pith. sign in

arxiv: 2605.17557 · v1 · pith:FGPC4CVZnew · submitted 2026-05-17 · 💻 cs.GR · cs.CV

Real-Time Neural Hair Denoising

Chenghao Wu , Yuefan Shen , Tao Huang , Kai Yan , Zahra Montazeri , Kui Wu This is my paper

Pith reviewed 2026-05-19 22:18 UTC · model grok-4.3

classification 💻 cs.GR cs.CV
keywords neural renderinghair reconstructionreal-time denoisingG-buffersstrand-based hairdeferred shadingtemporal accumulationhair rendering
0
0 comments X

The pith

Neural spatial and temporal reconstruction recovers accurate hair coverage and tangents from undersampled inputs to enable high-quality real-time shading.

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

The paper develops a lightweight real-time pipeline to reconstruct strand-based hair G-Buffers from severely undersampled rasterized inputs. It begins with neural spatial reconstruction and temporal accumulation to obtain fractional hair visibility and tangent directions. A tangent-guided step then completes the position data for use in physically based deferred shading. Sympathetic readers would care because hair strands are difficult to render accurately with limited samples, often causing visual artifacts in real-time graphics.

Core claim

We propose a lightweight real-time method for reconstructing strand-based hair G-Buffers from severely undersampled rasterized inputs. Our pipeline first applies neural spatial reconstruction and temporal accumulation to recover hair coverage, i.e., fractional hair visibility within a pixel, and tangent. It then uses a tangent-guided reconstruction step to complete the position, which is subsequently used for physically based deferred hair shading. We evaluate our method across a diverse set of hairstyles, including straight, wavy, afro, and ponytail styles, under both static and dynamic scenarios. Our method achieves higher hair reconstruction quality than existing hair-specific denoising 1

What carries the argument

Neural spatial reconstruction and temporal accumulation to recover hair coverage and tangent, followed by tangent-guided position reconstruction.

If this is right

  • The method produces higher reconstruction quality than hair-specific denoising techniques and general solutions such as DLSS and FSR.
  • It works across diverse hairstyles including straight, wavy, afro, and ponytail under static and dynamic conditions.
  • Reconstructed positions support physically based deferred hair shading.
  • The lightweight design enables real-time performance in graphics applications.

Where Pith is reading between the lines

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

  • This technique for thin geometry reconstruction might extend to other fine details such as fur or grass in real-time scenes.
  • Lower sample rates could become viable in production pipelines while maintaining visual quality.
  • Game engine integration could improve hair appearance during rapid character motion without added compute cost.
  • Performance on entirely new motion patterns or lighting would be a useful next verification step.

Load-bearing premise

The neural spatial reconstruction and temporal accumulation steps can reliably recover accurate fractional hair visibility and tangent from severely undersampled rasterized inputs across diverse hairstyles and motion.

What would settle it

A direct comparison showing inaccurate tangent directions or coverage on fast-moving afro or ponytail hairstyles with very low sample counts compared to ground truth would falsify the reliability claim.

Figures

Figures reproduced from arXiv: 2605.17557 by Chenghao Wu, Kai Yan, Kui Wu, Tao Huang, Yuefan Shen, Zahra Montazeri.

