Pith. sign in

REVIEW 4 major objections 7 minor 3 cited by

AFUNet: Cross-Iterative Alignment-Fusion Synergy for HDR Reconstruction via Deep Unfolding Paradigm

T0 review · 4 major / 7 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 AFUNet reconstructs HDR images from multi-exposure LDR inputs by alternating alignment and fusion at every iterative stage, and reports top results on three standard benchmarks.

desk verdict A competitive HDR deghosting network whose MAP-unfolding story is contradicted by its own modules; empirically solid, theoretically overstated. read the letter →

arxiv 2506.23537 v2 pith:YUETZWGZ submitted 2025-06-30 eess.IV cs.CV

Xinyue Li , Zhangkai Ni , Wenhan Yang This is my paper
classification eess.IVcs.CV
keywords HDRreconstructiondeepunfoldingMAPestimationmulti-exposureimagingdeghostingalignment-fusiontransformerimagerestoration
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 proposes AFUNet, a deep network for high-dynamic-range (HDR) reconstruction from three differently exposed low-dynamic-range images. Its central claim is that HDR reconstruction should be treated as a single optimization problem in which alignment and fusion are solved alternately rather than as separate pre-processing and merging steps. The authors cast the problem as maximum a posteriori (MAP) estimation with spatial correspondence priors, derive an alternating minimization scheme, and unfold that scheme into an end-to-end trainable network. They report that AFUNet consistently surpasses previous methods on the Kalantari, Hu, and Tel benchmarks, with gains attributed to the cross-iterative synergy between alignment and fusion. A sympathetic reader would care because the work offers a mathematical storyline for a task whose recent architectures are mostly empirical.

What carries the argument

The central object is the cross-iterative Alignment-Fusion Module (AFM), the network block obtained by unrolling one alternating-minimization step. Within an AFM, two window-based cross-attention Spatial Alignment Modules act as the gradient descent operators G1 and G3 of Eq. (4), aligning the non-reference features with the reference feature; a transformer-based Spatial Fusion Module followed by two channel-attention Channel Fusion Modules acts as the proximal operators of Eq. (7); and a Data Consistency Module implements the closed-form update of Eq. (8) as a learned combination of u, v, and the reference feature, with an MLP and residual connection refining the result. The mathematical roles are assigned by construction: the paper identifies SAM with gradient descent on the alignment priors, the SFM/CFM pair with proximal operators for the fusion priors, and DCM with the quadratic-regularized least-squares update. The network uses four default stages, with each stage sharing the same structure but independent parameters.

What would settle it

Train AFUNet in two configurations: the full model and a stripped model in which the SAM, CFM, and DCM are replaced by generic attention or convolution blocks of comparable capacity, and also record the value of the objective in Eq. (5) after each stage. If the stripped model matches the full model's accuracy, or if the unfolded objective does not decrease stage by stage, then the claimed correspondence between the modules and the optimization operators is not doing the work; conversely, if the full model strictly outperforms and the objective decreases, the theory is supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that multi-exposure HDR reconstruction is a joint alignment-and-fusion problem that should be optimized in alternation, not as sequential stages. Starting from a MAP objective over the reference exposure and two non-reference exposures, the authors introduce spatial correspondence priors p1 and p3, decompose the objective into alignment subproblems (solved by gradient descent operators G1 and G3) and a fusion subproblem (solved with half-quadratic splitting and proximal operators), and obtain a closed-form data-consistency update x_t = (D2^T D2 + (β1+β3)I)^{-1}(D2^T y2 + β1 u_t + β3 v_t). Unfolding this iteration yields AFUNet, whose T stages each contain a Spatial Alignment Module (window-based cross-attention), a Spatial Fusion Module plus two Channel Fusion Modules (channel-attention transformers), and a Data Consistency Module, with the degradation matrices $B^{{-1}}$ and D2^T learned by MLPs. The paper argues that because alignment sees the fusion state and fusion sees the alignment state at every stage, the two subtasks reinforce each other, producing better deghosting and detail than either the alignment-first or fusion-only paradigm. Empirically, the paper reports top PSNR, SSIM, and HDR-VDP2 numbers on the Kalantari, Hu, and Tel datasets, plus qualitative generalization to the Tursun dataset.

Load-bearing premise

The load-bearing premise is that the network's attention modules really perform the mathematical optimization steps they are named after (gradient descent, proximal updates, closed-form fusion); the paper provides no check that they do, so if that correspondence is only a name, the mathematical foundation is a post-hoc story and the real contribution is the empirical design.

Editorial extensions

If this is right

  • If the reported numbers hold, AFUNet is the best-performing HDR deghosting network on the Kalantari, Hu, and Tel benchmarks across PSNR, SSIM, and HDR-VDP2.
  • The alternating alignment-fusion pattern becomes a validated architectural template for multi-exposure reconstruction, since ablations show that adding SAM, CFM, and DCM each improves the result and that alignment-before-fusion beats fusion-before-alignment.
  • Because the degradation matrices B^{-1} and D2^T are learned rather than fixed, the same architecture can adapt to different camera response and degradation settings without re-deriving the model.
  • The 3-stage model already surpasses previous state-of-the-art in PSNR-µ and SSIM-µ, suggesting that the unfolding design achieves strong results with fewer iterations.
  • The deep-unfolding formulation means each stage refines a common objective rather than being a purely feed-forward stack, so the number of stages can be traded against quality and complexity.

