical work pages

[1]

Jiwoon Ahn and Suha Kwak. 2018. Learning pixel-level semantic affinity with image-level supervision for weakly supervised semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 4981–4990

2018
[2]

Nikita Araslanov and Stefan Roth. 2020. Single-stage semantic segmentation from image labels. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 4253–4262

2020
[3]

Amy Bearman, Olga Russakovsky, Vittorio Ferrari, and Li Fei-Fei. 2016. What’s the point: Semantic segmentation with point supervision. InEuropean Conference on Computer Vision (ECCV) . 549–565

2016
[4]

Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias Müller
[5]

Tim Brooks, Ben Mildenhall, Tianfan Xue, Jiawen Chen, Dillon Sharlet, and Jonathan T Barron. 2019. Unprocessing images for learned raw denoising. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 11036–11045

2019
[6]

Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Timofte, and Yulun Zhang
[7]

Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng- Ann Heng, and Stan Z Li. 2024. A survey on generative diffusion models. IEEE Transactions on Knowledge and Data Engineering (2024)

2024
[8]

In IEEE International Conference on Computer Vision (ICCV)

Retinexformer: One-stage retinex-based transformer for low-light image enhancement. In IEEE International Conference on Computer Vision (ICCV). 12504– 12513

Show all 88 references

[9]

Linwei Chen, Ying Fu, Kaixuan Wei, Dezhi Zheng, and Felix Heide. 2023. Instance segmentation in the dark. International Journal of Computer Vision 131, 8 (2023), 2198–2218

2023
[10]

Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. 2018. Learning to see in the dark. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 3291–3300

2018
[11]

Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus En- zweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. 2016. The cityscapes dataset for semantic urban scene understanding. In IEEE Conference on Computer Vision and Pattern Recognition (CV...

2016
[12]

Yadang Chen, Dingwei Zhang, Yuhui Zheng, Zhi-Xin Yang, Enhua Wu, and Haixing Zhao. 2023. Boosting video object segmentation via robust and efficient memory network. IEEE Transactions on Circuits and Systems for Video Technology 34, 5 (2023), 3340–3352

2023
[13]

Ziteng Cui, Guo-Jun Qi, Lin Gu, Shaodi You, Zenghui Zhang, and Tatsuya Harada
[14]

Florinel-Alin Croitoru, Vlad Hondru, Radu Tudor Ionescu, and Mubarak Shah
[15]

IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 9 (2023), 10850–10869

Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 9 (2023), 10850–10869

2023
[16]

Duc-Tien Dang-Nguyen, Cecilia Pasquini, Valentina Conotter, and Giulia Boato
[17]

Xueqing Deng, Peng Wang, Xiaochen Lian, and Shawn Newsam. 2022. NightLab: A dual-level architecture with hardness detection for segmentation at night. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 16938–16948

2022
[18]

Dengxin Dai and Luc Van Gool. 2018. Dark model adaptation: Semantic image segmentation from daytime to nighttime. In 2018 21st International Conference on Intelligent Transportation Systems (ITSC) . 3819–3824

2018
[19]

Jifeng Dai, Kaiming He, and Jian Sun. 2015. Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation. In International Conference on Computer Vision (ICCV)

2015
[20]

Jianping Gou, Baosheng Yu, Stephen J Maybank, and Dacheng Tao. 2021. Knowl- edge distillation: A survey. International Journal of Computer Vision 129, 6 (2021), 1789–1819

2021
[21]

Chunle Guo, Chongyi Li, Jichang Guo, Chen Change Loy, Junhui Hou, Sam Kwong, and Runmin Cong. 2020. Zero-reference deep curve estimation for low- light image enhancement. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 1780–1789

2020
[22]

Bharath Hariharan, Pablo Arbeláez, Lubomir Bourdev, Subhransu Maji, and Jitendra Malik. 2011. Semantic contours from inverse detectors. In International Conference on Computer Vision (ICCV) . 991–998