Figure 1
Figure 1. Figure 1: We showcase the denoising results of our method given severely undersampled strand-based hair G-buffers. Compared with TAA, DLSS, and FSR, our approach better preserves sparse hair silhouettes and fine strand details, producing results closer to the high-sample reference. Insets show the corresponding error maps, and blue boxes indicate zoomed regions. We propose a lightweight real-time method for reconstr… view at source ↗
Figure 2
Figure 2. Figure 2: The pipeline of our neural hair denoiser. Starting from low-sample, aliased hair renderings, we extract noisy hair G-buffers, including coverage C𝑖 , tangent T𝑖 , position P𝑖 , and depth D𝑖 . A spatiotemporal neural denoiser is applied to tangent and coverage through a spatial module 𝚿 and a temporal module 𝚪, while the noisy depth and position are directly fed into the reconstruction stage. Using consecut… view at source ↗
Figure 3
Figure 3. Figure 3: Left: given an empty pixel p with reconstructed tangent direc￾tion, we estimate its curvature center at cp. For its 3 × 3 neighborhood Np, we identify the neighbor s★ whose curvature center is closest to cp and propagate it to p. Right: from valid pixels s𝑎 and s𝑏 , which have already been marked as hair, we propagate along the tangent direction. For each reached missing pixel, we assign the value with the… view at source ↗
Figure 4
Figure 4. Figure 4: Comparisons on different static hairstyles. The bottom-left corner shows the error map. Input TAA DLSS FSR Ours Ref Frame 0 Frame 50 Frame 150 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison on dynamic hair sequence. Input w/o spatial w/o temporal w/o analytic FullGym (Ours) Ref Frame 0 Frame 50 Frame 150 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation studies on components of our method. The corresponding quantitative comparison is provided in the supplemental document [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Different color and lighting. The bottom-left corner shows the error map. Our method remains robust and yields the smallest error across different hair colors and lighting conditions, as it operates on G-buffers. The top-left number indicates the corresponding PSNR relative to the reference. Input (whole view) Input TAA DLSS FSR Ours Ref [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Zoom in comparison of extremely curved and highly tangled hairstyle under back lighting. These results demonstrate the benefit of reconstructing geometry-related G-buffers before shading, rather than directly filtering the final RGB image. Input Denoised Ref Denoised Zoom-in Ref Zoom-in Diff Zoom-in Currius et al. Ours [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison with the neural hair denoiser of Currius et al. [2022] on the same scene. Note that the input and reference images differ from ours because their method uses a different hair-shading model with direct illumination. Their method produces less accurate and blurred shading in regions with complex strand overlap, leading to a much larger perceptual error as measured by LPIPS (0.0622 versus our 0.009… view at source ↗
read the original abstract

We propose a lightweight real-time method for reconstructing strand-based hair G-Buffers from severely undersampled rasterized inputs. Our pipeline first applies neural spatial reconstruction and temporal accumulation to recover hair coverage, i.e., fractional hair visibility within a pixel, and tangent. It then uses a tangent-guided reconstruction step to complete the position, which is subsequently used for physically based deferred hair shading. We evaluate our method across a diverse set of hairstyles, including straight, wavy, afro, and ponytail styles, under both static and dynamic scenarios. Our method achieves higher hair reconstruction quality than existing hair-specific denoising techniques and general industrial neural reconstruction solutions such as DLSS and FSR.

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

2 major / 2 minor

Summary. The paper proposes a lightweight real-time neural pipeline for reconstructing strand-based hair G-buffers from severely undersampled rasterized inputs. The method first applies neural spatial reconstruction and temporal accumulation to recover fractional hair visibility (coverage) and tangent, then performs tangent-guided position reconstruction, and finally applies physically based deferred shading. It reports evaluation across straight, wavy, afro, and ponytail hairstyles in both static and dynamic scenarios, claiming higher reconstruction quality than existing hair-specific denoisers and general solutions such as DLSS and FSR.

Significance. If the central claims hold under quantitative scrutiny, the work could meaningfully advance real-time hair rendering in graphics applications by enabling higher-fidelity reconstruction of thin geometry without prohibitive sampling costs. The combination of learned spatial-temporal recovery with explicit tangent guidance for position completion is a targeted contribution to handling sub-pixel hair strands.

major comments (2)
  1. [Abstract] Abstract: the claim of superior quality over hair-specific denoisers and DLSS/FSR is presented without any quantitative metrics (e.g., PSNR, SSIM, or per-component error on visibility/tangent), error bars, or details on the training distribution and loss functions. This directly affects verifiability of the headline result, especially given that the method's quality advantage is asserted to rest on accurate recovery of fractional visibility and tangent from undersampled inputs.
  2. [Method and Evaluation] The weakest link for the quality claim is the neural spatial reconstruction plus temporal accumulation step that must produce reliable fractional hair visibility and tangent from severely undersampled raster inputs. Hair strands are sub-pixel thin; any systematic bias here propagates directly into the tangent-guided position reconstruction and deferred shading. The manuscript should provide targeted quantitative evaluation (e.g., ground-truth comparison of recovered coverage and tangent at varying sampling rates) rather than relying solely on final shaded-image comparisons.
minor comments (2)
  1. Clarify the exact network architecture, input feature channels, and temporal accumulation scheme (e.g., how many frames are used and whether motion vectors are incorporated) to support reproducibility.
  2. Add a limitations discussion addressing performance on extreme motion, very dense hairstyles, or lighting conditions outside the training distribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and have revised the paper to strengthen the quantitative presentation of our results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of superior quality over hair-specific denoisers and DLSS/FSR is presented without any quantitative metrics (e.g., PSNR, SSIM, or per-component error on visibility/tangent), error bars, or details on the training distribution and loss functions. This directly affects verifiability of the headline result, especially given that the method's quality advantage is asserted to rest on accurate recovery of fractional visibility and tangent from undersampled inputs.

