Pith. sign in

REVIEW 3 major objections 4 minor 64 references

Coarse-to-fine Framework for Generative MEF via Implicit Neural Representation

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read LIIFusion claims generative multi-exposure fusion can run at one-quarter the cost by fusing low-res with a diffusion prior, then refining full-res details with an implicit neural function.

desk verdict The speed-up is real and the architecture is sensible, but the structural-fidelity claim doesn't survive contact with the paper's own numbers, and there's a train/test mismatch in the fine stage that needs airing. read the letter →

arxiv 2607.17611 v1 pith:DLTWNPL4 submitted 2026-07-20 cs.CV cs.AI

classification cs.CVcs.AI
keywords multi-exposurefusiongenerativeMEFdiffusionmodelimplicitneuralrepresentationLIIFcoarse-to-fineframeworkadaptiveexposurecorrectioncomputationalphotography
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 generative multi-exposure fusion (MEF) practical for high-resolution images. It argues that the expensive parts of diffusion-based fusion — patch-wise sampling at fixed resolution and unstable synthesis in saturated regions — can be separated from the resolution problem. Its two-stage design runs the diffusion prior once on a downsampled pair, then uses a local implicit image function, conditioned on the original exposures and the coarse output, to reconstruct every output pixel. The reported result is up to a 3.5× speed-up over prior generative MEF with equal or better structural and perceptual quality, plus better handling of saturated regions via an adaptive exposure correction. If correct, this makes generative MEF competitive in speed with conventional fusion while keeping its deghosting and hallucination strengths.

What carries the argument

The load-bearing component is the Local Implicit Image Function (LIIF) repurposed for MEF. Instead of predicting a fixed pixel grid, it represents the output as a continuous function f(z_coarse, z_fine, [x,c]) — bilinearly interpolated features at query coordinate x, plus a cell size c — so a single MLP can produce pixel values at any resolution. The coarse stage supplies global generative structure; two lightweight encoders supply local evidence from the high-resolution exposures; training with an L1 loss against fused ground truth forces the function to fuse, not upscale. The adaptive exposure correction (a multiplicative, per-pixel gamma-weighted mask built from the luminance difference b

What would settle it

Take a scene with a heavily saturated sign, run LIIFusion, and also run a version whose coarse output has been corrupted by blurring that sign. If the final images are nearly identical in the sign region, the fine stage is not actually restoring structure from the HR sources. Alternatively, compare MEF-SSIM on saturated crops between coarse and final outputs: if the fine stage never raises structural fidelity there, the paper's central fix for saturation has no measurable effect.

Watch

Extended reading notes

Core claim

The central claim is that generative MEF does not need to be done at the output resolution. LIIFusion performs the diffusion-based fusion at a low resolution — near the model's native scale — obtaining a globally coherent coarse fused image, and then delegates high-resolution detail recovery to a resolution-agnostic implicit function. The fine stage is not a super-resolution head: it is trained as a multi-exposure fusion function, taking local features from the coarse output and from the aligned high-resolution over/under-exposed inputs, and fusing them at arbitrary query coordinates. The adaptive exposure correction attenuates saturated regions before the diffusion pass, so the coarse stage

Load-bearing premise

The fine INR stage can only refine, not reinterpret: if the low-resolution diffusion pass loses or hallucinates structure in saturated regions, the final output inherits that error.

Editorial extensions

If this is right

  • Generative MEF can be deployed on high-resolution images in minutes instead of hours, closing the speed gap with conventional methods.
  • The fine stage is resolution-agnostic, so sensors with different output sizes can use the same learned fusion function without retraining.
  • Training does not require expensive human-tuned ground truth: pseudo labels from a strong generative model are nearly as good for the fine stage.
  • Saturated regions inherit structure from the corrected coarse output, so the typical generative failure mode (hallucinated or blurred bright detail) is reduced.
  • Patch-wise sampling and boundary inconsistencies disappear because the diffusion pass runs once over the whole downsampled image.