Reading between the lines

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

  • The mathematical grounding is asserted but not verified: nothing in the training loss or ablations checks that the module outputs satisfy the fixed-point equations of the derived operators, so a cautious reading treats the MAP derivation as a design metaphor whose empirical payoff is what the tables show.
  • If the cross-iterative pattern is what matters, the same alternating alignment-fusion unfolding could transfer to other multi-image tasks such as burst denoising, video super-resolution, or multi-exposure fusion outside HDR, where alignment and fusion are also coupled.
  • A direct testable extension would be to add a consistency loss that penalizes deviation from the optimization equations, for example by enforcing that SAM's output equals a gradient step or that DCM matches Eq. (8) with shared B^{-1} and D2^T; such a constraint would either strengthen the theory-based claim or reveal where the analogy breaks.
  • Interpreting what the learned B^{-1} and D2^T matrices encode, such as whether they converge to expected degradation structure, could clarify how much of the model's behavior is truly unfolded optimization versus learned feature processing.
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

4 major / 7 minor

Summary. The paper proposes AFUNet, a deep unfolding network for multi-exposure HDR reconstruction. The authors formulate HDR reconstruction as a MAP estimation problem with two spatial-correspondence priors, decompose it into alternating alignment and fusion subproblems, and derive iterative updates involving gradient descent, proximal operators, and a closed-form least-squares step. These updates are then 'unfolded' into a T-stage network whose modules—SAM, SFM, CFM, and DCM—are claimed to implement the derived operators. The method is evaluated on Kalantari, Hu, and Tel datasets, with qualitative results on Tursun, and ablation studies on component contributions, paradigm ordering, and number of stages. The code is publicly available.

Significance. If the empirical results hold, AFUNet is a competitive HDR deghosting model, and the cross-iterative alignment-fusion design is a plausible design pattern; the ablations in Table 4-6 provide some evidence for the contribution of each module. However, the paper's central theoretical claim—that the network is 'theory-grounded' via MAP unfolding—is not substantiated: the priors p1 and p3 are never defined, and no evidence is given that the learned modules implement the derived mathematical operators. The empirical margins over the strongest baselines are small (often below 0.2 dB), with no error bars or significance tests. The contribution is therefore best viewed as an empirical architecture with a MAP-inspired narrative rather than a validated unfolding algorithm.

major comments (4)
  1. [Section 3.3, Eqs. (11)-(15)] The central claim that AFUNet is 'theory-grounded' is not supported by the module design. In Eq. (4), G_i is defined as a gradient descent operator on p_i(D_i x, alpha_i); for any differentiable p_i, the Jacobian of G_i with respect to alpha_i is I - zeta_i * Hessian_{alpha_i}(p_i), which is symmetric. SAM uses window-based cross-attention and an FFN (Eqs. 11-12), whose Jacobian is generically non-symmetric, so SAM cannot implement Eq. (4) for any differentiable prior. Similarly, CFM is described as a proximal operator for Eq. (7), but no property of the learned module (e.g., a minimization objective or fixed-point condition) is verified, and the proximal operator of p_i(D_i u, alpha_i) has a specific dependence on D_i and alpha_i that CFM does not reflect. Moreover, the paper states that all stages have independent parameters, so the T stages do not correspond to repeated iterations of a single objective. The authors should either provide constraints or training losses that enforce the operator identities, or explicitly reposition the MAP derivation as an architectural inspiration rather than an implemented optimization.
  2. [Section 3.3, Eqs. (8) and (14)] The closed-form update in Eq. (8) is derived in image space and requires B^{-1} = (D_2^T D_2 + (beta1 + beta3) I)^{-1}. The DCM learns B^{-1} and D_2^T with independent MLPs (Fig. 3d) and applies them to feature maps f_u, f_y2, and f_v, but no invertible mapping between image space and feature space is supplied. Thus Eq. (14) is not the feature-space version of Eq. (8) unless the MLPs are constrained to satisfy the Woodbury relation linking B^{-1}, D_2, and the beta weights. Please add consistency constraints or an experiment demonstrating that the learned matrices satisfy the required relation on the validation set.
  3. [Abstract, Section 4.2, Table 1] The statement that AFUNet 'consistently surpass[es] state-of-the-art methods' is overstated. On Kalantari PSNR-l, AFUNet (42.59 dB) ties LFDiff and is below FlexHDR (42.60 dB) and SAFNet (43.18 dB); the PSNR-mu margins over the strongest baselines are 0.15 dB (LFDiff) and 0.25 dB (SAFNet), respectively, and no error bars, multiple seeds, or significance tests are provided. Please quantify variability and either soften the claim or add a statistical test.
  4. [Section 3.2, Eq. (2)] The spatial correspondence priors p1 and p3 are never defined; the paper does not specify their functional form, differentiability, or how they relate to the training loss in Eq. (17). The variables alpha1 and alpha3 are initialized as features, not as aligned images, and no term in the training loss enforces that alpha_i is a spatially aligned version of y_i. Consequently, the MAP derivation is a scaffold that is not instantiated or verified. Please either define the priors and show the corresponding gradient and proximal operators, or state clearly that p1 and p3 are abstract placeholders and that the contribution is an empirical unfolding architecture.
minor comments (7)
  1. [Section 3.2, Eq. (7a)] There is an unmatched parenthesis in the proximal operator definition: 'prox_{lambda1/beta1 p1(·,alpha_t1))}(x^{t-1})' has an extra closing parenthesis; please fix the notation.
  2. [Section 3.3, after Eq. (10)] The text says 'The features f^{t-1}_{alpha1}, f^{t-1}_x, f^{t-1}_{alpha3} represent the outputs from stage t-2', but Eq. (10) indicates they are the outputs from stage t-1; the indexing should be made consistent.
  3. [Figure 3(a) caption] The caption says SAM aligns features with 'f^t_x', but Eq. (11) uses f^{t-1}_x; the notation should be aligned.
  4. [Table 3] Table 3 lists only a subset of the baselines included in Tables 1 and 2 (e.g., ADNet, APNT, HyHDR, LFDiff, RFG-HDR, SAFNet are omitted). Please either include the same baseline set or state explicitly why some methods are not compared on Tel's dataset.
  5. [Section 4.1, Implementation Details] The paper does not report the number of parameters, FLOPs, or runtime for the proposed model or the compared methods; please add these for reproducibility and practical assessment.
  6. [Section 4.2 and Figure 5] The dataset name is spelled 'Tursen' in the main text and figure caption; the correct name is 'Tursun'.
  7. [Section 3.4, Eq. (17)] The text says the loss is computed in the tone-mapped domain by applying the mu-law function, but the formula applies tau(x) and tau(hat x) without specifying how the predicted HDR values are clamped or normalized before tone mapping; please clarify the preprocessing.

