Pith. sign in

REVIEW 3 major objections 6 minor 54 references

DFVO: Learning Darkness-free Visible and Infrared Image Disentanglement and Fusion All at Once

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A joint cascaded multi-task network performs low-light enhancement and visible-infrared fusion in one pass, beating two-stage baselines on night fusion metrics.

desk verdict A worthwhile one-stage low-light fusion architecture, but the headline PSNR is unverifiable without a stated reference — and possibly trained into the model. read the letter →

arxiv 2505.04526 v1 pith:CZGJL5BJ submitted 2025-05-07 cs.CV cs.AI

classification cs.CVcs.AI
keywords visible-infraredimagefusionlow-lightenhancementcascadedmulti-tasklearningilluminationdisentanglementhypercross-attentionlatentfeatureextractionLLVIPdatasetnighttime
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 claims that the usual two-stage recipe for darkness-free infrared-visible fusion — first brighten the visible image, then fuse it with the infrared image — loses information at each transfer and can be replaced by a single network trained on three tasks at once: infrared reconstruction, illumination disentanglement, and fusion. If correct, DFVO gives clearer, better-lit fused images in night driving scenes with less noise and over-exposure than either conventional fusion or two-stage enhancement-then-fusion baselines, and it improves downstream pedestrian and vehicle detection. The headline evidence is 63.258 dB PSNR and 0.724 correlation with source images on the LLVIP dataset, plus a preference score of 62.5 percent in a human perception study.

What carries the argument

The load-bearing mechanism is the latent-common feature extractor (LCFE), which combines a details-extraction module (DEM) and a hyper cross-attention module (HCAM) to obtain high-frequency semantic and low-frequency texture features from both modalities. Three task-specific heads share these latent features: infrared image reconstruction, illumination disentanglement (via a separate encoder-decoder that estimates the illumination image under Retinex theory), and the main image fusion task, which concatenates visible and infrared latent features. Task-specific losses include a Retinex decomposition loss, a histogram-equalization prior loss, a content loss with adaptive weights that directly minimizes L2 distance between the fused image and the corrected-enhanced/infrared images, a structural loss, and an RGB color consistency loss; a growing task weight for the fusion loss balances the cascade during training.

What would settle it

Recompute PSNR, MSE, and CC with the visible image, the infrared image, and both sources (or a no-reference metric) as references on the same test set; if DFVO's margin over two-stage baselines disappears when the reference is not the infrared image, the headline metric is an artifact of the loss design.

Watch

Extended reading notes

Core claim

DFVO's central claim is that cascaded multi-task learning, in which shared latent features feed three task-specific heads trained simultaneously, outperforms the two-stage cascaded training used by existing darkness-free fusion methods. The network extracts common latent features from visible and infrared inputs, reconstructs the infrared image, decomposes the visible image into reflectance and illumination under Retinex theory, and fuses the features in one holistic pass. The authors report state-of-the-art quantitative results on the LLVIP dataset (PSNR 63.258 dB, MSE 0.032, SF 0.082, AG 6.783, CC 0.724), best perceptual preference in a human ranking, and higher YOLOv5 pedestrian precision and recall on fused images, which they attribute to avoiding the information entropy loss of hierarchical data transmission.

Load-bearing premise

The quantitative superiority rests on an unstated choice of reference image for PSNR and MSE; if the reference is the infrared image, the large PSNR partly follows from a loss term that pushes the fused image toward the infrared image.

Editorial extensions

If this is right

  • Night-vision fusion systems for autonomous driving could skip a dedicated low-light enhancement preprocessing stage, cutting runtime and avoiding error accumulation between stages.
  • Because the same shared latent features feed all three tasks, the network is trained end-to-end with no separate enhancement dataset or paired ground-truth illumination, making it applicable where only visible-infrared pairs are available.
  • The fused outputs improve object detection and pedestrian recognition in the paper's experiments, suggesting the all-at-once objective transfers to high-level vision tasks better than two-stage pipelines.
  • A single-stage network trained jointly on reconstruction, enhancement, and fusion generalizes across datasets (MSRS, SMOD, KAIST) without retraining, indicating the approach is not overfit to one data distribution.