Reading between the lines

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

  • The same coarse-to-fine split could apply to other generative restoration tasks (e.g., deblurring, low-light enhancement): use a diffusion prior globally at low resolution, then an INR to re-inject original high-resolution cues.
  • Because the fine stage is coordinate-based, it could be queried non-uniformly — e.g., spending more samples on saturated or high-texture regions — a cheap testable extension the paper does not explore.
  • The trade-off curve in Table 10 suggests that if a faster generative backbone replaces the diffusion model, the coarse stage could be pushed to higher resolution, potentially improving quality further.
  • The reliance on the coarse stage's structural reliability implies that the benefit of the fine stage is capped by the diffusion prior; a user interested in extreme saturation should treat the coarse output as the error floor.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes LIIFusion, a coarse-to-fine framework for generative multi-exposure fusion (MEF). The coarse stage runs the diffusion-based UltraFusion at low resolution, augmented by an adaptive exposure correction (AEC) module that attenuates saturated regions. The fine stage adapts the local implicit image function (LIIF) into a multi-exposure fusion function conditioned on the high-resolution OE/UE inputs and the coarse output, enabling resolution-agnostic coordinate-wise refinement. The paper claims up to a 3.5x speed-up over existing generative MEF methods while maintaining or improving structural fidelity and perceptual quality, and reports experiments on MEFB, RealHDRV, and the UltraFusion Benchmark, including ablations and a small human preference study.

Significance. If the quality claims hold, the work is practically significant: it reduces the cost of generative MEF from hours to tens of minutes and is, to my knowledge, the first INR-based MEF formulation. The speed-up is well evidenced by the reported runtimes (27 vs 101 min on RealHDRV; 59 vs 203 min on UltraFusion Benchmark). However, the structural-fidelity claim is not established by the reported metrics, and there is a training/inference distribution mismatch in the fine stage that threatens the central quality argument. The extensive ablation suite, including the pseudo-label experiment and exposure-input ablations, is a strength, but the key comparison to the baseline is undermined by the ambiguity described below.

major comments (3)
  1. [§4.1, Eq. (4), §3.2] The fine-stage training data construction is ambiguous and likely creates a train/test mismatch. The text says: 'SICE label data is used as ground truth, and we downsample them to random scale to obtain corresponding triplet {I_oe, I_ue, I_LR_mef} for training.' The natural reading is that I_LR_mef is the downsampled ground-truth image, not the actual low-resolution output of the frozen UltraFusion coarse stage used at inference (Eq. 2). At inference, the fine stage receives a generative coarse output that can be structurally unreliable in saturated regions, as the paper itself concedes in Sec. 5. The pseudo-label experiment (Tab. 7) replaces only the HR target with UltraFusion outputs, not the LR coarse input, so it does not close this gap. This distribution shift is load-bearing: the claimed ability of the INR to 'refine' the coarse output is only valid if the fine stage has seen reali
  2. [Abstract, Table 3, Table 2] The claim of 'maintaining or improving structural fidelity' is not supported by the evidence. The only structural metric reported, MEF-SSIM on MEFB (Tab. 3), is lower for LIIFusion (0.9201) than for UltraFusion (0.9266). On the dynamic benchmarks (Tab. 2), no structural metric is reported at all; all four metrics are no-reference perceptual measures. The NRIQA gains over UltraFusion are small (e.g., MUSIQ +1.98/+1.79, DeQA-Score -0.0091/-0.0316) and are presented without error bars or significance tests. Therefore the 'maintaining structural fidelity' part of the central claim is not established, even though the speed advantage is clear. The authors should either report MEF-SSIM on the dynamic benchmarks, add statistical significance/error bars, or soften the claim.
  3. [Supplementary §A.3, Table C vs main Table 3] There is an internal tension between the AEC ablation and the final structural result. In Supplementary Table C, applying AEC with α=0.2 improves MEF-SSIM to 0.9313 on low-resolution MEFB outputs, compared to 0.9266 for α=0.0 (which corresponds to the original UltraFusion behavior). Yet the final high-resolution MEF-SSIM in main Table 3 is 0.9201, below UltraFusion's 0.9266. If AEC genuinely helps the coarse stage, the fine stage appears to lose those structural gains, or the low-resolution structural metric is not predictive of final HR structure. This should be reconciled, especially since the fine stage is claimed to preserve and refine structure.
minor comments (4)
  1. [§4.4] The sentence 'All ablation studies except Sec. 4.4 are conducted on the UltraFusion Benchmark' is confusing because Sec. 4.4 is the ablation section itself; presumably one of the ablations is on MEFB (Sec. 4.4, Tab. 6) and the supplementary AEC ablation is on MEFB. Please clarify.
  2. [Fig. 2 caption] 'minimizing diffusion usage to up to 2 passes' is unclear: is this two diffusion forward passes total, or two per patch? The implementation details do not specify the number of inference passes at the chosen coarse resolution.
  3. [§3.3, Eq. (6)] The AEC weight map W=(1−α·L_diff)^{1/2.2} is introduced as a perceptual gamma correction, but the exponent and the clipping behavior are heuristic. The sensitivity analysis in Supplementary Table C is useful, but a brief motivation for the 1/2.2 exponent would improve reproducibility.
  4. [Table 4] The TFLOPs for UltraFusion (944.4) and LIIFusion (282.1) are for a single image of 1988×1326, while the abstract claims up to 3.5x speed-up. The factor is 3.35x for TFLOPs and 3.44x for runtime on the UltraFusion Benchmark; please clarify which ratio supports the '3.5x' statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central speed/fidelity claim is an engineering comparison against an external baseline, with no fitted quantity reused as a prediction.

