pith. machine review for the scientific record. sign in

arxiv: 2605.11628 · v1 · submitted 2026-05-12 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Single-Shot HDR Recovery via a Video Diffusion Prior

Chinmay Talegaonkar , Jinshi He , Christopher McKenna , Nicholas Antipa
Authors on Pith no claims yet

Pith reviewed 2026-05-13 01:41 UTC · model grok-4.3

classification 💻 cs.CV
keywords single-shot HDRvideo diffusionexposure bracketimage fusionHDR reconstructiongenerative modelsconditional generationdefocus deblurring
0
0 comments X

The pith

Single-shot HDR reconstruction works by generating an exposure video sequence from a diffusion model and fusing the frames.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that single-shot high dynamic range recovery improves when treated as conditional video generation instead of direct image prediction. A video diffusion model is fine-tuned to output a bracket of differently exposed frames from one low dynamic range input, after which a lightweight UNet predicts per-pixel weights to combine them into the final HDR result. This keeps the process interpretable because the intermediate exposures remain explicit rather than hidden inside an end-to-end guess. A sympathetic reader would care because the method removes the need for separate highlight and shadow models while maintaining higher fidelity to the original capture.

Core claim

We address these limitations by re-casting single-shot HDR reconstruction as conditional video generation and fusing the generated frames into an HDR image. We finetune a video diffusion model to generate an exposure bracket, conditioned on a low dynamic range (LDR) input. We fuse this image bracket using per-pixel weights predicted by a light-weight UNet. This formulation is simple, interpretable, and effective. Rather than directly hallucinating an HDR image, it explicitly reconstructs the intermediate exposure stack and fuses it into the final output. Our method eliminates the need for separate models across exposure regimes and produces HDR reconstructions with high input fidelity. On量化性

What carries the argument

Conditional video diffusion model that generates an exposure bracket from an LDR input, followed by lightweight UNet fusion of per-pixel weights.

If this is right

  • No separate models are required for different exposure regimes such as highlights and shadows.
  • Reconstruction maintains higher fidelity to the original LDR input than direct prediction methods.
  • Quantitative metrics on benchmarks exceed those of comparable generative baselines.
  • Human raters prefer the outputs in 72 percent of pairwise comparisons.
  • The same input-conditioned sequence generation and fusion extends to other tasks such as all-in-focus recovery from a defocused input.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework suggests video diffusion priors can supply consistent multi-frame outputs for any single-image inverse problem that benefits from explicit intermediate states.
  • Similar conditioning and fusion could be tested on other image restoration tasks where generating a short sequence enforces physical or geometric consistency.

Load-bearing premise

Fine-tuning a video diffusion model on LDR inputs will produce an exposure bracket whose frames are consistent and accurate enough for a simple UNet to fuse them into high-fidelity HDR without new artifacts or lost detail.

What would settle it

A side-by-side test on standard HDR benchmarks where the generated frames show visible misalignment or detail loss and the final PSNR or SSIM falls below direct generative baselines of similar capacity.

Figures

Figures reproduced from arXiv: 2605.11628 by Chinmay Talegaonkar, Christopher McKenna, Jinshi He, Nicholas Antipa.