Reading between the lines

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

  • The paper never defines what reference image the PSNR/MSE numbers are computed against; given that the content loss directly pushes the fused image toward the infrared image in L2, the headline 63.258 dB may largely measure similarity to the infrared input rather than fusion quality.
  • The same cascaded-multi-task principle could be tested on other modality pairs, such as RGB-depth or RGB-event fusion, where a two-stage restoration-then-fusion pipeline would suffer the same information-loss problem.
  • A straightforward ablation would hold the loss functions fixed and swap the joint training schedule for a sequentially frozen two-stage schedule with identical modules; if the gap persists, the 'all at once' claim is about optimization, not architecture.
  • The HCAM's quadratic attention complexity is acknowledged as a limitation; a linear-attention or windowed variant of the same cross-modal attention would be the natural test of whether the fusion gain survives a lightweight deployment.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes DFVO, a jointly trained network for visible-infrared image fusion under low-illumination conditions, replacing the conventional two-stage enhancement-then-fusion pipeline with a single cascaded multi-task model. The architecture combines a latent-common feature extractor with a details-extraction module (DEM) for high-frequency information and a hyper cross-attention module (HCAM) for low-frequency structural information, alongside task-specific losses for infrared reconstruction, illumination disentanglement, and fusion. Experiments are reported on LLVIP with additional generalization tests on MSRS, SMOD, and KAIST, including qualitative comparisons, nine quantitative metrics, ablation studies, a perceptual MOS study, and detection experiments with YOLOv5. The code is publicly released.

Significance. If the reported results hold, DFVO is a meaningful contribution to low-light infrared-visible fusion: a single-stage formulation that avoids error accumulation across separate enhancement and fusion stages is well motivated, and the design of DEM and HCAM is coherent. The paper provides code, extensive ablations, downstream detection experiments, and cross-dataset generalization, which are concrete strengths. However, the headline quantitative claims are currently undermined by an undefined evaluation reference and by the direct overlap between the training losses and the reported metrics, so the significance cannot be fully assessed from the manuscript as written.

major comments (3)
  1. [Section IV-A3, Tables I-II] The reference image for PSNR and MSE is never defined. In image fusion there is no ground-truth fused image, so PSNR/MSE must be computed against some reference (for example, a source image, an enhanced image, or a constructed ideal); the paper lists PSNR and MSE among the metrics but does not state this choice. This is load-bearing because the Abstract and Section IV-B3 claim the best LLVIP PSNR of 63.258 dB and MSE of 0.032. Moreover, Section IV-A1 states that training images are normalized to [0,1], yet PSNR=63.258 dB with MSE=0.032 is numerically consistent only if MSE is computed on a 0-255 scale. The authors must specify the reference image, the value range, and the exact computation for PSNR/MSE and for each other metric whose reference is ambiguous.
  2. [Eqs. (20), (23), (24) and Table IV] The reported quantitative gains overlap with the training losses in a way that makes the ranking partly by construction. Lcont (Eq. 20) is a weighted L2 penalty between the fused image and the corrected-enhanced and infrared images, which directly minimizes MSE and hence maximizes PSNR if either of those images is the metric reference; Lstr (Eq. 23) penalizes differences in gradient magnitude, directly improving spatial-frequency and average-gradient scores; and Lcos (Eq. 24) maximizes cosine similarity to the enhanced visible image, which would directly improve correlation-based metrics if the reference is the visible/enhanced image. Table IV is consistent with this concern: PSNR drops from 63.268 to 60.321 when Lcont is removed, and SF/AG drop sharply when Lstr is removed. To support the "best performance" claim, the authors should report metrics whose references are explicitly defined and that are not simply the training objectives, or show that the advantage persists under no-reference fusion metrics and on downstream tasks.
  3. [Section IV-A1, Tables I-III] The quantitative ranking is based on point estimates on a fixed test subset. The paper trains on 240 image pairs and reports validation results on 50 "typical" pairs from LLVIP (Section IV-A1); Tables I and II do not state sample size, variance, or significance. Given that some margins over baselines are small (for example, CC 0.724 vs 0.718 in Table I), the "best performance" claims need either the full test set, multiple splits, or statistical tests to rule out selection effects. The same applies to the generalization tables (Table III), which report results on 183, 331, and 25 pairs without error bars.
minor comments (6)
  1. [Section III-B1, Eq. (4)] The notation in Eq. (4) is difficult to parse: the summation indices n do not appear in the summands (e.g., \sum_{n=1}^{C/2} \Theta_t^I), and the channel-splitting notation is unclear; please rewrite the iteration with clearly indexed channel groups.
  2. [Fig. 9 caption and Table VI] The caption reports a "maximum mAP of 0.88 for vehicle recognition," but Table VI lists mAP@.5=0.861 and mAP@.5:.95=0.479 for pedestrian detection; please reconcile the numbers and clarify which detection task and metric are being reported.
  3. [Tables II and IV] The PSNR for the full model is 63.258 dB in Table II but 63.268 dB in Table IV; please unify the numbers.
  4. [Section IV-C2] The text says DFVO "leads in SF, PSNR, and MSR"; MSR is not defined or used elsewhere and appears to be a typo for MSE.
  5. [Section III-C, Eq. (25)] The iteration index n in Eq. (25) is defined only as "the n-th iteration," while training is reported in epochs (Section IV-A2); please clarify whether n counts epochs, iterations within an epoch, or total optimizer steps, since \zeta_3 takes very different values in each case.
  6. [Section IV-A3] The remaining metrics (EN, SF, MI, SD, VIF, AG, CC) are named but not defined; please provide definitions or citations, since the paper's ranking claims depend on them.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline PSNR/CC superiority is partly trained into the model: Eq. (20) directly minimizes MSE to the infrared/enhanced images, and the paper never defines the PSNR/MSE reference, so the reported 63.258 dB is not an independent measure.