full rationale

LIIFusion's derivation chain is not circular. The coarse stage is a frozen external diffusion model (UltraFusion, ref [8], not authored by the present team), and the fine INR stage is trained with L1 loss against SICE ground-truth labels (Sec. 4.1, Eq. 4), not against the baseline being compared. The speed-up claim is a runtime measurement (Tables 2 and 4), and the fidelity claim is evaluated on external benchmarks (MEFB, RealHDRV, UltraFusion Benchmark) with metrics defined independently (MEF-SSIM, MUSIQ, etc.). The pseudo-label experiment (Table 7, Sec. A.1) is explicitly a scalability/ablation study, not the main evidence, and it compares two training-supervision variants rather than claiming a prediction. The paper's own limitation statement — 'the final output is still influenced by the structural reliability of the coarse generative stage' (Sec. 5) — is a correctness caveat, not evidence that the output reduces to an input. The closest issue is a train/test mismatch: Sec. 4.1 describes I_LR_mef as downsampled SICE labels during training, while at inference it is the coarse generative output; this is a distribution-shift/validity concern, not a logical circle, and it is explicitly acknowledged. Hyperparameter alpha is selected on MEFB (Table C), which is a benchmarking concern, not circularity. No self-citation chain, uniqueness import, or ansatz-via-citation was found.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The paper's contribution is mainly architectural: it imports UltraFusion's diffusion prior and LIIF's implicit decoding, and adds a hand-designed AEC. The main hand-tuned free parameters are α, the coarse resolution, and the fine-encoder patch size. The AEC formula is an ad hoc heuristic. No new physical or mathematical entities are introduced.

free parameters (3)
  • α (AEC strength) = 0.2
    Controls attenuation of saturated OE pixels via W=(1-α·L_diff)^{1/2.2}; selected by sweeping on MEFB (Supp. Tab. C), not derived.
  • Coarse fusion resolution = 768×768 (used in main results)
    Tab. 10 evaluates 256/512/768/1024 and the main pipeline uses 768×768; chosen as a practical efficiency-quality tradeoff on RealHDRV.
  • Fine encoder patch size = 19×19
    Architecture choice for E_fine, not theoretically motivated; could affect HR detail recovery and is not ablated.
assumptions (6)
  • standard math LIIF's local implicit function formulation (Eq. 1): an MLP can predict RGB at arbitrary coordinates given interpolated local features and cell size.
    Assumed from Chen et al. [7]; not re-derived.
  • standard math RAFT optical flow and backward warping provide accurate alignment between exposure pairs.
    Used as a black box in coarse and fine stages; errors propagate if flow is inaccurate.
  • domain assumption MEF can be treated as guided inpainting in the latent space of a frozen diffusion model (UltraFusion).
    The coarse stage inherits this formulation from [8].
  • domain assumption A low-resolution diffusion output plus HR exposure-conditioned INR can recover full-resolution detail.
    Central architectural premise; if coarse output is structurally unreliable, the fine stage cannot fix it (Sec. 5).
  • domain assumption NRIQA metrics (MUSIQ, PAQ2PIQ, HyperIQA, DeQA-Score) are valid proxies for fused-image quality.
    Used as primary evaluation; no full-reference metrics or significance tests.
  • ad hoc to paper AEC multiplicative weight W=(1-α·L_diff)^{1/2.2} attenuates saturated regions; the clipping and gamma exponent are heuristics.
    Not derived; α and exponent are chosen by hand.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Coarse-to-fine Framework for Generative MEF via Implicit Neural Representation." pith.science (2026). https://pith.science/paper/DLTWNPL4