Figure 1
Figure 1. Figure 1: Video diffusion models contain implicit priors to generate exposure bursts. Given an input LDR image (a) captured at a high exposure, we prompt an off-the-shelf video model (Runway Gen 4): "Reduce the brightness of the image over the course of the video, generating an exposure burst stack from high to low exposure". The resulting video frames are plotted in (b), demonstrating that the video model contains … view at source ↗
Figure 2
Figure 2. Figure 2: Method overview. (a) Given an input LDR image, we fine-tune a latent video diffusion model to generate an LDR exposure bracket corresponding to the scene. (b) A fusion UNet then predicts per-pixel weights to fuse this bracket into an HDR image. The fusion network is supervised in PU-21 encoded space, and at (c) inference time, we sample the video diffusion model to synthesize an LDR bracket, which is fused… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative HDR reconstruction results. We show challenging ill-exposed examples where our method produces higher-fidelity HDR reconstructions than the baselines. Our method is better at denoising under-exposed regions (rows 2, 3, 4) and generates more plausible content in over-exposed regions (rows 1, 4). All results are normalized to the same median luminance and visualized with Reinhard tone mapping Rei… view at source ↗
Figure 5
Figure 5. Figure 5: Failure cases. Our method can struggle with fine text (row 1), as the VAE encoding can lose out on small, high-frequency details. Ex￾tremely dark inputs (row 2) remain challenging for all methods due to a low signal-to-noise ratio. trained on LDR brackets synthesized from the GT exposure distribution used to fine-tune the video model, allowing it to learn a fusion strategy better suited to our predicted LD… view at source ↗
Figure 6
Figure 6. Figure 6: Fusion method comparison. (Left) UNet fusion weights on a test scene. (Right) Visual comparison of Mertens vs. UNet fusion on a test scene. Our Fusion UNet generates spatially smoother and less noisy fusion weights as compared to Mertens [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Extending our method to all-in-focus (AIF) image recovery. We fine-tune the video model to take a defocused image as input and generate a focal stack of images with varying focus distances. We fuse the generated focal stack into an AIF image using a separate Fusion UNet. lower deviation (0.233) than the base SVD model (0.326), corresponding to an approximately 29% improvement in bracket consistency. See Ap… view at source ↗
Figure 8
Figure 8. Figure 8: Exposure brackets generated by our fine-tuned model. Each row corresponds to one scene from the SI-HDR test set. We show the ground-truth LDR exposure bracket under the header GT and our predictions under Pred. Exposure increases from left to right under each header. Our predictions match the relative exposure progression of the GT brackets with high fidelity in most cases. In rows 3, 4, and 6, we observe … view at source ↗
Figure 9
Figure 9. Figure 9: User-study trial interface. The ground-truth HDR image is displayed on the top half (centered) and the two candidates (ours vs. a randomly chosen baseline) are displayed side-by-side on the bottom half. The participant uses the left/right arrow keys to pick the candidate that best matches the ground truth. Stimuli and pre-display pipeline: Each trial draws on a single scene from the test split, where the s… view at source ↗
Figure 10
Figure 10. Figure 10: User study failure cases. Reconstructions for the five test-set scenes on which participants most often preferred a competing baseline over ours. amount of high-frequency details visible in the input. Intuitively, when fewer regions are saturated or under-exposed, more edges remain measurable in the input. Based on this insight, we use the edge density, defined below, as a simple proxy for visible high-fr… view at source ↗
Figure 11
Figure 11. Figure 11: Per-scene ∆ SSIM vs. input edge density (n = 96). Each point is one test scene. ∆ SSIM = SSIMOurs − SSIMbaselines (positive ⇒ Ours better). The Pearson correlation coefficient between edge density and ∆ SSIM is r = −0.79: We outperform the baselines on inputs with low edge-density. After controlling for the mean LDR luminance (normalized to [0, 1]), edge density remains strongly correlated with ∆ SSIM (pa… view at source ↗
Figure 12
Figure 12. Figure 12: Failure-case qualitative comparison. We show five well-exposed scenes containing high-frequency textures. HDR images are visualized with Reinhard tone-mapping. In these cases, our method loses some fine texture detail compared to existing methods, consistent with the VAE bottleneck analysis in section J.1. Nevertheless, the reconstructions remain visually comparable overall [PITH_FULL_IMAGE:figures/full_… view at source ↗
Figure 13
Figure 13. Figure 13: Bright-input scenes where our method outperforms most baselines. Input LDR thumb￾nails (top row) and tone-mapped ground-truth HDR images (bottom row) for the 10 brightly exposed test scenes (mean LDR luminance ≥ 0.40). These examples show that over-exposure can reduce apparent edge density in the input LDR, even when the underlying HDR scene contains substantial structure. In such bright-input, moderate-t… view at source ↗
read the original abstract