  1. other [Eq. (20) in Section III-C; metric list in Section IV-A3]
    "The content loss Lcont forces the network to learn more texture details from corrected-enhanced and infrared images, which can be defined as: Lcont =ωR·∥If− ˆIR∥2 +ωir·∥If−Iir∥2"

    PSNR is a monotone decreasing transform of MSE. Eq. (20) is a weighted MSE between the fused image If and the two training-derived references Ihat_R and Iir. Section IV-A3 lists PSNR/MSE as evaluation metrics but never states the reference image; in fusion there is no ground-truth fused image, so the metric must be computed against some input-derived image. If the reference is Iir or Ihat_R (or their average), the headline 63.258 dB is a direct evaluation of the training loss, not an independent quality measure. The ablation in Table IV reinforces this: removing Lcont lowers PSNR from 63.268 to 60.321 and collapses CC to -0.029.

  2. other [Eqs. (23)-(24) in Section III-C; metrics SF, AG, CC in Section IV-A3]
    "Lstr =∥|∇If|− max(|∇ˆIR|,|∇Iir|)∥2 ... Lcos = 1/HWC Σ_{c∈(R,G,B)} cos(If_RGB, IR_RGB)"

    SF and AG are aggregate gradient-magnitude statistics; Lstr explicitly drives the fused gradient magnitude toward the pixelwise maximum of the source gradients, so the SF/AG margins are a direct consequence of the loss. CC measures similarity to a reference; Lcos maximizes cosine similarity between the fused RGB image and the enhanced visible RGB image, so a CC computed against the visible image is also a training target. These metric gains are therefore partly by construction rather than independent evidence of fusion quality.

full rationale

The architecture derivation itself is not circular: the modules and loss functions are defined from the source images and the network parameters, and the paper does not invoke a load-bearing self-citation or an imported uniqueness theorem. The circularity is in the quantitative evaluation of the central empirical claim. No ground-truth fused image exists, so PSNR/MSE must be evaluated against some input-derived reference; Section IV-A3 never defines which reference is used. Eq. (20) is a weighted MSE against the corrected-enhanced and infrared images, so if PSNR uses either of those references, the headline 63.258 dB is a direct evaluation of the training objective. The same holds for gradient metrics SF/AG versus Eq. (23), and for CC versus Eq. (24). The ablation table confirms that PSNR and CC track the presence of Lcont. Some independent support does exist, including the human MOS experiment and the YOLOv5 detection results, so the paper is not wholly circular; however, the strongest numerical claim, the best PSNR/CC on LLVIP, is partly trained into the model and is presented without the reference definition needed to verify it. This warrants a partial circularity score of 6.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central empirical results depend on a set of hand-set loss weights and thresholds (listed above) and on several domain priors (Retinex, histogram equalization, dark channel prior, information-loss assumption) that are taken from the literature without re-validation on nighttime driving data. No new physical entities are introduced.