    Authors: We agree that the abstract would benefit from explicit quantitative references to improve verifiability. We have revised the abstract to include summary metrics (e.g., average PSNR/SSIM gains and per-component errors on visibility/tangent) drawn from our evaluations. Full details on the training distribution (diverse hairstyles with static/dynamic sequences), loss functions (L1 plus perceptual terms for reconstruction), and error bars (from repeated runs) are provided in Sections 3 and 4; we have added cross-references in the abstract for clarity. These changes address the concern without exceeding length limits. revision: yes

  2. Referee: [Method and Evaluation] The weakest link for the quality claim is the neural spatial reconstruction plus temporal accumulation step that must produce reliable fractional hair visibility and tangent from severely undersampled raster inputs. Hair strands are sub-pixel thin; any systematic bias here propagates directly into the tangent-guided position reconstruction and deferred shading. The manuscript should provide targeted quantitative evaluation (e.g., ground-truth comparison of recovered coverage and tangent at varying sampling rates) rather than relying solely on final shaded-image comparisons.

    Authors: We concur that targeted evaluation of the intermediate neural step is essential to substantiate the pipeline. While end-to-end shaded results demonstrate practical utility, we have added a new subsection (4.2) with ground-truth comparisons of recovered coverage and tangent at sampling rates from 1spp to 8spp. This includes aggregate errors (MAE), per-pixel error visualizations, and bias analysis, confirming low systematic error in the spatial-temporal reconstruction. These additions directly validate that accurate fractional visibility and tangent recovery underpin the subsequent position completion and shading improvements. revision: yes

Circularity Check

0 steps flagged

No circularity: pipeline is a trained neural reconstruction with independent evaluation

full rationale

The paper describes a new neural pipeline (spatial reconstruction + temporal accumulation for visibility/tangent, followed by tangent-guided position completion and deferred shading) trained to recover G-Buffers from undersampled raster inputs. No equations are presented that define a target quantity in terms of itself, rename a fitted parameter as a 'prediction', or reduce the central quality claim to a self-citation chain. The method is externally evaluated against DLSS, FSR, and hair-specific denoisers on diverse hairstyles; the derivation does not collapse to tautology or fitted-input renaming. This is the normal non-circular case for a learned reconstruction technique.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the assumption that a lightweight neural network can accurately infer hair coverage, tangent, and position from undersampled inputs; no free parameters, axioms, or invented entities are explicitly listed in the abstract.

pith-pipeline@v0.9.0 · 5642 in / 1124 out tokens · 18872 ms · 2026-05-19T22:18:05.706416+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

59 extracted references · 59 canonical work pages

  1. [1]

    2023 , eprint =

    Robust Average Networks for Monte Carlo Denoising , author =. 2023 , eprint =. doi:10.48550/arXiv.2310.04080 , url =

    work page doi:10.48550/arxiv.2310.04080 2023

  2. [2]

    Kajiya, J. T. and Kay, T. L. , title =. SIGGRAPH Comput. Graph. , month =. 1989 , issue_date =

    work page 1989

  3. [3]

    Marschner and Henrik Wann Jensen and Mike Cammarano and Steve Worley and Pat Hanrahan , title =

    Stephen R. Marschner and Henrik Wann Jensen and Mike Cammarano and Steve Worley and Pat Hanrahan , title =

    work page

  4. [4]

    ACM SIGGRAPH 2010 Papers , articleno =

    Sadeghi, Iman and Pritchett, Heather and Jensen, Henrik Wann and Tamstorf, Rasmus , title =. ACM SIGGRAPH 2010 Papers , articleno =. 2010 , isbn =

    work page 2010

  5. [5]