Circularity Check

3 steps flagged · score 6.0 of 10

AFUNet's 'theory-grounded' claim is self-definitional: the MAP priors are never specified, the derived operators B−1 and D2T of Eq. (8) are learned by unconstrained MLPs, and SAM's cross-attention update cannot be a gradient step of any differentiable prior; the architecture defines the theory rather than being derived from it.

  1. self definitional [Section 3.2, Eq. (8), and Section 3.3 / Fig. 3(d)]
    "The matrix inverse is computationally expensive, so we treat (DT2 D2 + (β1 + β3)I)−1 as a single entity, denoted as B−1. To efficiently handle this, we design neural networks to learn the complex degradation matrices B−1 and DT2."

    Eq. (8) is the paper's derived closed-form fusion update whose defining relation is B−1 = (DT2 D2 + (β1 + β3)I)−1. The paper then declares this matrix inverse 'computationally expensive' and assigns its computation, together with DT2, to neural networks (independent MLPs, Fig. 3(d)), and names a module DCM that 'obtains f t xp according to Eq. (8)'. No constraint enforces the defining relation B−1 = (DT2 D2 + (β1 + β3)I)−1 or ties the two MLPs together. The supposedly derived operator therefore has no definition independent of the trained network: whatever the MLP computes is, by naming, the closed-form update. The claim that the network implements the derivation is thus true only by construction, in the manner of a fitted parameter re-labeled as a prediction.

  2. self definitional [Section 3.3, subsection A (Feature Alignment), Eqs. (4), (11)-(12)]
    "There are two SAMs act as gradient descent operators G1(·) and G3(·) in Eq. (4a) and (4b), respectively, producing the aligned features f t α1 , f t α3. ... we construct the SAM as a window-based cross-attention transformer block [19], which can be formulated as: f t αi = FFN(WCAA(f t−1 αi , f t−1 x )), (11)"

    For any twice-differentiable prior p_i, the gradient-descent map Gi(α) = α − ς ∇α p_i(D_i x, α) of Eq. (4) has Jacobian I − ς·Hessian(p_i), which is symmetric. The SAM update of Eqs. (11)-(12) is a windowed cross-attention block plus FFN whose Jacobian is generically non-symmetric, so no differentiable p_i exists for which SAM equals Gi. The asserted identity 'two SAMs act as gradient descent operators' therefore cannot be a consequence of the derivation. It can hold only if the prior p_i and operator Gi are defined to be the trained SAM, i.e., the theory is reverse-engineered from the module and the gradient-descent label is attached after the fact. The alignment update 'derived' in Eq. (4) thus places no constraint on the architecture.

1 more flagged steps
  1. other [Section 3.3, first paragraph (HDR Feature Reconstruction)]
    "Notably, as outlined in Section 3.2, the solution to the unfolding paradigm is performed in the image space, where the LDR images y1, y2, y3 are directly involved in the optimization process. In contrast, in this section, we apply the iterative optimization and refinement at the feature space and propose a learnable solution using a deep unfolding network."