free parameters (5)
  • Loss weights for illumination disentanglement (lambda_1..lambda_4) = 500, 1.35, 1.55, 2.5
    Chosen in Section IV-A2 without sensitivity analysis; they determine the balance of Retinex, smoothness, and histogram priors.
  • Loss weights for fusion task (alpha_1..alpha_3) = 1.75, 0.65, 0.35
    Set in Section IV-A2; they control how strongly content, structure, and color losses shape the fused image and the reported metrics.
  • Task importance weights (zeta_1, zeta_2, mu) = 10, 1, 0.1
    Eq. (10) and (25); the fusion task weight grows linearly with iteration. Values are hand-selected.
  • Over-exposure threshold eta and gradient control c = eta=7, c=10, epsilon=0.01
    Used in Eq. (14) and Eq. (18) to strip over-exposed pixels; values are fixed without analysis.
  • Iteration count T in details-extraction module = 3
    Set in Section III-B1; more iterations are not tested.
assumptions (4)
  • domain assumption Retinex decomposition: enhanced reflectance times illumination equals the visible luminance channel (Eq. 13).
    Forms the basis of the illumination disentanglement task; known to be approximate for noisy, colored nighttime images.
  • domain assumption Histogram equalization provides a valid self-supervised target for enhancement (Eq. 15).
    Assumes the histogram-equalized visible image contains the structural and color information needed to train the enhancement branch without ground truth.
  • domain assumption The dark channel prior identifies over-exposed pixels (Eq. 18-19).
    Transfers a dehazing prior to over-exposure detection in nighttime driving scenes; no validation is provided for this transfer.
  • ad hoc to paper Each cascaded processing stage causes information entropy loss (Section I, citing [25]).
    Used to justify replacing two-stage training; this is asserted rather than demonstrated in the setting of image fusion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DFVO: Learning Darkness-free Visible and Infrared Image Disentanglement and Fusion All at Once." pith.science (2026). https://pith.science/paper/CZGJL5BJ

@misc{pith2026250504526,
  author       = {Pith},
  title        = {Pith review of: DFVO: Learning Darkness-free Visible and Infrared Image Disentanglement and Fusion All at Once},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CZGJL5BJ}},
  note         = {Machine review of arXiv:2505.04526}
}
read the original abstract

Visible and infrared image fusion is one of the most crucial tasks in the field of image fusion, aiming to generate fused images with clear structural information and high-quality texture features for high-level vision tasks. However, when faced with severe illumination degradation in visible images, the fusion results of existing image fusion methods often exhibit blurry and dim visual effects, posing major challenges for autonomous driving. To this end, a Darkness-Free network is proposed to handle Visible and infrared image disentanglement and fusion all at Once (DFVO), which employs a cascaded multi-task approach to replace the traditional two-stage cascaded training (enhancement and fusion), addressing the issue of information entropy loss caused by hierarchical data transmission. Specifically, we construct a latent-common feature extractor (LCFE) to obtain latent features for the cascaded tasks strategy. Firstly, a details-extraction module (DEM) is devised to acquire high-frequency semantic information. Secondly, we design a hyper cross-attention module (HCAM) to extract low-frequency information and preserve texture features from source images. Finally, a relevant loss function is designed to guide the holistic network learning, thereby achieving better image fusion. Extensive experiments demonstrate that our proposed approach outperforms state-of-the-art alternatives in terms of qualitative and quantitative evaluations. Particularly, DFVO can generate clearer, more informative, and more evenly illuminated fusion results in the dark environments, achieving best performance on the LLVIP dataset with 63.258 dB PSNR and 0.724 CC, providing more effective information for high-level vision tasks. Our code is publicly accessible at https://github.com/DaVin-Qi530/DFVO.