Recent generative methods for single-shot high dynamic range (HDR) image reconstruction show promising results, but often struggle with preserving fidelity to the input image. They require separate models to handle highlights and shadows, or sacrifice interpretability by directly predicting the final HDR image. We address these limitations by re-casting single-shot HDR reconstruction as conditional video generation and fusing the generated frames into an HDR image. We finetune a video diffusion model to generate an exposure bracket, conditioned on a low dynamic range (LDR) input. We fuse this image bracket using per-pixel weights predicted by a light-weight UNet. This formulation is simple, interpretable, and effective. Rather than directly hallucinating an HDR image, it explicitly reconstructs the intermediate exposure stack and fuses it into the final output. Our method eliminates the need for separate models across exposure regimes and produces HDR reconstructions with high input fidelity. On quantitative benchmarks, we outperform state-of-the-art generative baselines with comparable model capacity on several reconstruction metrics. Human evaluators further prefer our results in 72% of pairwise comparisons against existing methods. Finally, we show that this input-conditioned sequence generation and fusion framework extends beyond HDR to other image reconstruction tasks, such as all-in-focus image recovery from a single defocus-blurred input.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper re-casts single-shot HDR reconstruction as conditional video generation: a video diffusion model is fine-tuned to produce an exposure bracket conditioned on a single LDR input, after which a lightweight UNet predicts per-pixel fusion weights to recover the final HDR image. The approach is presented as simple and interpretable because it explicitly reconstructs the intermediate bracket rather than directly hallucinating HDR values. The manuscript reports quantitative outperformance over generative baselines of comparable capacity on several reconstruction metrics and states that human evaluators prefer the results in 72% of pairwise comparisons; it also claims the framework extends to other reconstruction tasks such as all-in-focus recovery.

Significance. If the generated brackets prove sufficiently consistent, the method supplies an interpretable route for transferring video diffusion priors to static multi-exposure problems without separate highlight/shadow models, while preserving input fidelity. The reported human preference and extension to other tasks would indicate practical utility beyond current generative HDR baselines.

major comments (3)
  1. [Method] Method section (exposure-bracket generation): the central assumption that fine-tuning a video diffusion model on LDR inputs yields frames differing only by exposure level and remaining scene-consistent is load-bearing, yet the manuscript provides no bracket-level consistency metrics (e.g., optical-flow error, feature-matching scores, or per-pixel variance across the generated stack) nor any ablation replacing the video prior with independent image diffusion.
  2. [Experiments] Experiments section: the claim of outperformance “on several reconstruction metrics” with “comparable model capacity” is stated without enumerating the exact metrics (PSNR, SSIM, HDR-VDP, etc.), the datasets, the precise baselines, or the capacity measurements, preventing verification that the quantitative superiority supports the central claim.
  3. [Experiments] Human evaluation: the 72% pairwise preference is reported without the number of evaluators, the protocol for selecting comparison pairs, or statistical significance testing, which is required to assess whether this result reliably corroborates the method’s advantage.
minor comments (2)
  1. [Abstract] The abstract refers to “several reconstruction metrics” without listing them; an explicit enumeration in the abstract or a table reference would improve clarity.
  2. [Method] Notation for the lightweight UNet fusion weights is introduced without an equation number or diagram; adding a simple equation or figure label would aid reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We appreciate the referee's detailed review and constructive suggestions. Below, we provide point-by-point responses to the major comments and outline the revisions made to the manuscript.

read point-by-point responses

  1. Referee: [Method] Method section (exposure-bracket generation): the central assumption that fine-tuning a video diffusion model on LDR inputs yields frames differing only by exposure level and remaining scene-consistent is load-bearing, yet the manuscript provides no bracket-level consistency metrics (e.g., optical-flow error, feature-matching scores, or per-pixel variance across the generated stack) nor any ablation replacing the video prior with independent image diffusion.

    Authors: We concur that validating the scene consistency of the generated exposure brackets is essential for supporting the method's core premise. To address this, we have incorporated bracket-level consistency metrics into the revised manuscript, specifically reporting the average optical flow error between generated frames and the per-pixel standard deviation across the exposure stack. These additions demonstrate that the generated frames maintain high consistency, differing primarily in exposure levels. Additionally, we have included an ablation study comparing our video diffusion approach to independent image diffusion generations (using the same backbone but without temporal modeling), which shows a clear degradation in both consistency and final HDR quality, thereby justifying the use of the video prior. revision: yes

  2. Referee: [Experiments] Experiments section: the claim of outperformance “on several reconstruction metrics” with “comparable model capacity” is stated without enumerating the exact metrics (PSNR, SSIM, HDR-VDP, etc.), the datasets, the precise baselines, or the capacity measurements, preventing verification that the quantitative superiority supports the central claim.