@misc{pith2026260717611,
  author       = {Pith},
  title        = {Pith review of: Coarse-to-fine Framework for Generative MEF via Implicit Neural Representation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DLTWNPL4}},
  note         = {Machine review of arXiv:2607.17611}
}
abstract

Multi-exposure fusion (MEF) expands the luminance range beyond what a single exposure can capture. Combining images taken at different exposure levels requires handling geometric differences while naturally merging their complementary brightness information. It often demands generative completion where details are missing. Diffusion-based generative methods address these challenges, however, they are computationally expensive and struggle to preserve fine structures in saturated regions. We propose LIIFusion, a coarse-to-fine framework that balances fusion quality and efficiency in generative MEF. The coarse stage performs low resolution generative fusion, enhanced by an adaptive exposure correction that recovers structure lost in saturated over-exposed areas. The fine stage adapts a local implicit image function into a multi-exposure fusion function: conditioned on the HR OE/UE sources and the coarse output, it queries arbitrary target coordinates and fuses source evidence regardless of the HR input resolution. LIIFusion achieves up to 3.5$\times$ speed-up over existing generative methods while maintaining or improving structural fidelity and perceptual quality. We believe this framework provides an effective pathway toward making generative MEF more practical in real-world applications.

Figures

Figures reproduced from arXiv: 2607.17611 by the authors.

Figure 1
Figure 1. Our coarse-to-fine generative MEF framework preserves the natural synthesis of generative MEF (color tone and reduced ghosting) while achieving detailed structural preservation comparable to conventional MEF. It is about 3.5× faster than the previous generative MEF method (Ultrafusion [8]) and runs close to conventional MEF (HSDS￾MEF [47]); speed is measured on a single NVIDIA RTX A5000 with a 1228 × 1638- resolutio… view at source ↗
Figure 2
Figure 2. Overview of LIIFusion pipeline. (a) Previous generative MEF models oper￾ate in a patch-wise manner, requiring many sampling passes. (b) Our proposed coarse￾to-fine framework, LIIFusion, performs low-resolution generative fusion with adaptive exposure correction to obtain a structurally reliable result, minimizing diffusion usage to up to 2 passes depending on the image resolution ratio, followed by a LIIF-based fine… view at source ↗
Figure 3
Figure 3. Qualitative comparison on ultra HDR scenes in the UltraFusion Benchmark. Compared to prior MEF models, our approach achieves more stable exposure fusion, retaining structural detail and natural color appearance in both day and night scenes. MEF-LUT HSDS-MEF UltraFusion Ours UltraFusion Ours MEF-LUT HSDS-MEF OE UE Zoom-In OE UE Zoom-In [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Qualitative results on the RealHDRV dataset, showing that our method pro￾vides cleaner fusion with improved guided inpainting, robust occlusion handling, and reduced ghosting. operates on patch-wise latent inpainting, its contextual range is limited, some￾times leading…
Figure 5
Figure 5. Figure 5: Visual ablation of the adaptive ex￾posure correction module on the UltraFu￾sion Benchmark at low resolution, show￾ing improved reconstruction of details lost in saturated OE regions [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison between the pre-trained LIIF and our LIIFusion on the UltraFusion Benchmark. Our LIIFusion restores finer textual and structural details, such as characters and patterns, which are blurred or lost in the pre-trained LIIF outputs. As shown in Tab. 6, t…
Figure 7
Figure 7. Figure 7: Qualitative comparison between the original SICE labels and our gener￾ated pseudo labels [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Comparison of different exposure￾conditioning strategies for LIIF. Using both OE and UE exposures preserves fine details and reduces residual artifacts com￾pared with configurations using only one exposure or none. annotations. This also suggests that stronger generati…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 7 linked inside Pith

  1. [1]

    In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition Workshops

    Agustsson, E., Timofte, R.: Ntire 2017 challenge on single image super-resolution: Dataset and study. In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition Workshops. pp. 126–135 (2017)