Figures

Figures reproduced from arXiv: 2505.04526 by the authors.

Figure 1
Figure 1. Existing fusion methods in low-light environments [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall architecture of our method. The parallel cascaded tasks include the infrared image-reconstruction task, illumination disentanglement task, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) The visual results of iteration process in the Details-Extraction [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Vision quality comparison of five SOTA fusion methods on the LLVIP dataset. (a)-(b) Source images. (c) SeAFusion [ [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Human perception comparison of five SOTA fusion methods on the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Vision quality comparison of two-stage fusion methods on the LLVIP dataset. (a)-(b) Source images. (c) Zero-DCE [ [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Vision quality comparison of two-stage fusion methods on the MSRS, SMOD, and KAIST datasets. (a)-(b) Source images. (c) Zero-DCE [ [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Vision quality comparison of the ablation study on important modules [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Detection performance of our fused images with four SOTA fusion [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 34 canonical work pages

  1. [1]

    Stdfusionnet: An infrared and visible image fusion network based on salient target detection,

    J. Ma, L. Tang, M. Xu, H. Zhang, and G. Xiao, “Stdfusionnet: An infrared and visible image fusion network based on salient target detection,” IEEE Transactions on Instrumentation and Measurement , vol. 70, pp. 1–13, 2021

  2. [2]

    Diff-mosaic: augmenting realistic representations in infrared small target detection via 11 diffusion prior,

    Y . Shi, Y . Lin, P. Wei, X. Xian, T. Chen, and L. Lin, “Diff-mosaic: augmenting realistic representations in infrared small target detection via 11 diffusion prior,” IEEE Transactions on Geoscience and Remote Sensing , 2024

  3. [3]

    Sirst-5k: Exploring massive negatives synthesis with self-supervised learning for robust infrared small target detection,

    Y . Lu, Y . Lin, H. Wu, X. Xian, Y . Shi, and L. Lin, “Sirst-5k: Exploring massive negatives synthesis with self-supervised learning for robust infrared small target detection,” IEEE Transactions on Geoscience and Remote Sensing, 2024

  4. [4]

    Yolactfusion: An instance segmentation method for rgb-nir multimodal image fusion based on an attention mechanism,

    C. Liu, Q. Feng, Y . Sun, Y . Li, M. Ru, and L. Xu, “Yolactfusion: An instance segmentation method for rgb-nir multimodal image fusion based on an attention mechanism,” Computers and Electronics in Agriculture , vol. 213, p. 108186, 2023

  5. [5]

    A novel image fusion framework for night- vision navigation and surveillance,

    G. Bhatnagar and Z. Liu, “A novel image fusion framework for night- vision navigation and surveillance,” Signal, Image and Video Processing, vol. 9, pp. 165–175, 2015

  6. [6]

    The nonsubsampled contourlet transform: theory, design, and applications,

    A. L. Da Cunha, J. Zhou, and M. N. Do, “The nonsubsampled contourlet transform: theory, design, and applications,” IEEE transactions on image processing, vol. 15, no. 10, pp. 3089–3101, 2006

  7. [7]

    Crossfuse: Learning infrared and visible image fusion by cross-sensor top-k vision alignment and beyond,

    Y . Shi, C. Shi, Z. Weng, Y . Tian, X. Xian, and L. Lin, “Crossfuse: Learning infrared and visible image fusion by cross-sensor top-k vision alignment and beyond,” IEEE Transactions on Circuits and Systems for Video Technology, 2025

  8. [8]

    Fast saliency-aware multi- modality image fusion,

    J. Han, E. J. Pauwels, and P. De Zeeuw, “Fast saliency-aware multi- modality image fusion,” Neurocomputing, vol. 111, pp. 70–80, 2013

Show all 54 references
  1. [9]

    Multiple feature fusion by subspace learning,

    Y . Fu, L. Cao, G. Guo, and T. S. Huang, “Multiple feature fusion by subspace learning,” in Proceedings of the 2008 international conference on Content-based image and video retrieval , 2008, pp. 127–134

  2. [10]