2011
[23]

Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. 2010. The pascal visual object classes (voc) challenge. Inter- national Journal of Computer Vision 88, 2 (2010), 303–338

2010
[24]

1987.Digital image processing

Rafael C Gonzales and Paul Wintz. 1987.Digital image processing. Addison-Wesley Longman Publishing Co., Inc

1987
[25]

Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural Information Processing Systems (NeurIPS) 33 (2020), 6840–6851

2020
[26]

Jinhui Hou, Zhiyu Zhu, Junhui Hou, Hui Liu, Huanqiang Zeng, and Hui Yuan
[27]

Tao Huang, Shan You, Fei Wang, Chen Qian, and Chang Xu. 2022. Knowledge distillation from a stronger teacher. Advances in Neural Information Processing Systems (NeurIPS) 35 (2022), 33716–33727

2022
[28]

Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2014. Distilling the knowledge in a neural network. Neural Information Processing Systems (NeurIPS)

2014
[29]

Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the knowledge in a neural network. arXiv preprint (2015)

2015
[30]

Zhicheng Ji, Huan Zheng, Zhao Zhang, Qiaolin Ye, Yang Zhao, and Mingliang Xu
[31]

Hai Jiang, Ao Luo, Haoqiang Fan, Songchen Han, and Shuaicheng Liu. 2023. Low- light image enhancement with wavelet-based diffusion models.ACM Transactions on Graphics (TOG) 42, 6 (2023), 1–14

2023
[32]

Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jian- chao Yang, Pan Zhou, and Zhangyang Wang. 2021. Enlightengan: Deep light enhancement without paired supervision. IEEE transactions on image processing 30 (2021), 2340–2349

2021
[33]

Seonggwan Ko, Jinsun Park, Byungjoo Chae, and Donghyeon Cho. 2021. Learning lightweight low-light enhancement network using pseudo well-exposed images. IEEE Signal Processing Letters 29 (2021), 289–293

2021
[34]

Tao Huang, Yuan Zhang, Mingkai Zheng, Shan You, Fei Wang, Chen Qian, and Chang Xu. 2024. Knowledge diffusion for distillation. Advances in Neural Infor- mation Processing Systems (NeurIPS) 36 (2024)

2024
[35]

Young-Min Jeong, Tae-Sung Park, Jeong-Hyeok Park, and Jong-Ok Kim. 2023. Low-Light Image Enhancement via Distillation of NIR-to-RGB Conversion Knowl- edge. In 2023 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). IEEE, 714–718

2023
[36]

Mohit Lamba and Kaushik Mitra. 2021. Restoring extremely dark images in real time. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 3487–3497

2021
[37]

IEEE Transactions on Consumer Electronics 70, 1 (2023), 3626–3634

Multi-scale interaction network for low-light stereo image enhancement. IEEE Transactions on Consumer Electronics 70, 1 (2023), 3626–3634

2023
[38]

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. In European conference on computer vision (ECCV) . 740–755

2014
[39]

Risheng Liu, Long Ma, Jiaao Zhang, Xin Fan, and Zhongxuan Luo. 2021. Retinex- inspired unrolling with cooperative prior architecture search for low-light image enhancement. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 10561–10570

2021
[40]

Wenyu Liu, Wentong Li, Jianke Zhu, Miaomiao Cui, Xuansong Xie, and Lei Zhang
[41]

Viveka Kulharia, Siddhartha Chandra, Amit Agrawal, Philip Torr, and Ambrish Tyagi. 2020. Box2seg: Attention weighted loss and discriminative feature learning for weakly supervised segmentation. In European Conference on Computer Vision (ECCV). 290–308

2020
[42]

Hyeokjun Kweon and Kuk-Jin Yoon. 2024. From sam to cams: Exploring segment anything model for weakly supervised semantic segmentation. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 19499–19509

2024
[43]