    The derivation of Eqs. (2)-(8) is conducted on image variables x, α1, α3, u, v with degradation operators D1, D2, D3, and its closed form mixes y2, u, v as images. The paper concedes the translation: the solution is 'performed in the image space' while the network operates 'at the feature space'. All variables are replaced by features f_x, f_αi, f_u, f_v produced by a single convolutional layer (SFEMi, Eq. (9)), with no invertible map supplied between image y2 or x and their feature counterparts. Hence the modules are not the derived iterations; they are feature-space networks that inherit the names of image-space variables. The 'unfolding' step is a renaming convention rather than a construction from Eq. (8), so the claimed derivation does not connect to the implemented modules.

full rationale

The empirical contribution is genuine and benchmark-grounded: Tables 1-3 compare AFUNet against prior methods on the held-out Kalantari, Hu, and Tel test sets, and Tables 4-6 measure component and stage-count effects on held-out data. On that axis the paper is self-contained and not circular, which is why the score is not higher. (The abstract's 'consistently surpassing state-of-the-art' is overstated: on Kalantari PSNR-l AFUNet ties LFDiff at 42.59 and trails SAFNet at 43.18, but that is a correctness/accuracy concern, not circularity.) The circularity lies in the paper's central novelty claim: that the architecture is 'theory-grounded' because it is unfolded from a MAP objective (Eqs. 1-8). That claim reduces by construction. First, the priors p1 and p3 in Eqs. (2)-(7) are never given functional forms, and the 'proximal operators' and 'gradient descent operators' are asserted to be the learned modules themselves. Second, the only non-trivial derived quantity, the closed-form matrix B−1 in Eq. (8), is replaced by an independent learned MLP with no constraint enforcing its defining relation to D2. Third, the identity 'SAM = Gi' is mathematically incompatible with Eq. (4) for any differentiable prior (the gradient-step Jacobian must be symmetric; cross-attention's is not), so the gradient-descent step is a post hoc label: Gi is defined to be SAM. Fourth, the derivation is performed in image space but implemented in feature space with no invertible variable map, and all stages 'share the same structure but have independent parameters', so the stages are not iterations of one fixed objective. Each of these moves ensures the derived equations place no constraint on the network: the network is the data and the MAP symbols are fitted to it afterward. No self-citation chain is involved; the circularity is of the self-definitional kind, and it affects the theoretical contribution rather than the external benchmark results.

Assumptions & free parameters 4 free parameters · 5 assumptions · 3 invented entities

The central claim rests on the existence and differentiability of two unspecified priors, the assumption that learned attention modules implement the operators of the derivation, and the assumption that the degradation model is linear with learnable matrices. These are domain assumptions or ad hoc scaffolding rather than established facts.

free parameters (4)
  • Number of unfolding stages T = 4
    Chosen by hand as the default balancing performance and complexity, based on the ablation in Section 4.3 (Table 6).
  • Perceptual loss weight eta = 0.005
    Set manually in Eq. (17); a standard value, not fitted to the data.
  • Balancing coefficients lambda1, lambda3, beta1, beta3 and step sizes xi_i = unspecified
    Appear in the MAP derivation Eqs. (2)-(8) but are not used in the network; their values are never reported or tuned, making the derivation decorative.
  • Learning rate and schedule = 5e-4 to 5e-6 cosine decay
    Standard training hyperparameters; not part of the scientific claim.
assumptions (5)
  • domain assumption The degradation process is modeled as y = Dx + n with additive Gaussian noise, justifying the l2 data fidelity term.
    Section 3.2 Eq. (1) assumes a linear degradation model and l2 fidelity. Real HDR-to-LDR degradation includes nonlinear tone mapping and saturation, which is only approximately linear.
  • ad hoc to paper Spatial correspondence priors p1 and p3 exist, are differentiable, and are strong enough to make the alternating optimization meaningful.
    The priors are never defined; they are placeholders. The actual alignment and fusion are performed by learned attention networks, so the priors impose no verifiable constraint. Section 3.2.
  • ad hoc to paper The hand-designed modules (SAM, CFM, DCM) compute the corresponding mathematical operators (gradient descent, proximal, closed-form) from Eqs. (4)-(8).
    No constraint or proof links the modules to the operators; they are named after the operators. Section 3.3.
  • ad hoc to paper The matrices B^{-1} and D2^T can be learned by MLPs and still correspond to the degradation model used in Eq. (8).
    The closed-form solution is not computed; these matrices are learned from data, so the derivation does not determine the update. Section 3.2.
  • standard math Half-Quadratic Splitting converges to the solution of Eq. (2) under the chosen priors.
    HQS is a standard optimization technique; convergence requires conditions on the priors that are not verified. Section 3.2.
invented entities (3)
  • Spatial correspondence priors p1(D1x, alpha1) and p3(D3x, alpha3)
    purpose: Intended to constrain the spatial correspondence between non-reference LDR images and the reference, bridging alignment and fusion.
    The priors are never explicitly defined, and the modules that replace them are trained end-to-end; no testable prediction follows from the priors themselves.
  • Spatially aligned versions alpha1 and alpha3 of the non-reference images
    purpose: Intermediate optimization variables representing aligned versions of y1 and y3 with respect to the reference.
    These are internal latent variables produced by the SAM network; they are not validated against any ground-truth correspondence.
  • Learned degradation matrices B^{-1} and D2^T
    purpose: Replace the matrix inverse in the closed-form solution Eq. (8) with learned transformations.
    The degradation model is unknown; learning these matrices means the claimed closed-form solution is actually a learned network component.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AFUNet: Cross-Iterative Alignment-Fusion Synergy for HDR Reconstruction via Deep Unfolding Paradigm." pith.science (2026). https://pith.science/paper/YUETZWGZ