    Infrared and visible image fusion using joint convolution sparse coding,

    C. Zhang, Z. Yue, D. Yan, and X. Yang, “Infrared and visible image fusion using joint convolution sparse coding,” in 2019 International Conference on Image and Video Processing, and Artificial Intelligence , vol. 11321. SPIE, 2019, pp. 181–189

  3. [11]

    Ufa-fuse: A novel deep supervised and hybrid model for multifocus image fusion,

    Y . Zang, D. Zhou, C. Wang, R. Nie, and Y . Guo, “Ufa-fuse: A novel deep supervised and hybrid model for multifocus image fusion,” IEEE Transactions on Instrumentation and Measurement , vol. 70, pp. 1–17, 2021

  4. [12]

    Ifcnn: A general image fusion framework based on convolutional neural network,

    Y . Zhang, Y . Liu, P. Sun, H. Yan, X. Zhao, and L. Zhang, “Ifcnn: A general image fusion framework based on convolutional neural network,” Information Fusion, vol. 54, pp. 99–118, 2020

  5. [13]

    Infrared and visible image fusion based on variational auto-encoder and infrared feature compensation,

    L. Ren, Z. Pan, J. Cao, and J. Liao, “Infrared and visible image fusion based on variational auto-encoder and infrared feature compensation,” Infrared Physics & Technology , vol. 117, p. 103839, 2021

  6. [14]

    Mrfddgan: Multireceptive field feature transfer and dual discriminator-driven generative adversarial network for infrared and color visible image fusion,

    J. Li, B. Li, Y . Jiang, L. Tian, and W. Cai, “Mrfddgan: Multireceptive field feature transfer and dual discriminator-driven generative adversarial network for infrared and color visible image fusion,” IEEE Transactions on Instrumentation and Measurement , vol. 72, pp. 1–28, 2023

  7. [15]

    Cgtf: Convolution-guided transformer for infrared and visible image fusion,

    J. Li, J. Zhu, C. Li, X. Chen, and B. Yang, “Cgtf: Convolution-guided transformer for infrared and visible image fusion,” IEEE Transactions on Instrumentation and Measurement , vol. 71, pp. 1–14, 2022

  8. [16]

    Swinfuse: A residual swin transformer fusion network for infrared and visible images,

    Z. Wang, Y . Chen, W. Shao, H. Li, and L. Zhang, “Swinfuse: A residual swin transformer fusion network for infrared and visible images,” IEEE Transactions on Instrumentation and Measurement , vol. 71, pp. 1–12, 2022

  9. [17]

    Divfusion: Darkness-free infrared and visible image fusion,

    L. Tang, X. Xiang, H. Zhang, M. Gong, and J. Ma, “Divfusion: Darkness-free infrared and visible image fusion,” Information Fusion , vol. 91, pp. 477–493, 2023

  10. [18]

    Ev-fusion: A novel infrared and low-light color visible image fusion network integrating unsuper- vised visible image enhancement,

    X. Zhang, X. Wang, C. Yan, and Q. Sun, “Ev-fusion: A novel infrared and low-light color visible image fusion network integrating unsuper- vised visible image enhancement,” IEEE Sensors Journal , 2024

  11. [19]

    L2fusion: Low-light oriented infrared and visible image fusion,

    X. Gao, G. Lv, A. Dong, Z. Wei, and J. Cheng, “L2fusion: Low-light oriented infrared and visible image fusion,” in 2023 IEEE International Conference on Image Processing (ICIP) . IEEE, 2023, pp. 2405–2409

  12. [20]

    Lenfusion: A joint low- light enhancement and fusion network for nighttime infrared and visible image fusion,

    J. Chen, L. Yang, W. Liu, X. Tian, and J. Ma, “Lenfusion: A joint low- light enhancement and fusion network for nighttime infrared and visible image fusion,” IEEE Transactions on Instrumentation and Measurement, 2024

  13. [21]

    Enlighten fusion multiscale network for infrared and visible image fusion in dark environments,

    H. Wang, C. Shu, and X. Li, “Enlighten fusion multiscale network for infrared and visible image fusion in dark environments,” IEEE Signal Processing Letters, 2023

  14. [22]

    Crose: Low-light enhancement by cross-sensor interaction for nighttime driving scenes,