Jeong-Hyeok Park, Tae-Hyeon Kim, and Jong-Ok Kim. 2022. Dual-teacher distilla- tion for low-light image enhancement. In2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) . IEEE, 1351– 1355

2022
[44]

Di Lin, Jifeng Dai, Jiaya Jia, Kaiming He, and Jian Sun. 2016. Scribblesup: Scribble- supervised convolutional networks for semantic segmentation. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR)

2016
[45]

Abhijith Punnappurath, Abdullah Abuolaim, Abdelrahman Abdelhamed, Alex Levinshtein, and Michael S Brown. 2022. Day-to-night image synthesis for training nighttime neural isps. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 10769–10778

2022
[46]

René Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, and Vladlen Koltun
[47]

Lixiang Ru, Yibing Zhan, Baosheng Yu, and Bo Du. 2022. Learning affinity from attention: end-to-end weakly-supervised semantic segmentation with trans- formers. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 16846–16855

2022
[48]

IEEE Transactions on Circuits and Systems for Video Technology 33, 10 (2023), 5855–5867

Improving nighttime driving-scene segmentation via dual image-adaptive learnable filters. IEEE Transactions on Circuits and Systems for Video Technology 33, 10 (2023), 5855–5867

2023
[49]

Hongmin Mu, Gang Zhang, MengChu Zhou, and Zhengcai Cao. 2024. End-to-end Semantic Segmentation Network for Low-Light Scenes. In IEEE International Conference on Robotics and Automation (ICRA) . 7725–7731

2024
[50]

Junwen Pan, Pengfei Zhu, Kaihua Zhang, Bing Cao, Yu Wang, Dingwen Zhang, Junwei Han, and Qinghua Hu. 2022. Learning self-supervised low-rank network for single-stage weakly and semi-supervised semantic segmentation. Interna- tional Journal of Computer Vision 130, 5 (2022), 1181–1195

2022
[51]

Christos Sakaridis, Dengxin Dai, and Luc Van Gool. 2021. ACDC: The adverse con- ditions dataset with correspondences for semantic driving scene understanding. In IEEE International Conference on Computer Vision (ICCV . 10765–10775

2021
[52]

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in Neural Information Processing System...

2019
[53]

Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. 2021. Segmenter: Transformer for semantic segmentation. In IEEE International Conference on Computer Vision (ICCV). 7262–7272

2021
[54]

Xin Tan, Ke Xu, Ying Cao, Yiheng Zhang, Lizhuang Ma, and Rynson WH Lau
[55]

Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, et al . 2024. Yolov10: Real-time end-to-end object detection. Advances in Neural Information Processing Systems 37 (2024), 107984–108011

2024
[56]

Chunyan Wang, Dong Zhang, Liyan Zhang, and Jinhui Tang. 2023. Coupling Global Context and Local Contents for Weakly-Supervised Semantic Segmenta- tion. IEEE Transactions on Neural Networks and Learning Systems (2023)

2023
[57]

Lixiang Ru, Heliang Zheng, Yibing Zhan, and Bo Du. 2023. Token contrast for weakly-supervised semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 3093–3102

2023
[58]

Christos Sakaridis, Dengxin Dai, and Luc Van Gool. 2019. Guided curriculum model adaptation and uncertainty-aware evaluation for semantic nighttime image segmentation. In IEEE International Conference on Computer Vision (ICCV . 7374– 7383

2019
[59]

Christos Sakaridis, Dengxin Dai, and Luc Van Gool. 2020. Map-guided curriculum domain adaptation and uncertainty-aware evaluation for semantic nighttime im- age segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 6 (2020), 3139–3153

2020
[60]

Kaixuan Wei, Ying Fu, Yinqiang Zheng, and Jiaolong Yang. 2021. Physics-based noise modeling for extreme low-light photography. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 11 (2021), 8520–8537

2021
[61]