@misc{pith2026250623537,
  author       = {Pith},
  title        = {Pith review of: AFUNet: Cross-Iterative Alignment-Fusion Synergy for HDR Reconstruction via Deep Unfolding Paradigm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YUETZWGZ}},
  note         = {Machine review of arXiv:2506.23537}
}
read the original abstract

Existing learning-based methods effectively reconstruct HDR images from multi-exposure LDR inputs with extended dynamic range and improved detail, but they rely more on empirical design rather than theoretical foundation, which can impact their reliability. To address these limitations, we propose the cross-iterative Alignment and Fusion deep Unfolding Network (AFUNet), where HDR reconstruction is systematically decoupled into two interleaved subtasks -- alignment and fusion -- optimized through alternating refinement, achieving synergy between the two subtasks to enhance the overall performance. Our method formulates multi-exposure HDR reconstruction from a Maximum A Posteriori (MAP) estimation perspective, explicitly incorporating spatial correspondence priors across LDR images and naturally bridging the alignment and fusion subproblems through joint constraints. Building on the mathematical foundation, we reimagine traditional iterative optimization through unfolding -- transforming the conventional solution process into an end-to-end trainable AFUNet with carefully designed modules that work progressively. Specifically, each iteration of AFUNet incorporates an Alignment-Fusion Module (AFM) that alternates between a Spatial Alignment Module (SAM) for alignment and a Channel Fusion Module (CFM) for adaptive feature fusion, progressively bridging misaligned content and exposure discrepancies. Extensive qualitative and quantitative evaluations demonstrate AFUNet's superior performance, consistently surpassing state-of-the-art methods. Our code is available at: https://github.com/eezkni/AFUNet

Figures

Figures reproduced from arXiv: 2506.23537 by the authors.

Figure 1
Figure 1. Comparison of three HDR reconstruction paradigms: [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Framework of the AFUNet for HDR reconstruction consists of three main processes: Initialization, HDR Feature Reconstruction, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) SAM uses window-based cross-attention to align [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison between our method and state-of-the-art methods on three datasets: (a) Kalantari’s dataset [ [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison between our method and state [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction

    cs.CV 2026-03 conditional novelty 6.0 of 10

    Dual-stream HDR video can be reconstructed without fragile cross-exposure warping by backbone-guided collaborative attention plus sequence-level residual refinement.

  2. GMODiff: One-Step Gain Map Refinement with Diffusion Priors for HDR Reconstruction

    cs.CV 2025-12 conditional novelty 6.0 of 10

    HDR reconstruction is reformulated as one-step gain map refinement, enabling a pre-trained latent diffusion model plus regression priors to produce high-quality HDR at a fraction of the cost of prior diffusion approaches.

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

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A coarse-to-fine multi-exposure fusion method that fuses low-res diffusion output with implicit-neural high-res detail reconstruction, achieving ~3.5x speedup over a diffusion-only baseline.

Reference graph

Works this paper leans on

50 extracted references · 47 canonical work pages · cited by 3 Pith papers

  1. [1]

    Extending dynamic range of monochrome and color images through fusion

    Luca Bogoni. Extending dynamic range of monochrome and color images through fusion. In IEEE International Confer- ence on Pattern Recognition, pages 7–12, 2000. 1

    work page 2000

  2. [2]

    FlexHDR: Modeling alignment and exposure uncertainties for flexible HDR imaging

    Sibi Catley-Chandar, Thomas Tanay, Lucas Vandroux, Ale ˇs Leonardis, Gregory Slabaugh, and Eduardo P ´erez-Pellitero. FlexHDR: Modeling alignment and exposure uncertainties for flexible HDR imaging. IEEE Transactions on Image Pro- cessing, 31:5923–5935, 2022. 2, 6

    work page 2022

  3. [3]

    Attention-guided progressive neural tex- ture fusion for high dynamic range image restoration

    Jie Chen, Zaifeng Yang, Tsz Nam Chan, Hui Li, Junhui Hou, and Lap-Pui Chau. Attention-guided progressive neural tex- ture fusion for high dynamic range image restoration. IEEE Transactions on Image Processing, 31:2661–2672, 2022. 6

    work page 2022

  4. [4]