    Authors: We regret the omission of these specifics in the initial submission, which indeed hinders verification. In the revised manuscript, we have detailed the exact metrics employed—PSNR, SSIM, HDR-VDP-2, and LPIPS—along with the evaluation datasets (HDR-Eye, the dataset from Kalantari et al., and a held-out test set from our training data). We also specify the baselines and capacity comparisons via parameter counts and inference FLOPs. These clarifications confirm that our approach achieves superior performance with comparable model capacity. revision: yes

  3. Referee: [Experiments] Human evaluation: the 72% pairwise preference is reported without the number of evaluators, the protocol for selecting comparison pairs, or statistical significance testing, which is required to assess whether this result reliably corroborates the method’s advantage.

    Authors: We acknowledge that additional details on the human study are necessary for proper assessment. The revised paper now specifies the number of evaluators, the protocol for selecting comparison pairs (random sampling from the test set with balanced conditions), and the results of statistical significance testing, confirming the reliability of the 72% preference rate. revision: yes

Circularity Check

0 steps flagged

No circularity: standard fine-tuning of video diffusion prior plus separate lightweight fusion network

full rationale

The paper's derivation consists of (1) re-casting HDR recovery as conditional video generation, (2) fine-tuning an off-the-shelf video diffusion model on LDR-to-exposure-bracket pairs, and (3) training a separate lightweight UNet to predict per-pixel fusion weights. None of these steps reduce to self-definition, fitted parameters renamed as predictions, or load-bearing self-citations. The central claims are empirical (quantitative metrics and human preference) and rest on the external video diffusion architecture plus the added fusion module; they are not forced by construction from the inputs or prior self-work. This is a conventional ML pipeline with no detectable circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on the assumption that pre-trained video diffusion models can be fine-tuned to produce usable exposure brackets from LDR inputs and that the resulting bracket contains enough information for a simple fusion network to succeed. No new physical entities are introduced.

free parameters (1)
  • UNet fusion weights
    The per-pixel blending weights are learned parameters of the lightweight UNet trained on the generated brackets.
axioms (1)
  • domain assumption Video diffusion models, when conditioned on an LDR image, can generate a temporally consistent exposure bracket suitable for HDR fusion.
    This is the core modeling assumption that allows the re-casting of HDR as sequence generation.

pith-pipeline@v0.9.0 · 5529 in / 1516 out tokens · 46280 ms · 2026-05-13T01:41:13.326133+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We fine-tune a video diffusion model to generate an exposure bracket, conditioned on a low dynamic range (LDR) input. We fuse this image bracket using per-pixel weights predicted by a light-weight UNet.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 5 internal anchors

  1. [1]

    FirstName Alpher , title =

    work page

  2. [2]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

    work page

  3. [3]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

    work page

  4. [4]

    FirstName Alpher and FirstName Gamow , title =

    work page

  5. [5]

    Computer Vision -- ECCV 2022 , year =

    work page 2022

  6. [6]

    In: Proceedings of the 24th Annual Conference on Computer Graphics and Interactive Techniques

    Debevec, Paul E. and Malik, Jitendra , title =. Proceedings of the 24th Annual Conference on Computer Graphics and Interactive Techniques , pages =. 1997 , isbn =. doi:10.1145/258734.258884 , abstract =

    work page doi:10.1145/258734.258884 1997

  7. [7]

    15th Pacific Conference on Computer Graphics and Applications (PG'07) , pages=

    Exposure fusion , author=. 15th Pacific Conference on Computer Graphics and Applications (PG'07) , pages=. 2007 , organization=

    work page 2007

  8. [8]

    ACM transactions on graphics (TOG) , volume=

    HDR image reconstruction from a single exposure using deep CNNs , author=. ACM transactions on graphics (TOG) , volume=. 2017 , publisher=

    work page 2017

  9. [9]

    Computer graphics forum , volume=

    Expandnet: A deep convolutional neural network for high dynamic range expansion from low dynamic range content , author=. Computer graphics forum , volume=. 2018 , organization=

    work page 2018

  10. [10]

    High Dynamic Range Imaging , booktitle =

    Mantiuk, Rafa. High Dynamic Range Imaging , booktitle =. doi:https://doi.org/10.1002/047134608X.W8265 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/047134608X.W8265 , year =

    work page doi:10.1002/047134608x.w8265

  11. [11]

    Advances in neural information processing systems , volume=

    Improved distribution matching distillation for fast image synthesis , author=. Advances in neural information processing systems , volume=

    work page

  12. [12]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Single-image HDR reconstruction by learning to reverse the camera pipeline , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  13. [13]

    Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    Single-image hdr reconstruction by multi-exposure generation , author=. Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    work page