Jiaming Song, Chenlin Meng, and Stefano Ermon. [n. d.]. Denoising Diffusion Implicit Models. In International Conference on Learning Representations
[62]

Zifeng Wu, Chunhua Shen, and Anton Van Den Hengel. 2019. Wider or deeper: Revisiting the resnet model for visual recognition. Pattern Recognition 90 (2019), 119–133

2019
[63]

Mengfan Xu, Wei Huang, and Rui Huang. 2023. MADA: Multi-Level Align- ment in Domain Adaptation Network for Nighttime Semantic Segmentation. In International Conference on Image, Vision and Computing (ICIVC) . IEEE, 352–357

2023
[64]

IEEE Transactions on Image Processing 30 (2021), 9085–9098

Night-time scene parsing with a large real dataset. IEEE Transactions on Image Processing 30 (2021), 9085–9098

2021
[65]

Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Heng- shuang Zhao. 2024. Depth anything: Unleashing the power of large-scale unla- beled data. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 10371–10381

2024
[66]

Sung-Hoon Yoon, Hoyong Kwon, Hyeonseong Kim, and Kuk-Jin Yoon. 2024. Class tokens infusion for weakly supervised semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 3595–3605

2024
[67]

Wenjing Wang, Chen Wei, Wenhan Yang, and Jiaying Liu. 2018. Gladnet: Low- light enhancement network with global awareness. In2018 13th IEEE international conference on automatic face & gesture recognition (FG 2018) . IEEE, 751–755

2018
[68]

Xintao Wang, Ke Yu, Chao Dong, and Chen Change Loy. 2018. Recovering realistic texture in image super-resolution by deep spatial feature transform. In IEEE conference on computer vision and pattern recognition (CVPR) . 606–615

2018
[69]

Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. 2018. Deep Retinex Decomposition for Low-Light Enhancement. InBritish Machine Vision Conference (BMVC). British Machine Vision Association

2018
[70]

Dong Zhang, Hanwang Zhang, Jinhui Tang, Xian-Sheng Hua, and Qianru Sun
[71]

Zhixiang Wei, Lin Chen, Tao Tu, Pengyang Ling, Huaian Chen, and Yi Jin. 2023. Disentangle then Parse: Night-time Semantic Segmentation with Illumination Disentanglement. In IEEE International Conference on Computer Vision (ICCV) . 21593–21603

2023
[72]

Xiaofeng Zhang, Zishan Xu, Hao Tang, Chaochen Gu, Wei Chen, and Abdulmo- taleb El Saddik. 2025. Wakeup-Darkness: When Multimodal Meets Unsupervised Low-light Image Enhancement. ACM Transactions on Multimedia Computing, Communications and Applications (2025)

2025
[73]

Xinliang Zhang, Lei Zhu, Hangzhou He, Lujia Jin, and Yanye Lu. 2024. Scribble Hides Class: Promoting Scribble-Based Weakly-Supervised Semantic Segmen- tation with Its Class Label. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), Vol. 38. 7332–7340

2024
[74]

Qingsen Yan, Yixu Feng, Cheng Zhang, Guansong Pang, Kangbiao Shi, Peng Wu, Wei Dong, Jinqiu Sun, and Yanning Zhang. 2025. HVI: A New color space for Low-light Image Enhancement. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

2025
[75]

Yuanhao Zhao, Genyun Sun, Ziyan Ling, Aizhu Zhang, and Xiuping Jia. 2024. Point Based Weakly Supervised Deep Learning for Semantic Segmentation of Remote Sensing Images. IEEE Transactions on Geoscience and Remote Sensing (2024)

2024
[76]

Dewei Zhou, Zongxin Yang, and Yi Yang. 2023. Pyramid diffusion models for low-light image enhancement. arXiv preprint arXiv:2305.10028 (2023)

2023 arXiv
[77]

Bingfeng Zhang, Siyue Yu, Yunchao Wei, Yao Zhao, and Jimin Xiao. 2024. Frozen clip: A strong backbone for weakly supervised semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 3796–3806