    Improving dynamic HDR imaging with fusion transformer

    Rufeng Chen, Bolun Zheng, Hua Zhang, Quan Chen, Cheng- gang Yan, Gregory Slabaugh, and Shanxin Yuan. Improving dynamic HDR imaging with fusion transformer. In Proceed- ings of the AAAI Conference on Artificial Intelligence, pages 340–349, 2023. 2

    work page 2023

  5. [5]

    Recovering high dy- namic range radiance maps from photographs

    Paul E Debevec and Jitendra Malik. Recovering high dy- namic range radiance maps from photographs. In Proceed- ings of the 24th Annual Conference on Computer Graphics and Interactive Techniques, pages 369–378, 1997. 1

    work page 1997

  6. [6]

    Nonlinear image recov- ery with half-quadratic regularization

    Donald Geman and Chengda Yang. Nonlinear image recov- ery with half-quadratic regularization. IEEE Transactions on Image Processing, 4(7):932–946, 1995. 3

    work page 1995

  7. [7]

    Fast and robust high dynamic range image generation with camera and object movement

    Thorsten Grosch et al. Fast and robust high dynamic range image generation with camera and object movement. Vi- sion, Modeling and Visualization, RWTH Aachen , 277284 (3):2, 2006. 1

    work page 2006

  8. [8]

    CPP-Net: Embracing multi- scale feature fusion into deep unfolding CP-PPA network for compressive sensing

    Zhen Guo and Hongping Gan. CPP-Net: Embracing multi- scale feature fusion into deep unfolding CP-PPA network for compressive sensing. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 25086–25095, 2024. 2

    work page 2024

Show all 50 references
  1. [9]

    MERF: a prac- tical HDR-like image generator via mutual-guided learning between multi-exposure registration and fusion.IEEE Trans- actions on Image Processing, 33:2361–2376, 2024

    Wenhui Hong, Hao Zhang, and Jiayi Ma. MERF: a prac- tical HDR-like image generator via mutual-guided learning between multi-exposure registration and fusion.IEEE Trans- actions on Image Processing, 33:2361–2376, 2024. 2

    2024

  2. [10]

    HDR deghosting: How to deal with saturation? In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 1163–1170, 2013

    Jun Hu, Orazio Gallo, Kari Pulli, and Xiaobai Sun. HDR deghosting: How to deal with saturation? In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 1163–1170, 2013. 1, 6, 7

    2013

  3. [11]

    Gen- erating content for HDR deghosting from frequency view

    Tao Hu, Qingsen Yan, Yuankai Qi, and Yanning Zhang. Gen- erating content for HDR deghosting from frequency view. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 25732–25741, 2024. 6, 7

    2024

  4. [12]

    Un- folding the alternating optimization for blind super resolu- tion

    Yan Huang, Shang Li, Liang Wang, Tieniu Tan, et al. Un- folding the alternating optimization for blind super resolu- tion. Advances in Neural Information Processing Systems , 33:5632–5643, 2020. 2

    2020

  5. [13]

    Deep high dynamic range imaging of dynamic scenes

    Nima Khademi Kalantari and Ravi Ramamoorthi. Deep high dynamic range imaging of dynamic scenes. ACM Transac- tions on Graphics, 36(4):1–12, 2017. 2, 6, 7, 8

    2017

  6. [14]

    High dynamic range video

    Sing Bing Kang, Matthew Uyttendaele, Simon Winder, and Richard Szeliski. High dynamic range video. ACM Transac- tions on Graphics, 22(3):319–325, 2003. 1

    2003

  7. [15]

    SAFNet: Selective alignment fusion net- work for efficient HDR imaging

    Lingtong Kong, Bo Li, Yike Xiong, Hao Zhang, Hong Gu, and Jinwei Chen. SAFNet: Selective alignment fusion net- work for efficient HDR imaging. In Proceedings of the Eu- ropean Conference on Computer Vision, 2024. 2, 6, 7

    2024

  8. [16]

    Ghost-free high dynamic range imaging via rank minimization

    Chul Lee, Yuelong Li, and Vishal Monga. Ghost-free high dynamic range imaging via rank minimization. IEEE Signal Processing Letters, 21(9):1045–1049, 2014. 1

    2014

  9. [17]

    RFG-HDR: Representative feature-guided transformer for multi-exposure high dynamic range imaging

    Keuntek Lee, Jaehyun Park, Gu Yong Park, and Nam Ik Cho. RFG-HDR: Representative feature-guided transformer for multi-exposure high dynamic range imaging. In 2024 IEEE International Conference on Image Processing, pages 1521–1527. IEEE, 2024. 6

    2024

  10. [18]

    ADNet: Attention-guided deformable convolutional network for high dynamic range imaging

    Zhen Liu, Wenjie Lin, Xinpeng Li, Qing Rao, Ting Jiang, Mingyan Han, Haoqiang Fan, Jian Sun, and Shuaicheng Liu. ADNet: Attention-guided deformable convolutional network for high dynamic range imaging. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2021