    A practical approach for photometric acquisition of hair color , journal =

    Arno Zinke and Martin Rump and Tom. A practical approach for photometric acquisition of hair color , journal =

    work page

  6. [6]

    An Energy-Conserving Hair Reflectance Model , journal =

    Eugene d'Eon and Guillaume Fran. An Energy-Conserving Hair Reflectance Model , journal =

    work page

  7. [7]

    A Practical and Controllable Hair and Fur Model for Production Path Tracing , journal =

    Matt Jen. A Practical and Controllable Hair and Fur Model for Production Path Tracing , journal =

    work page

  8. [8]

    ACM Trans

    Yan, Ling-Qi and Tseng, Chi-Wei and Jensen, Henrik Wann and Ramamoorthi, Ravi , title =. ACM Trans. Graph. , month =. 2015 , issue_date =

    work page 2015

  9. [9]

    Pramook Khungurn and Daniel Schroeder and Shuang Zhao and Kavita Bala and Steve Marschner , title =

    work page

  10. [10]

    Computer Graphics Forum , year =

    A Practical and Hierarchical Yarn-based Shading Model for Cloth , author =. Computer Graphics Forum , year =

    work page

  11. [11]

    and Hullin, M

    Huang, W. and Hullin, M. B. and Hanika, J. , title =. Computer Graphics Forum , volume =

    work page

  12. [12]

    ACM Trans

    Xia, Mengqi (Mandy) and Walter, Bruce and Michielssen, Eric and Bindel, David and Marschner, Steve , title =. ACM Trans. Graph. , month =. 2020 , issue_date =

    work page 2020

  13. [13]

    ACM Trans

    Xia, Mengqi and Walter, Bruce and Hery, Christophe and Maury, Olivier and Michielssen, Eric and Marschner, Steve , title =. ACM Trans. Graph. , month =. 2023 , issue_date =

    work page 2023

  14. [14]

    Moon and Stephen R

    Jonathan T. Moon and Stephen R. Marschner , title =

    work page

  15. [15]

    ACM Trans

    Khungurn, Pramook and Marschner, Steve , title =. ACM Trans. Graph. , month =. 2017 , issue_date =

    work page 2017

  16. [16]

    and Walter, Bruce and Marschner, Steve , title =

    Moon, Jonathan T. and Walter, Bruce and Marschner, Steve , title =. ACM SIGGRAPH 2008 Papers , articleno =. 2008 , isbn =

    work page 2008

  17. [17]

    KT, Aakash and Jarabo, Adrian and Aliaga, Carlos and Chiang, Matt Jen-Yuan and Maury, Olivier and Hery, Christophe and Narayanan, P. J. and Nam, Giljoo , title =. Computer Graphics Forum , volume =

    work page

  18. [18]

    Arno Zinke and Cem Yuksel and Andreas Weber and John Keyser , title =

    work page

  19. [19]

    Computer Graphics Forum (Proceedings of EUROGRAPHICS 2008) , year =

    Cem Yuksel and John Keyser , title =. Computer Graphics Forum (Proceedings of EUROGRAPHICS 2008) , year =

    work page 2008

  20. [20]

    ACM SIGGRAPH 2021 Talks , year=

    A Deep Dive into Nanite Virtualized Geometry , author=. ACM SIGGRAPH 2021 Talks , year=

    work page 2021

  21. [21]

    , title =

    Garland, Michael and Heckbert, Paul S. , title =. Proceedings of the 24th Annual Conference on Computer Graphics and Interactive Techniques , pages =. 1997 , isbn =

    work page 1997

  22. [22]

    Proceedings of the 23rd Annual Conference on Computer Graphics and Interactive Techniques , pages =

    Hoppe, Hugues , title =. Proceedings of the 23rd Annual Conference on Computer Graphics and Interactive Techniques , pages =. 1996 , isbn =

    work page 1996

  23. [23]

    and Reddy, Martin and Varshney, Amitabh , title =

    Luebke, David and Watson, Benjamin and Cohen, Jonathan D. and Reddy, Martin and Varshney, Amitabh , title =. 2002 , isbn =

    work page 2002

  24. [24]