  14. [14]

    ACM Transactions on Graphics , volume=

    Single-shot HDR using conventional image sensor shutter functions and optical randomization , author=. ACM Transactions on Graphics , volume=. 2025 , publisher=

    work page 2025

  15. [15]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Lediff: Latent exposure diffusion for hdr generation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  16. [16]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    UltraFusion: Ultra high dynamic imaging using exposure fusion , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  17. [17]

    arXiv preprint arXiv:2602.04814 , year=

    X2HDR: HDR Image Generation in a Perceptually Uniform Space , author=. arXiv preprint arXiv:2602.04814 , year=

    work page arXiv

  18. [18]

    arXiv preprint arXiv:2510.07741 , year=

    Ultraled: Learning to see everything in ultra-high dynamic range scenes , author=. arXiv preprint arXiv:2510.07741 , year=

    work page arXiv

  19. [19]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Repurposing pre-trained video diffusion models for event-based video interpolation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  20. [20]

    ACM Transactions on Graphics (TOG) , volume=

    Generating the Past, Present and Future from a Motion-Blurred Image , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    work page 2025

  21. [21]

    DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models

    DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models , author=. arXiv preprint arXiv:2604.06161 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    International Conference on Medical image computing and computer-assisted intervention , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. International Conference on Medical image computing and computer-assisted intervention , pages=. 2015 , organization=

    work page 2015

  23. [23]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Training neural networks on RAW and HDR images for restoration tasks , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  24. [24]

    Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    Learning to Refocus with Video Diffusion Models , author=. Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    work page 2025

  25. [25]

    , author=

    Deep high dynamic range imaging of dynamic scenes. , author=. ACM Trans. Graph. , volume=

    work page

  26. [26]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Attention-guided network for ghost-free high dynamic range imaging , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  27. [27]

    IEEE Transactions on Image Processing , volume=

    HDR-GAN: HDR image reconstruction from multi-exposed LDR images with large motions , author=. IEEE Transactions on Image Processing , volume=. 2021 , publisher=

    work page 2021

  28. [28]

    Advances in neural information processing systems , volume=

    Elucidating the design space of diffusion-based generative models , author=. Advances in neural information processing systems , volume=

    work page

  29. [29]

    European Conference on computer vision , pages=

    Ghost-free high dynamic range imaging with context-aware transformer , author=. European Conference on computer vision , pages=. 2022 , organization=

    work page 2022

  30. [30]

    Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets

    Stable video diffusion: Scaling latent video diffusion models to large datasets , author=. arXiv preprint arXiv:2311.15127 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  31. [31]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Wan: Open and advanced large-scale video generative models , author=. arXiv preprint arXiv:2503.20314 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    2304.13625 , archivePrefix=

    HDR-VDP-3: A multi-metric for predicting image differences, quality and contrast distortions in high dynamic range and regular content , author=. arXiv preprint arXiv:2304.13625 , year=

    work page arXiv

  33. [33]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Perceptual assessment and optimization of HDR image rendering , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  34. [34]

    Color and imaging conference , volume=

    The HDR photographic survey , author=. Color and imaging conference , volume=. 2007 , organization=

    work page 2007

  35. [35]

    Classifier-Free Diffusion Guidance

    Classifier-free diffusion guidance , author=. arXiv preprint arXiv:2207.12598 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [36]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Rawhdr: High dynamic range image reconstruction from a single raw image , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  37. [37]

    ACM SIGGRAPH 2022 conference proceedings , pages=

    Comparison of single image HDR reconstruction methods—the caveats of quality assessment , author=. ACM SIGGRAPH 2022 conference proceedings , pages=

    work page 2022

  38. [38]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Repurposing diffusion-based image generators for monocular depth estimation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  39. [39]

    Computer Graphics Forum , volume=

    Bracket diffusion: Hdr image generation by consistent ldr denoising , author=. Computer Graphics Forum , volume=. 2025 , organization=

    work page 2025

  40. [40]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  41. [41]

    ECCV , year =

    Silberman, Nathan and Hoiem, Derek and Kohli, Pushmeet and Fergus, Rob , title=. ECCV , year =

    work page

  42. [42]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Gcc: Generative color constancy via diffusing a color checker , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  43. [43]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Deep High Dynamic Range Imaging with Large Foreground Motions , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    work page

  44. [44]