  11. [19]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10012–10022, 2021. 5

    2021

  12. [20]

    Ghost-free high dynamic range imaging with context-aware transformer

    Zhen Liu, Yinglong Wang, Bing Zeng, and Shuaicheng Liu. Ghost-free high dynamic range imaging with context-aware transformer. In Proceedings of the European Conference on Computer Vision, pages 344–360, 2022. 2, 6, 7

    2022

  13. [21]

    Robust multi-exposure image fusion: a struc- tural patch decomposition approach

    Kede Ma, Hui Li, Hongwei Yong, Zhou Wang, Deyu Meng, and Lei Zhang. Robust multi-exposure image fusion: a struc- tural patch decomposition approach. IEEE Transactions on Image Processing, 26(5):2519–2532, 2017. 1

    2017

  14. [22]

    Deep guided learning for fast multi- exposure image fusion

    Kede Ma, Zhengfang Duanmu, Hanwei Zhu, Yuming Fang, and Zhou Wang. Deep guided learning for fast multi- exposure image fusion. IEEE Transactions on Image Pro- cessing, 29:2808–2819, 2020. 1

    2020

  15. [23]

    Deep unrolled low-rank tensor completion for high dynamic range imaging

    Truong Thanh Nhat Mai, Edmund Y Lam, and Chul Lee. Deep unrolled low-rank tensor completion for high dynamic range imaging. IEEE Transactions on Image Processing, 31: 5774–5787, 2022. 2

    2022

  16. [24]

    HDR-VDP-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions

    Rafał Mantiuk, Kil Joong Kim, Allan G Rempel, and Wolf- gang Heidrich. HDR-VDP-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions. ACM Transactions on graphics, 30(4):1–14, 2011. 6

    2011

  17. [25]

    Progressive high-frequency re- construction for pan-sharpening with implicit neural repre- sentation

    Ge Meng, Jingjia Huang, Yingying Wang, Zhenqi Fu, Xing- hao Ding, and Yue Huang. Progressive high-frequency re- construction for pan-sharpening with implicit neural repre- sentation. In Proceedings of the AAAI Conference on Artifi- cial Intelligence, pages 4189–4197, 2024. 2

    2024

  18. [26]

    HDR-GAN: HDR image reconstruction from multi-exposed LDR images with large motions

    Yuzhen Niu, Jianbin Wu, Wenxi Liu, Wenzhong Guo, and Rynson WH Lau. HDR-GAN: HDR image reconstruction from multi-exposed LDR images with large motions. IEEE Transactions on Image Processing, 30:3885–3896, 2021. 6

    2021

  19. [27]

    Robust high dynamic range imaging by rank min- imization

    Tae-Hyun Oh, Joon-Young Lee, Yu-Wing Tai, and In So Kweon. Robust high dynamic range imaging by rank min- imization. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 37(6):1219–1232, 2014. 1 9

    2014

  20. [28]

    Ram Prabhakar, V Sai Srikar, and R

    K. Ram Prabhakar, V Sai Srikar, and R. Venkatesh Babu. DeepFuse: A deep unsupervised approach for exposure fu- sion with extreme exposure image pairs. In Proceedings of the IEEE International Conference on Computer Vision ,

  21. [29]

    Robust patch-based HDR reconstruction of dynamic scenes

    Pradeep Sen, Nima Khademi Kalantari, Maziar Yaesoubi, Soheil Darabi, Dan B Goldman, and Eli Shechtman. Robust patch-based HDR reconstruction of dynamic scenes. ACM Transactions on Graphics, 31(6):203, 2012. 1

    2012

  22. [30]

    Very deep convolutional net- works for large-scale image recognition

    K Simonyan and A Zisserman. Very deep convolutional net- works for large-scale image recognition. In International Conference on Learning Representations, 2015. 6

    2015

  23. [31]

    Dynamic path- controllable deep unfolding network for compressive sens- ing

    Jiechong Song, Bin Chen, and Jian Zhang. Dynamic path- controllable deep unfolding network for compressive sens- ing. IEEE Transactions on Image Processing , 32:2202– 2214, 2023. 2

    2023

  24. [32]

    Selective TransHDR: Transformer-based selective HDR imaging using ghost region mask

    Jou Won Song, Ye-In Park, Kyeongbo Kong, Jaeho Kwak, and Suk-Ju Kang. Selective TransHDR: Transformer-based selective HDR imaging using ghost region mask. In Eu- ropean Conference on Computer Vision , pages 288–304. Springer, 2022. 2

    2022

  25. [33]

    Alignment-free HDR deghosting with semantics consistent transformer

    Steven Tel, Zongwei Wu, Yulun Zhang, Barth ´el´emy Heyr- man, C ´edric Demonceaux, Radu Timofte, and Dominique Ginhac. Alignment-free HDR deghosting with semantics consistent transformer. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 1283...