    ACM Trans

    Cohen-Steiner, David and Alliez, Pierre and Desbrun, Mathieu , title =. ACM Trans. Graph. , month =. 2004 , issue_date =

    work page 2004

  25. [25]

    ISPRS Journal of Photogrammetry and Remote Sensing , keywords =

    Feature-preserving 3D mesh simplification for urban buildings. ISPRS Journal of Photogrammetry and Remote Sensing , keywords =

    work page

  26. [26]

    PolyFit: Polygonal Surface Reconstruction from Point Clouds , year=

    Nan, Liangliang and Wonka, Peter , booktitle=. PolyFit: Polygonal Surface Reconstruction from Point Clouds , year=

    work page

  27. [27]

    ACM Trans

    Bauchet, Jean-Philippe and Lafarge, Florent , title =. ACM Trans. Graph. , month =. 2020 , issue_date =

    work page 2020

  28. [28]

    ACM Trans

    Weier, Philippe and Zirr, Tobias and Kaplanyan, Anton and Yan, Ling-Qi and Slusallek, Philipp , title =. ACM Trans. Graph. , month =. 2023 , issue_date =

    work page 2023

  29. [29]

    and Kim, T-Y

    Bertails, F. and Kim, T-Y. and Cani, M-P. and Neumann, U. , title =. Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Computer Animation , pages =. 2003 , isbn =

    work page 2003

  30. [30]

    Modeling hair using level-of-detail representations , year=

    Ward, Kelly and Lin, Ming C and Joohi, Lee and Fisher, Susan and Macri, Dean , booktitle=. Modeling hair using level-of-detail representations , year=

    work page

  31. [31]

    Proceedings of the Eurographic Workshop on Computer Animation and Simulation , pages =

    Koh, Chuan Koon and Huang, Zhiyong , title =. Proceedings of the Eurographic Workshop on Computer Animation and Simulation , pages =. 2001 , isbn =

    work page 2001

  32. [32]

    The process of creating volumetric-based materials in uncharted 4 , title =

    Jiang, Yibing , journal=. The process of creating volumetric-based materials in uncharted 4 , title =. 2016 , volume =

    work page 2016

  33. [33]

    A Practical Ply-Based Appearance Model of Woven Fabrics , year =

    Montazeri, Zahra and Gammelmark, S. A Practical Ply-Based Appearance Model of Woven Fabrics , year =. ACM Trans. Graph. , month =

    work page

  34. [34]

    A Practical Ply-Based Appearance Modeling for Knitted Fabrics

    Zahra Montazeri and Gammelmark, \ Soren B.\ and Jensen, \ Henrik Wann\ and Shuang Zhao. A Practical Ply-Based Appearance Modeling for Knitted Fabrics. Eurographics Symposium on Rendering. 2021

    work page 2021

  35. [35]

    and Halstead, John and Planck, Maxwell and Ryu, David , title =

    Cook, Robert L. and Halstead, John and Planck, Maxwell and Ryu, David , title =. ACM Trans. Graph. , month =. 2007 , issue_date =

    work page 2007

  36. [36]

    Practical level-of-detail aggregation of fur appearance , journal =

    Junqiu Zhu and Sizhe Zhao and Lu Wang and Yanning Xu and Ling. Practical level-of-detail aggregation of fur appearance , journal =

    work page

  37. [37]

    and Zhou, Y

    Huang, T. and Zhou, Y. and Lin, D. and Zhu, J. and Yan, L. and Wu, K. , title =. Computer Graphics Forum , volume =

    work page

  38. [38]

    ACM Trans

    Huang, Tao and Shi, Haoyang and Wang, Mengdi and Qiu, Yuxing and Yang, Yin and Wu, Kui , title =. ACM Trans. Graph. , month = jul, articleno =. 2025 , issue_date =

    work page 2025

  39. [39]

    ACM Trans

    Zheng, Zhongtian and Huang, Tao and Su, Haozhe and Ma, Xueqi and Shen, Yuefan and Wang, Tongtong and Yang, Yin and Gao, Xifeng and Pan, Zherong and Wu, Kui , title =. ACM Trans. Graph. , month = dec, articleno =. 2025 , issue_date =

    work page 2025

  40. [40]