    X. Xian, Q. Zhou, J. Qin, X. Yang, Y . Tian, Y . Shi, and D. Tian, “Crose: Low-light enhancement by cross-sensor interaction for nighttime driving scenes,” Expert Systems with Applications , vol. 248, p. 123470, 2024

  15. [23]

    Nitedr: Nighttime image de-raining with cross-view sensor cooperative learning for dynamic driving scenes,

    C. Shi, L. Fang, H. Wu, X. Xian, Y . Shi, and L. Lin, “Nitedr: Nighttime image de-raining with cross-view sensor cooperative learning for dynamic driving scenes,” IEEE Transactions on Multimedia , 2024

  16. [24]

    Reference- free low-light image enhancement by associating hierarchical wavelet representations,

    X. Yang, J. Gong, L. Wu, Z. Yang, Y . Shi, and F. Nie, “Reference- free low-light image enhancement by associating hierarchical wavelet representations,” Expert Systems with Applications , vol. 213, p. 118920, 2023

  17. [25]

    An introduction to information theory and entropy,

    T. Carter, “An introduction to information theory and entropy,” Complex systems summer school, Santa Fe , 2007

  18. [26]

    Densefuse: A fusion approach to infrared and visible images,

    H. Li and X.-J. Wu, “Densefuse: A fusion approach to infrared and visible images,” IEEE Transactions on Image Processing, vol. 28, no. 5, pp. 2614–2623, 2018

  19. [27]

    U2fusion: A unified unsupervised image fusion network,

    H. Xu, J. Ma, J. Jiang, X. Guo, and H. Ling, “U2fusion: A unified unsupervised image fusion network,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 44, no. 1, pp. 502–518, 2020

  20. [28]

    Ganmcc: A generative adversarial network with multiclassification constraints for infrared and visible image fusion,

    J. Ma, H. Zhang, Z. Shao, P. Liang, and H. Xu, “Ganmcc: A generative adversarial network with multiclassification constraints for infrared and visible image fusion,” IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–14, 2020

  21. [29]

    Target-aware dual adversarial learning and a multi-scenario multi- modality benchmark to fuse infrared and visible for object detection,

    J. Liu, X. Fan, Z. Huang, G. Wu, R. Liu, W. Zhong, and Z. Luo, “Target-aware dual adversarial learning and a multi-scenario multi- modality benchmark to fuse infrared and visible for object detection,” in Proceedings of the IEEE/CVF conference on computer vision and pattern re...

  22. [30]

    Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network,

    L. Tang, J. Yuan, and J. Ma, “Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network,” Information Fusion, vol. 82, pp. 28–42, 2022

  23. [31]

    Nice: Non-linear independent components estimation,

    L. Dinh, D. Krueger, and Y . Bengio, “Nice: Non-linear independent components estimation,” arXiv preprint arXiv:1410.8516 , 2014

  24. [32]

    iscmis: Spatial-channel attention based deep invertible network for multi-image steganography,

    F. Li, Y . Sheng, X. Zhang, and C. Qin, “iscmis: Spatial-channel attention based deep invertible network for multi-image steganography,” IEEE Transactions on Multimedia , 2023

  25. [33]

    Enhanced invertible encoding for learned image compression,

    Y . Xie, K. L. Cheng, and Q. Chen, “Enhanced invertible encoding for learned image compression,” in Proceedings of the 29th ACM international conference on multimedia , 2021, pp. 162–170

  26. [34]

    Invertible image rescaling,

    M. Xiao, S. Zheng, C. Liu, Y . Wang, D. He, G. Ke, J. Bian, Z. Lin, and T.-Y . Liu, “Invertible image rescaling,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part I 16 . Springer, 2020, pp. 126–144

  27. [35]

    Guided image generation with conditional invertible neural networks,

    L. Ardizzone, C. L ¨uth, J. Kruse, C. Rother, and U. K¨othe, “Guided image generation with conditional invertible neural networks,” arXiv preprint arXiv:1907.02392, 2019

  28. [36]

    Invertible network for classification and biomarker selection for asd,

    J. Zhuang, N. C. Dvornek, X. Li, P. Ventola, and J. S. Duncan, “Invertible network for classification and biomarker selection for asd,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 201...