    2023

  26. [34]

    An objective deghosting quality metric for HDR images

    Okan Tarhan Tursun, Ahmet O˘guz Aky¨uz, Aykut Erdem, and Erkut Erdem. An objective deghosting quality metric for HDR images. In Computer Graphics Forum, pages 139–152,

  27. [35]

    Progressive divide-and-conquer via subsam- pling decomposition for accelerated mri

    Chong Wang, Lanqing Guo, Yufei Wang, Hao Cheng, Yi Yu, and Bihan Wen. Progressive divide-and-conquer via subsam- pling decomposition for accelerated mri. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25128–25137, 2024. 2

    2024

  28. [36]

    Deep unfolded network with intrinsic super- vision for pan-sharpening

    Hebaixu Wang, Meiqi Gong, Xiaoguang Mei, Hao Zhang, and Jiayi Ma. Deep unfolded network with intrinsic super- vision for pan-sharpening. In Proceedings of the AAAI Con- ference on Artificial Intelligence, pages 5419–5426, 2024. 2

    2024

  29. [37]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 6

    2004

  30. [38]

    Chi- Keung

    Shangzhe Wu, Xu Jiarui, Tai Yu-Wing, and Tang. Chi- Keung. Deep high dynamic range imaging with large fore- ground motions. In Proceedings of the European Conference on Computer Vision, pages 117–132, 2018. 6

    2018

  31. [39]

    Attention- guided network for ghost-free high dynamic range imaging

    Qingsen Yan, Dong Gong, Qinfeng Shi, Anton van den Hen- gel, Chunhua Shen, Ian Reid, and Yanning Zhang. Attention- guided network for ghost-free high dynamic range imaging. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1751–1760, 2...

    2019

  32. [40]

    Deep HDR imaging via a non- local network

    Qingsen Yan, Lei Zhang, Yu Liu, Yu Zhu, Jinqiu Sun, Qin- feng Shi, and Yanning Zhang. Deep HDR imaging via a non- local network. IEEE Transactions on Image Processing, 29: 4308–4322, 2020. 2, 6

    2020

  33. [41]

    A unified HDR imaging method with pixel and patch level

    Qingsen Yan, Weiye Chen, Song Zhang, Yu Zhu, Jinqiu Sun, and Yanning Zhang. A unified HDR imaging method with pixel and patch level. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 22211–22220, 2023. 2, 6

    2023

  34. [42]

    Towards high- quality HDR deghosting with conditional diffusion models

    Qingsen Yan, Tao Hu, Yuan Sun, Hao Tang, Yu Zhu, Wei Dong, Luc Van Gool, and Yanning Zhang. Towards high- quality HDR deghosting with conditional diffusion models. IEEE Transactions on Circuits and Systems for Video Tech- nology, 2023. 6

    2023

  35. [43]

    Memory-augmented deep conditional unfolding network for pan-sharpening

    Gang Yang, Man Zhou, Keyu Yan, Aiping Liu, Xueyang Fu, and Fan Wang. Memory-augmented deep conditional unfolding network for pan-sharpening. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1788–1797, 2022. 2

    2022

  36. [44]

    Progressive and selective fusion network for high dynamic range imaging

    Qian Ye, Jun Xiao, Kin-man Lam, and Takayuki Okatani. Progressive and selective fusion network for high dynamic range imaging. In Proceedings of the 29th ACM Interna- tional Conference on Multimedia , pages 5290–5297, 2021. 2

    2021

  37. [45]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5728–5739,

  38. [46]

    Deep unfold- ing network for image super-resolution

    Kai Zhang, Luc Van Gool, and Radu Timofte. Deep unfold- ing network for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3217–3226, 2020. 2

    2020

  39. [47]

    Gradient-directed multiex- posure composition

    Wei Zhang and Wai-Kuen Cham. Gradient-directed multiex- posure composition. IEEE Transactions on Image Process- ing, 21(4):2318–2323, 2011. 1

    2011

  40. [48]

    Hl-hdr: Multi-exposure high dy- namic range reconstruction with high-low frequency decom- position

    Xiang Zhang, Genggeng Chen, Tao Hu, Kangzhen Yang, Fan Zhang, and Qingsen Yan. Hl-hdr: Multi-exposure high dy- namic range reconstruction with high-low frequency decom- position. In 2024 International Joint Conference on Neural Networks, pages 1–9. IEEE, 2024. 2

    2024

  41. [49]

    Memory-augmented deep unfolding net- work for guided image super-resolution

    Man Zhou, Keyu Yan, Jinshan Pan, Wenqi Ren, Qi Xie, and Xiangyong Cao. Memory-augmented deep unfolding net- work for guided image super-resolution. International Jour- nal of Computer Vision, 131(1):215–242, 2023. 2

    2023

  42. [50]

    Freehand HDR imaging of moving scenes with simultane- ous resolution enhancement

    Henning Zimmer, Andr ´es Bruhn, and Joachim Weickert. Freehand HDR imaging of moving scenes with simultane- ous resolution enhancement. In Computer Graphics Forum, pages 405–414, 2011. 1 10

    2011

Pith tools

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