    2025 , howpublished =

    DLSS 4: Transforming Real-Time Graphics with AI , author =. 2025 , howpublished =

    work page 2025

  41. [41]

    AMD FidelityFX Super Resolution , year =

    work page

  42. [42]

    Computer Graphics Forum , title =

    Huang, Tao and Yuan, JunPing and Hu, Ruike and Wang, Lu and Guo, Yanwen and Chen, Bin and Guo, Jie and Zhu, Junqiu , year =. Computer Graphics Forum , title =

    work page

  43. [43]

    ACM SIGGRAPH 2024 Conference Papers , articleno =

    Bhokare, Gaurav and Montalvo, Eisen and Diaz, Elie and Yuksel, Cem , title =. ACM SIGGRAPH 2024 Conference Papers , articleno =. 2024 , isbn =

    work page 2024

  44. [44]

    and Assarsson, Ulf and Sintorn, Erik , title =

    Currius, Roc R. and Assarsson, Ulf and Sintorn, Erik , title =. Proc. ACM Comput. Graph. Interact. Tech. , month = may, articleno =. 2022 , issue_date =

    work page 2022

  45. [45]

    Proceedings of the 20th ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games , pages =

    Wu, Kui and Yuksel, Cem , title =. Proceedings of the 20th ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games , pages =. 2016 , isbn =

    work page 2016

  46. [46]

    ACM Trans

    Yuksel, Cem and Schaefer, Scott and Keyser, John , title =. ACM Trans. Graph. , month = dec, pages =. 2009 , issue_date =

    work page 2009

  47. [47]

    2023 , note =

    Xihao Fu , title =. 2023 , note =

    work page 2023

  48. [48]

    2009 , month = feb, howpublished =

    Timothy Lottes , title =. 2009 , month = feb, howpublished =

    work page 2009

  49. [49]

    Computer Graphics Forum , volume =

    Yang, Lei and Liu, Shiqiu and Salvi, Marco , title =. Computer Graphics Forum , volume =

    work page

  50. [50]

    ACM SIGGRAPH Courses: Advances in Real-Time Rendering in Games Course , year=

    Strand-based Hair Rendering in Frostbite , author=. ACM SIGGRAPH Courses: Advances in Real-Time Rendering in Games Course , year=

    work page

  51. [51]

    ACM SIGGRAPH Courses: Advances in Real-Time Rendering in Games Course , year=

    Strand-based hair and fur rendering in Indiana Jones and the Great Circle , author=. ACM SIGGRAPH Courses: Advances in Real-Time Rendering in Games Course , year=

    work page

  52. [52]

    Unreal Engine , url =

    work page

  53. [53]

    and Munkberg, J

    Hasselgren, J. and Munkberg, J. and Salvi, M. and Patney, A. and Lefohn, A. , title =. Computer Graphics Forum , volume =

    work page

  54. [54]

    and Blanc-Feraud, L

    Charbonnier, P. and Blanc-Feraud, L. and Aubert, G. and Barlaud, M. , booktitle=. Two deterministic half-quadratic regularization algorithms for computed imaging , year=

    work page

  55. [55]

    Denoising with kernel prediction and asymmetric loss functions , year =

    Vogels, Thijs and Rousselle, Fabrice and Mcwilliams, Brian and R\". Denoising with kernel prediction and asymmetric loss functions , year =. ACM Trans. Graph. , month = jul, articleno =

    work page

  56. [56]

    and Ermon, Stefano and Rudra, Atri and R\'

    Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R\'. FLASHATTENTION: fast and memory-efficient exact attention with IO-awareness , year =. Proceedings of the 36th International Conference on Neural Information Processing Systems , articleno =

    work page

  57. [57]

    , booktitle=

    Edison, Anitha and Jiji, C.V. , booktitle=. Optical Acceleration for Motion Description in Videos , year=

    work page

  58. [58]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015. 2015

    work page 2015

  59. [59]

    FovVideoVDP: a visible difference predictor for wide field-of-view video , year =

    Mantiuk, Rafa. FovVideoVDP: a visible difference predictor for wide field-of-view video , year =. ACM Trans. Graph. , month = jul, articleno =

    work page