pith. sign in

arxiv: 2606.22481 · v1 · pith:R6LMUWEEnew · submitted 2026-06-21 · 💻 cs.GR · cs.CV

Lighting-Consistent Object Transfer Across Radiance Fields

Nicol\'as Violante , George Kopanas , Linus Franke , Julien Philip , George Drettakis This is my paper

Pith reviewed 2026-06-26 09:44 UTC · model grok-4.3

classification 💻 cs.GR cs.CV
keywords 3D Gaussian Splattingobject compositinglighting harmonizationdiffusion modelsradiance fieldsscene editingview synthesisimage harmonization
0
0 comments X

The pith

A diffusion model harmonizes lighting when objects are transferred between 3D Gaussian Splatting scenes.

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

The paper tackles the problem that naively pasting an object extracted from one captured scene into another produces unrealistic results because the lighting conditions do not match. It trains a diffusion model on a mixed collection of inconsistent composite images paired with consistent target images drawn from synthetic, generated, and real sources. The pipeline renders multiple views of the target scene containing the pasted object, applies the diffusion model to correct lighting in each view independently, and then merges the corrected views into a single coherent 3D Gaussian Splatting model through a post-optimization step. This yields a practical end-to-end solution for object transfer that works across common capture pipelines.

Core claim

Extracting an object from a source 3DGS scene and compositing it into a target scene produces inconsistent lighting across rendered views; a diffusion model trained on heterogeneous pairs of inconsistent composites and consistent outputs can correct each view separately, after which the harmonized images can be consolidated into a new 3DGS representation by post-optimization, producing visually consistent results.

What carries the argument

Diffusion model trained on inconsistent-composite to consistent-output image pairs, applied per rendered view before 3DGS post-optimization consolidation.

If this is right

  • A user can extract an object from one captured scene and insert it into another while obtaining consistent lighting across all viewpoints.
  • The same diffusion model works on rendered views from any pair of 3DGS scenes without requiring per-scene retraining.
  • The final output is a single editable 3DGS asset usable in standard rendering pipelines for VFX, design, or marketing.
  • Quality exceeds prior object-transfer methods that lacked explicit lighting harmonization.

Where Pith is reading between the lines

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

  • The per-view harmonization step could be replaced by a single 3D-aware model if the diffusion architecture is extended to operate directly on the Gaussian representation.
  • Because the training set mixes synthetic and real data, the approach may transfer to scenes captured under uncontrolled outdoor lighting without additional fine-tuning.
  • The post-optimization step implicitly enforces multi-view consistency; removing it would likely leave residual view-dependent lighting errors.

Load-bearing premise

The diffusion model will generalize from its training pairs to correct lighting accurately on new rendered composites, and the post-optimization step will produce a coherent 3D representation without introducing fresh inconsistencies.

What would settle it

Rendering novel views from the final post-optimized 3DGS and observing visible lighting mismatches or new artifacts between the transferred object and the target scene would show the method has failed to deliver consistent transfer.

Figures

Figures reproduced from arXiv: 2606.22481 by George Drettakis, George Kopanas, Julien Philip, Linus Franke, Nicol\'as Violante.

Figure 1
Figure 1. Figure 1: We transfer an object from one captured 3DGS scene into another and harmonize its lighting so that shadows and reflec￾tions match the new environment. Naively compositing the source object (the car, from Ref-NeRF [VHM*21]) into the target scene (from DL3DV [LST*24]) leaves it inconsistently lit, with missing shadows and wrong reflection highlights (upper row). Our method harmonizes the object and produces … view at source ↗
Figure 2
Figure 2. Figure 2: Object Composition Interface. We propose a user inter￾face for object composition across radiance fields. The user loads the extracted object from the source scene, inserts it into the tar￾get scene, and can control its final position, orientation, and scale within the target scene. Composite images are rendered using the camera viewpoints from the target scene [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Method Overview. Step 1: Single-View Harmonization: Conditioned on the composite image and the object mask, our har￾monization model transforms a noisy latent into a harmonized im￾age with plausible lighting, shadows, and reflections. Step 2: Multi￾View Consolidation: Our 3DGS post-optimization consolidates the independent harmonized single-views into a multi-view consistent representation. 3. Overview Our… view at source ↗
Figure 4
Figure 4. Figure 4: Data for Harmonization Model. Following Eq. 2, we form an input image by compositing a background image of a scene with an object lit under a different illumination. The target im￾age contains the harmonized object within the scene. During train￾ing, pairs of input and target images are sampled on-the-fly. We have three data sources: our synthetic dataset of Blender scenes, a dataset of images generated wi… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative 2D Results. From left to right: Background, Composite, LBM [CTSA25], MV-CoLight [RBX*25], Nano Ba￾nana [Rai25], and Ours. From top to bottom, the composited scenes are: (i) Truck from Tanks and Temples into Garden from Mip-NeRF-360, (ii) Desk from BlenderKit into London Bedroom from Zip-NeRF, (iii) Locomotive from BlenderKit into Train from Tanks and Temples, (iv) Plant from BlenderKit into Bou… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative 3D Results. From left to right: Background, Composite, GS-IR [LZF*24], Gaussian Shader [JTL*24], 3DGS￾DR [YHZ24], and Ours. From top to bottom, the composited scenes are: (i) Truck from Tanks and Temples into Garden from Mip-NeRF-360, (ii) 508850 into 4cea29, both from DL3DV, (iii) Desk from BlenderKit into London Bedroom from Zip-NeRF, (iv) Plant from BlenderKit into Bouquet from LERF, (v) Loc… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative 3D Results. We provide nine high-quality synthetic scenes with ground truth for numerical comparisons. From left to right: Composite, GS-IR [LZF*24], Gaussian Shader [JTL*24], 3DGS-DR [YHZ24], Ours, and Ground Truth. © 2026 The Author(s). Computer Graphics Forum published by Eurographics and John Wiley & Sons Ltd [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Illumination Consistency. Results of our harmonization method on three synthetic scenes with an inserted object at three different rotations. Our method produces consistent multi-view il￾lumination across all configurations. © 2026 The Author(s). Computer Graphics Forum published by Eurographics and John Wiley & Sons Ltd [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Iterative Dataset Update Ablation. We experimented with two iterative dataset update schedules for the consolidation: linear from 1.0 to 0.1 and geometric from 1.0 to 0.1. In this example we perform five dataset updates (only updates 1, 3, and 5 are shown here). But we still found that a single consolidation produced satis￾factory results without having to update the dataset multiple times (see [PITH_FULL… view at source ↗
Figure 10
Figure 10. Figure 10: Photometric vs Perceptual Loss Ablation. For 3DGS post-optimization, a perceptual loss better preserves details and sharp edges. See zoomed-in regions (green) for details [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Limitations. (top) Bonsai into Kitchen, both from Mip￾NeRF-360 [BMV*22]. Contains block-like artifacts and blur due to very small details with high-frequency content. (bottom) Kettle from NERO [LWL*23] into Bouquet from LERF [KKG*23]. De￾spite the harmonization looking realistic, the appearance of the ob￾ject changes from a metallic surface to glass. 8. Conclusions We presented DOT3D, a method that transf… view at source ↗
read the original abstract