2024
[78]

Dong Zhang and Kwang-Ting Cheng. 2025. Generalized Task-Driven Medical Image Quality Enhancement With Gradient Promotion. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

2025
[79]

Dingwei Zhang, Hui Yan, Yadang Chen, Dichao Li, and Chuanyan Hao. 2024. Cross-domain few-shot learning based on feature adaptive distillation. Neural Computing and Applications 36, 8 (2024), 4451–4465

2024
[81]

Ad- vances in neural information processing systems 33 (2020), 655–666

Causal intervention for weakly-supervised semantic segmentation. Ad- vances in neural information processing systems 33 (2020), 655–666

2020
[82]

Dong Zhang, Hanwang Zhang, Jinhui Tang, Meng Wang, Xiansheng Hua, and Qianru Sun. 2020. Feature pyramid transformer. In European conference on computer vision (ECCV). 323–339

2020
[85]

Yian Zhao, Wenyu Lv, Shangliang Xu, Jinman Wei, Guanzhong Wang, Qingqing Dang, Yi Liu, and Jie Chen. 2024. Detrs beat yolos on real-time object detection. In IEEE conference on Computer Vision and Pattern Recognition (CVPR). 16965–16974

2024
[88]

Tianfei Zhou and Wenguan Wang. 2024. Cross-image pixel contrasting for semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence (2024). Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation , , A Supplementar...

2024
[2015]

In Proceedings of the 6th ACM multimedia systems conference

Raise: A raw images dataset for digital image forensics. In Proceedings of the 6th ACM multimedia systems conference . 219–224
[2020]

IEEE transactions on pattern analysis and machine intelligence 44, 3 (2020), 1623–1637

Towards robust monocular depth estimation: Mixing datasets for zero- shot cross-dataset transfer. IEEE transactions on pattern analysis and machine intelligence 44, 3 (2020), 1623–1637. , , Chunyan Wang, Dong Zhang, and Jinhui Tang

2020
[2021]

In IEEE International Conference on Computer Vision (ICCV)

Multitask aet with orthogonal tangent regularity for dark object detection. In IEEE International Conference on Computer Vision (ICCV) . 2553–2562
[2023]

Zoedepth: Zero-shot transfer by combining relative and metric depth.arXiv preprint arXiv:2302.12288 (2023)

2023 arXiv
[2024]

Advances in Neural Information Processing Systems (NeurIPS) 36 (2024)

Global structure-aware diffusion process for low-light image enhancement. Advances in Neural Information Processing Systems (NeurIPS) 36 (2024)

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Jiwoon Ahn and Suha Kwak. 2018. Learning pixel-level semantic affinity with image-level supervision for weakly supervised semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 4981–4990

2018

[2] [2]

Nikita Araslanov and Stefan Roth. 2020. Single-stage semantic segmentation from image labels. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 4253–4262

2020

[3] [3]

Amy Bearman, Olga Russakovsky, Vittorio Ferrari, and Li Fei-Fei. 2016. What’s the point: Semantic segmentation with point supervision. InEuropean Conference on Computer Vision (ECCV) . 549–565

2016

[4] [4]

Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias Müller

[5] [5]

Tim Brooks, Ben Mildenhall, Tianfan Xue, Jiawen Chen, Dillon Sharlet, and Jonathan T Barron. 2019. Unprocessing images for learned raw denoising. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 11036–11045

2019

[6] [6]

Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Timofte, and Yulun Zhang

[7] [7]

Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng- Ann Heng, and Stan Z Li. 2024. A survey on generative diffusion models. IEEE Transactions on Knowledge and Data Engineering (2024)

2024

[8] [8]

In IEEE International Conference on Computer Vision (ICCV)

Retinexformer: One-stage retinex-based transformer for low-light image enhancement. In IEEE International Conference on Computer Vision (ICCV). 12504– 12513