    European Conference on Computer Vision (ECCV) , pages=

    Ghost-free High Dynamic Range Imaging with Context-aware Transformer , author=. European Conference on Computer Vision (ECCV) , pages=. 2022 , organization=

    work page 2022

  45. [45]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Single image depth estimation trained via depth from defocus cues , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  46. [46]

    Journal of graphics tools , volume=

    Parameter estimation for photographic tone reproduction , author=. Journal of graphics tools , volume=. 2002 , publisher=

    work page 2002

  47. [47]

    Proceedings of the European Conference on Computer Vision (ECCV) , year=

    SAFNet: Selective Alignment Fusion Network for Efficient HDR Imaging , author=. Proceedings of the European Conference on Computer Vision (ECCV) , year=

    work page

  48. [48]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    AFUNet: Cross-Iterative Alignment-Fusion Synergy for HDR Reconstruction via Deep Unfolding Paradigm , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    work page

  49. [49]

    Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages=

    DeepFuse: A Deep Unsupervised Approach for Exposure Fusion with Extreme Exposure Image Pairs , author=. Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages=

    work page

  50. [50]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    MEFLUT: Unsupervised 1D Lookup Tables for Multi-exposure Image Fusion , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  51. [51]

    IEEE Transactions on Computational Imaging , volume=

    Robust estimation of exposure ratios in multi-exposure image stacks , author=. IEEE Transactions on Computational Imaging , volume=. 2023 , publisher=

    work page 2023

  52. [52]

    ACM SIGGRAPH Computer Graphics , volume=

    A lens and aperture camera model for synthetic image generation , author=. ACM SIGGRAPH Computer Graphics , volume=. 1981 , publisher=

    work page 1981

  53. [53]

    HDR Video Generation via Latent Alignment with Logarithmic Encoding

    HDR Video Generation via Latent Alignment with Logarithmic Encoding , author=. arXiv preprint arXiv:2604.11788 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  54. [54]

    2011 , publisher=

    The art and science of HDR imaging , author=. 2011 , publisher=

    work page 2011

  55. [55]

    Vision system with high dynamic range for optical surface defect inspection , volume =

    Zhaolou Cao and Fenping Cui and Chunjie Zhai , journal =. Vision system with high dynamic range for optical surface defect inspection , volume =. 2018 , url =

    work page 2018

  56. [56]

    Proceedings IEEE Conference on Computer Vision and Pattern Recognition

    High dynamic range imaging: Spatially varying pixel exposures , author=. Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No. PR00662) , volume=. 2000 , organization=

    work page 2000

  57. [57]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Deep optics for single-shot high-dynamic-range imaging , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  58. [58]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Learning rank-1 diffractive optics for single-shot high dynamic range imaging , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  59. [59]

    ACM Transactions on Graphics (TOG) , volume=

    Split-aperture 2-in-1 computational cameras , author=. ACM Transactions on Graphics (TOG) , volume=. 2024 , publisher=

    work page 2024

  60. [60]

    2022 IEEE International Conference on Computational Photography (ICCP) , pages=

    Mantissacam: Learning snapshot high-dynamic-range imaging with perceptually-based in-pixel irradiance encoding , author=. 2022 IEEE International Conference on Computational Photography (ICCP) , pages=. 2022 , organization=

    work page 2022

  61. [61]

    2024 , howpublished=

    Black Forest Labs , title=. 2024 , howpublished=

    work page 2024

  62. [62]

    a talk at the Stanford Artificial Project in , volume=

    A 3x3 isotropic gradient operator for image processing , author=. a talk at the Stanford Artificial Project in , volume=

    work page

  63. [63]

    CoRR abs/2004.07728(2020),https://arxiv.org/abs/ 2004.07728

    Keyan Ding and Kede Ma and Shiqi Wang and Eero P. Simoncelli , title =. CoRR , volume =. 2020 , url =. 2004.07728 , timestamp =

    work page arXiv 2020

  64. [64]

    IEEE Transactions on Image processing , volume=

    Image fusion with guided filtering , author=. IEEE Transactions on Image processing , volume=. 2013 , publisher=

    work page 2013

  65. [65]

    IEEE Transactions on Image Processing , volume=

    Robust all-in-focus super-resolution for focal stack photography , author=. IEEE Transactions on Image Processing , volume=. 2016 , publisher=

    work page 2016