  2. [2]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Barron, J.T., Mildenhall, B., Tancik, M., Hedman, P., Martin-Brualla, R., Srini- vasan, P.P.: Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 5855–5864 (2021)

  3. [3]

    Computer Graphics Forum (Proceedings of Eurographics 2025)44(2) (2025), arXiv:2405.14304

    Bemana, M., Leimkuehler, T., Myszkowski, K., Seidel, H.P., Ritschel, T.: Bracket diffusion: Hdr image generation by consistent ldr denoising. Computer Graphics Forum (Proceedings of Eurographics 2025)44(2) (2025), arXiv:2405.14304

  4. [4]

    IEEE transactions on image processing27(4), 2049–2062 (2018)

    Cai, J., Gu, S., Zhang, L.: Learning a deep single image contrast enhancer from multi-exposure images. IEEE transactions on image processing27(4), 2049–2062 (2018)

  5. [5]

    In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition

    Cao, J., Wang, Q., Xian, Y., Li, Y., Ni, B., Pi, Z., Zhang, K., Zhang, Y., Timofte, R., Van Gool, L.: Ciaosr: Continuous implicit attention-in-attention network for arbitrary-scale image super-resolution. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition. pp. 1796–1807 (2023)

  6. [6]

    IEEE Transactions on Image Processing31, 2661–2672 (2022)

    Chen, J., Yang, Z., Chan, T.N., Li, H., Hou, J., Chau, L.P.: Attention-guided progressive neural texture fusion for high dynamic range image restoration. IEEE Transactions on Image Processing31, 2661–2672 (2022)

  7. [7]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Chen, Y., Liu, S., Wang, X.: Learning continuous image representation with local implicit image function. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 8628–8638 (2021)

  8. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Chen, Z., Wang, Y., Cai, X., You, Z., Lu, Z., Zhang, F., Guo, S., Xue, T.: Ultra- fusion: Ultra high dynamic imaging using exposure fusion. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 16111– 16121 (2025)

Show all 64 references
  1. [9]

    In: European Confer- ence on Computer Vision

    Chobola, T., Liu, Y., Zhang, H., Schnabel, J.A., Peng, T.: Fast context-based low- light image enhancement via neural implicit representations. In: European Confer- ence on Computer Vision. p. 413–430 (2024), arXiv:2407.12511

  2. [10]

    IEEE transactions on pattern analysis and machine intelligence 38(2), 295–307 (2015)

    Dong, C., Loy, C.C., He, K., Tang, X.: Image super-resolution using deep convo- lutional networks. IEEE transactions on pattern analysis and machine intelligence 38(2), 295–307 (2015)

  3. [11]

    In: European Conference on Computer Vision

    Dong, C., Loy, C.C., Tang, X.: Accelerating the super-resolution convolutional neu- ral network. In: European Conference on Computer Vision. pp. 391–407. Springer (2016)

  4. [12]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Fang, W., Tang, Y., Guo, H., Yuan, M., Mok, T.C.W., Yan, K., Yao, J., Chen, X., Liu, Z., Lu, L., Zhang, L., Xu, M.: Cycleinr: Cycle implicit neural representation for arbitrary-scale volumetric super-resolution of medical data. In: Proceedings of the IEEE/CVF Conference on Com...

  5. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Feng, Y., Shang, Y., Li, X., Shao, T., Jiang, C., Yang, Y.: Pie-nerf: Physics-based interactive elastodynamics with nerf. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4450–4461 (2024)

  6. [14]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Jiang, H., Guan, B., Liu, Z., Liu, X., Yu, J., Liu, Z., Han, S., Liu, S.: Learning to see in the extremely dark. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 7676–7685 (2025) Coarse-to-fine Framework for Generative MEF via INR 9

  7. [15]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Jiang, T., Wang, C., Li, X., Li, R., Fan, H., Liu, S.: Meflut: Unsupervised 1d lookup tables for multi-exposure image fusion. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 10542–10551 (2023)

  8. [16]

    ACM Transactions on Graphics (TOG)36(4), 144:1–144:12 (2017)

    Kalantari, N.K., Ramamoorthi, R.: Deep high dynamic range imaging of dynamic scenes. ACM Transactions on Graphics (TOG)36(4), 144:1–144:12 (2017)