  29. [37]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

  30. [38]

    Multigrained attention network for infrared and visible image fusion,

    J. Li, H. Huo, C. Li, R. Wang, C. Sui, and Z. Liu, “Multigrained attention network for infrared and visible image fusion,” IEEE Transactions on Instrumentation and Measurement , vol. 70, pp. 1–12, 2020

  31. [39]

    Defect detection method of aluminum profile surface using deep self-attention mechanism under hybrid noise conditions,

    R. Chen, D. Cai, X. Hu, Z. Zhan, and S. Wang, “Defect detection method of aluminum profile surface using deep self-attention mechanism under hybrid noise conditions,” IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–9, 2021

  32. [40]

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

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

  33. [41]

    Perceive-ir: Learning to perceive degradation better for all-in-one image restoration,

    X. Zhang, J. Ma, G. Wang, Q. Zhang, H. Zhang, and L. Zhang, “Perceive-ir: Learning to perceive degradation better for all-in-one image restoration,” arXiv preprint arXiv:2408.15994 , 2024

  34. [42]

    Transformer-based end-to-end anatomical and functional image fusion,

    J. Zhang, A. Liu, D. Wang, Y . Liu, Z. J. Wang, and X. Chen, “Transformer-based end-to-end anatomical and functional image fusion,” IEEE Transactions on Instrumentation and Measurement , vol. 71, pp. 1–11, 2022

  35. [43]

    Transcending pixels: boosting saliency detection via scene understanding from aerial imagery,

    Y . Liu, Z. Xiong, Y . Yuan, and Q. Wang, “Transcending pixels: boosting saliency detection via scene understanding from aerial imagery,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–16, 2023

  36. [44]

    Vision transformer with enhanced self-attention for few shot ship target recognition in complex environments,

    Y . Tian, H. Meng, F. Yuan, Y . Ling, and N. Yuan, “Vision transformer with enhanced self-attention for few shot ship target recognition in complex environments,” IEEE Transactions on Instrumentation and Measurement, 2023

  37. [45]

    Lightness and retinex theory,

    E. H. Land and J. J. McCann, “Lightness and retinex theory,” Josa, vol. 61, no. 1, pp. 1–11, 1971

  38. [46]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 4510–4520. 12

  39. [47]

    Kindling the darkness: A practical low-light image enhancer,

    Y . Zhang, J. Zhang, and X. Guo, “Kindling the darkness: A practical low-light image enhancer,” inProceedings of the 27th ACM international conference on multimedia , 2019, pp. 1632–1640

  40. [48]

    Deep retinex decomposition for low-light enhancement,

    C. Wei, W. Wang, W. Yang, and J. Liu, “Deep retinex decomposition for low-light enhancement,” arXiv preprint arXiv:1808.04560 , 2018

  41. [49]

    Self-supervised low- light image enhancement via histogram equalization prior,

    F. Zhang, Y . Shao, Y . Sun, C. Gao, and N. Sang, “Self-supervised low- light image enhancement via histogram equalization prior,” in Chinese Conference on Pattern Recognition and Computer Vision (PRCV) . Springer, 2023, pp. 63–75

  42. [50]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

  43. [51]

    Single image haze removal using dark channel prior,

    K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” IEEE transactions on pattern analysis and machine intelligence, vol. 33, no. 12, pp. 2341–2353, 2010

  44. [52]

    Llvip: A visible-infrared paired dataset for low-light vision,

    X. Jia, C. Zhu, M. Li, W. Tang, and W. Zhou, “Llvip: A visible-infrared paired dataset for low-light vision,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 3496–3504

  45. [53]

    Zero- reference deep curve estimation for low-light image enhancement,

    C. Guo, C. Li, J. Guo, C. C. Loy, J. Hou, S. Kwong, and R. Cong, “Zero- reference deep curve estimation for low-light image enhancement,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 1780–1789

  46. [54]

    Toward fast, flexible, and robust low-light image enhancement,

    L. Ma, T. Ma, R. Liu, X. Fan, and Z. Luo, “Toward fast, flexible, and robust low-light image enhancement,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5637– 5646

Pith tools

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