  9. [17]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Ke, J., Wang, Q., Wang, Y., Milanfar, P., Yang, F.: Musiq: Multi-scale image quality transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 5148–5157 (2021)

  10. [18]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kim, J., Kim, T.K.: Arbitrary-scale image generation and upsampling using latent diffusion model and implicit neural decoder. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9202–9211 (2024)

  11. [19]

    arXiv preprint arXiv:2210.07762 (2022)

    Kim, S., Min, Y., Jung, Y., Kim, S.: Controllable style transfer via test-time train- ing of implicit neural representation. arXiv preprint arXiv:2210.07762 (2022)

  12. [20]

    In: European Conference on Computer Vision

    Kong, L., Li, B., Xiong, Y., Zhang, H., Gu, H., Chen, J.: Safnet: Selective alignment fusion network for efficient hdr imaging. In: European Conference on Computer Vision. pp. 256–273. Springer (2024)

  13. [21]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Lee, J., Jin, K.H.: Local texture estimator for implicit representation function. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 1929–1938 (2022)

  14. [22]

    IEEE Transactions on Image Processing29, 5805– 5816 (2020)

    Li, H., Ma, K., Yong, H., Zhang, L.: Fast multiscale structural patch decomposition for multi-exposure image fusion. IEEE Transactions on Image Processing29, 5805– 5816 (2020)

  15. [23]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025), arXiv:2506.23537

    Li, X., Ni, Z., Yang, W.: Afunet: Cross-iterative alignment–fusion synergy for hdr reconstruction via deep unfolding paradigm. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025), arXiv:2506.23537

  16. [24]

    IEEE Transactions on Image Processing26(3), 1243–1252 (2017)

    Li, Z., Wei, Z., Wen, C., Zheng, J.: Detail-enhanced multi-scale exposure fusion. IEEE Transactions on Image Processing26(3), 1243–1252 (2017)

  17. [25]

    Liang, J., Cao, J., Sun, G., Zhang, K., Van Gool, L., Timofte, R.: Swinir: Image restorationusingswintransformer.In:ProceedingsoftheIEEE/CVFInternational Conference on Computer Vision. pp. 1833–1844 (2021)

  18. [26]

    In: European Conference on Computer Vision

    Liang, P., Jiang, J., Liu, X., Ma, J.: Fusion from decomposition: A self-supervised decomposition approach for image fusion. In: European Conference on Computer Vision. pp. 719–735. Springer (2022)

  19. [27]

    In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition workshops

    Lim,B.,Son,S.,Kim,H.,Nah,S.,MuLee,K.:Enhanceddeepresidualnetworksfor single image super-resolution. In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition workshops. pp. 136–144 (2017)

  20. [28]

    In: European Conference on Computer Vision

    Liu, Z., Wang, Y., Zeng, B., Liu, S.: Ghost-free high dynamic range imaging with context-aware transformer. In: European Conference on Computer Vision. pp. 344–

  21. [29]

    IEEE Transactions on Image Processing29, 2808– 2819 (2019)

    Ma, K., Duanmu, Z., Zhu, H., Fang, Y., Wang, Z.: Deep guided learning for fast multi-exposure image fusion. IEEE Transactions on Image Processing29, 2808– 2819 (2019)

  22. [30]

    IEEE Transactions on Image Processing26(5), 2519–2532 (2017)

    Ma, K., Li, H., Yong, H., Wang, Z., Meng, D., Zhang, L.: Robust multi-exposure image fusion: A structural patch decomposition approach. IEEE Transactions on Image Processing26(5), 2519–2532 (2017)

  23. [31]

    IEEE Transactions on Image Processing24(11), 3345–3356 (2015)

    Ma, K., Zeng, K., Wang, Z.: Perceptual quality assessment for multi-exposure image fusion. IEEE Transactions on Image Processing24(11), 3345–3356 (2015)

  24. [32]

    In: Pacific Conference on Computer Graphics and Applications (PG’07)

    Mertens, T., Kautz, J., Reeth, F.V.: Exposure fusion. In: Pacific Conference on Computer Graphics and Applications (PG’07). pp. 382–390 (2007) 10 S.Han et al

  25. [33]

    In: Eu- ropean Conference on Computer Vision

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: Eu- ropean Conference on Computer Vision. pp. 405–421 (2020)

  26. [34]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Peng, S., Dong, J., Wang, Q., Zhang, S., Shuai, Q., Zhou, X., Bao, H.: Animatable neural radiance fields for modeling dynamic human bodies. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 14314–14323 (2021)

  27. [35]

    In: European Con- ference on Computer Vision

    Prabhakar, K.R., Agrawal, S., Singh, D.K., Ashwath, B., Babu, R.V.: Towards practical and efficient high-resolution hdr deghosting with cnn. In: European Con- ference on Computer Vision. pp. 497–513 (2020)

  28. [36]

    In: IEEE International Conference on Computational Photography

    Prabhakar, K.R., Arora, R., Swaminathan, A., Singh, K.P., Babu, R.V.: A fast, scalable, and reliable deghosting method for extreme exposure fusion. In: IEEE International Conference on Computational Photography. pp. 1–8 (2019)

  29. [37]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Pumarola, A., Corona, E., Pons-Moll, G., Moreno-Noguer, F.: D-nerf: Neural ra- diance fields for dynamic scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10318–10327 (2021)

  30. [38]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Ram Prabhakar, K., Sai Srikar, V., Venkatesh Babu, R.: Deepfuse: A deep un- supervised approach for exposure fusion with extreme exposure image pairs. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 4714–4722 (2017)

  31. [39]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Sabour, S., Vora, S., Duckworth, D., Krasin, I., Fleet, D.J., Tagliasacchi, A.: Ro- bustnerf: Ignoring distractors with robust losses. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20626–20636 (2023)

  32. [40]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Shi, K., Zhou, X., Gu, S.: Improved implicit neural representation with fourier reparameterized training. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 25985–25994 (2024)

  33. [41]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Shu, Y., Shen, L., Hu, X., Li, M., Zhou, Z.: Towards real-world hdr video recon- struction: A large-scale benchmark dataset and a two-stage alignment network. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2879–2888. IEEE (2024), rea...

  34. [42]

    In: European Conference on Computer Vision

    Song, J.W., Park, Y.I., Kong, K., Kwak, J., Kang, S.J.: Selective transhdr: Transformer-based selective hdr imaging using ghost region mask. In: European Conference on Computer Vision. pp. 288–304 (2022)

  35. [43]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Su, S., Yan, Q., Zhu, Y., Zhang, C., Ge, X., Sun, J., Zhang, Y.: Blindly assess image quality in the wild guided by a self-adaptive hyper network. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3667–3676 (2020)

  36. [44]

    In: European Conference on Computer Vision

    Teed, Z., Deng, J.: Raft: Recurrent all-pairs field transforms for optical flow. In: European Conference on Computer Vision. pp. 402–419. Springer (2020)

  37. [45]

    arXiv preprint arXiv:2305.18135 (2023)

    Tel, S., Wu, Z., Zhang, Y., Heyrman, B., Demonceaux, C., Timofte, R., Ginhac, D.: Alignment-free hdr deghosting with semantics consistent transformer. arXiv preprint arXiv:2305.18135 (2023)

  38. [46]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wang, C., Xia, Z., Leimkuehler, T., Myszkowski, K., Zhang, X.: Lediff: La- tent exposure diffusion for hdr generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 453–464 (2025), arXiv:2412.14456

  39. [47]

    In: Pro- ceedings of the AAAI conference on artificial intelligence

    Wu, G., Fu, H., Liu, J., Ma, L., Fan, X., Liu, R.: Hybrid-supervised dual-search: Leveraging automatic learning for loss-free multi-exposure image fusion. In: Pro- ceedings of the AAAI conference on artificial intelligence. pp. 5985–5993 (2024)

  40. [48]

    In: European Conference on Computer Vision

    Wu, S., Xu, J., Tai, Y.W., Tang, C.K.: Deep high dynamic range imaging with large foreground motions. In: European Conference on Computer Vision. pp. 117–132 (2018) Coarse-to-fine Framework for Generative MEF via INR 11

  41. [49]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Xu, G., Wang, Y., Gu, J., Xue, T., Yang, X.: Hdrflow: Real-time hdr video re- construction with large motions. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 24851–24860 (2024)

  42. [50]

    IEEE transactions on pattern analysis and machine intelligence 44(1), 502–518 (2020)

    Xu, H., Ma, J., Jiang, J., Guo, X., Ling, H.: U2fusion: A unified unsupervised image fusion network. IEEE transactions on pattern analysis and machine intelligence 44(1), 502–518 (2020)

  43. [51]

    IEEE Transactions on Image Processing29, 7203–7216 (2020)

    Xu, H., Ma, J., Zhang, X.P.: Mef-gan: Multi-exposure image fusion via genera- tive adversarial networks. IEEE Transactions on Image Processing29, 7203–7216 (2020)

  44. [52]

    International Journal of Computer Vision127(8), 1106–1125 (2019)

    Xue, T., Chen, B., Wu, J., Wei, D., Freeman, W.T.: Video enhancement with task-oriented flow. International Journal of Computer Vision127(8), 1106–1125 (2019)

  45. [53]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yan, Q., Gong, D., Shi, Q., van den Hengel, A., Shen, C., Reid, I., Zhang, Y.: Attention-guided network for ghost-free high dynamic range imaging. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 1751–1760 (2019)

  46. [54]

    IEEE Transactions on Image Processing29, 4308–4322 (2020)

    Yan, Q., Zhang, L., Liu, Y., Zhu, Y., Sun, J., Shi, Q., Zhang, Y.: Deep hdr imaging via a non-local network. IEEE Transactions on Image Processing29, 4308–4322 (2020)

  47. [55]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Ying, Z., Niu, H., Gupta, P., Mahajan, D., Ghadiyaram, D., Bovik, A.: From patches to pictures (paq-2-piq): Mapping the perceptual space of picture quality. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3575–3585 (2020)

  48. [56]

    You, Z., Cai, X., Gu, J., Xue, T., Dong, C.: Teaching large language models to regressaccurateimagequalityscoresusingscoredistribution.In:Proceedingsofthe IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 14483– 14494 (2025)

  49. [57]

    In: European Conference on Computer Vision, pp

    Yuan, W., Zhu, Q., Liu, X., Ding, Y., Zhang, H., Zhang, C.: Sobolev training for implicit neural representations with approximated image derivatives. In: European Conference on Computer Vision, pp. 72–88. Springer (2022)

  50. [58]

    Information Fusion74, 111–131 (2021), mEFB dataset

    Zhang, X.: Benchmarking and comparing multi-exposure image fusion algorithms. Information Fusion74, 111–131 (2021), mEFB dataset

  51. [59]

    Zhang, Y., Tian, Y., Kong, Y., Zhong, B., Fu, Y.: Residual dense network for imagesuper-resolution.In:ProceedingsoftheIEEE/CVFConferenceonComputer Vision and Pattern Recognition. pp. 2472–2481 (2018)

  52. [60]

    arXiv preprint arXiv:2402.02235 (2024)

    Zhao, Z., Deng, L., Bai, H., Cui, Y., Zhang, Z., Zhang, Y., Qin, H., Chen, D., Zhang, J., Wang, P., et al.: Image fusion via vision-language model. arXiv preprint arXiv:2402.02235 (2024)

  53. [61]

    In: IEEE International Conference on Image Processing

    Zheng, J., Li, Z.: Superpixel based patch match for differently exposed images with moving objects and camera movements. In: IEEE International Conference on Image Processing. pp. 4516–4520 (2015)

  54. [62]

    IEEE Transactions on Image Processing22(12), 5190–5201 (2013)

    Zheng, J., Li, Z., Zhu, Z., Wu, S., Rahardja, S.: Hybrid patching for a sequence of differently exposed images with moving objects. IEEE Transactions on Image Processing22(12), 5190–5201 (2013)

  55. [63]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhu, R., Xu, S., Liu, P., Li, S., Lu, Y., Niu, D., Liu, Z., Meng, Z., Li, Z., Chen, X., Fan, Y.: Zero-shot structure-preserving diffusion model for high dynamic range tone mapping. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 26130...

  56. [64]

    diffusion models

    Zhu, R., Xu, S., Liu, P., Liu, J., Lu, Y., Niu, D., Zheng, H., Chen, Y.K., Jing, M., Fan, Y.: A flexible zero-shot approach to tone mapping via structure-preserving 12 S.Han et al. diffusion models. IEEE Transactions on Circuits and Systems for Video Technology (2025)

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.