3D Gaussian Splatting (3DGS) is widely used to capture and render real scenes. Compositing objects from one capture into another has applications in many domains, such as VFX, architecture and interior design, or marketing. However, extracting an object from a source scene and naively pasting it into a target scene will fail to produce realistic results due to the different lighting conditions between the two scenes. To address this problem, we introduce a diffusion model that harmonizes naively composited images with inconsistent lighting. The model is trained with a heterogeneous dataset of image pairs (inconsistent composite input, consistent output), combining synthetic, generated, and real data. Our complete 3D solution allows a user to extract an object from the source scene and composite it into the target scene. From this, the (inconsistent) views of the target scene with the composite object are rendered. Our diffusion model harmonizes each one of these views, which are finally consolidated in a 3DGS representation with a post-optimization step. Our method provides visually compelling results, making object transfer between 3DGS easy to use and significantly improving quality compared to previous methods.

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 presents a pipeline for lighting-consistent object transfer across 3D Gaussian Splatting (3DGS) scenes. An object is extracted from a source scene and naively composited into a target scene; inconsistent rendered views are then passed through a diffusion model trained on heterogeneous pairs of inconsistent composites and consistent outputs (synthetic, generated, and real data). The harmonized 2D views are consolidated into a new 3DGS representation via post-optimization. The central claim is that this yields visually compelling, lighting-consistent results that are easy to use and significantly outperform prior methods.

Significance. If the pipeline produces coherent 3D results without new artifacts, the work would supply a practical end-to-end tool for object compositing in radiance fields, directly addressing a common pain point in VFX, interior design, and marketing. The combination of per-view diffusion harmonization with 3DGS post-optimization is a reasonable engineering approach to the multi-view consistency problem. Credit is due for the heterogeneous training data strategy and the explicit 3D consolidation step, both of which move beyond purely 2D harmonization baselines.

major comments (3)
  1. [Abstract / Results] Abstract and Results: the claim that the method 'significantly improving quality compared to previous methods' is unsupported by any quantitative metrics (PSNR, SSIM, LPIPS, user study), ablation studies, or error analysis. This absence is load-bearing for the central contribution.
  2. [Method] Method (diffusion harmonization step): the model is applied independently to each rendered view with no explicit multi-view consistency mechanism (shared latent, geometric conditioning, or cross-view loss). The subsequent 3DGS post-optimization optimizes per-Gaussian appearance parameters without solving global illumination, leaving open whether harmonized outputs remain consistent enough to avoid new lighting or geometry artifacts when source/target lighting differs strongly.
  3. [Experiments / Evaluation] Evaluation: no analysis is provided of failure cases when rendered composites contain 3DGS-specific artifacts absent from the training pairs, nor of how well the post-optimization recovers coherence under large lighting mismatches.
minor comments (2)
  1. [Training Data] The proportions of synthetic, generated, and real data in the training set are not quantified; a table or paragraph stating the exact mix would improve reproducibility.
  2. [Method] Notation for the diffusion model input/output (inconsistent composite vs. harmonized image) should be introduced once and used consistently rather than described only in prose.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's report. We address each of the major comments below and outline the revisions we plan to make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results: the claim that the method 'significantly improving quality compared to previous methods' is unsupported by any quantitative metrics (PSNR, SSIM, LPIPS, user study), ablation studies, or error analysis. This absence is load-bearing for the central contribution.

    Authors: We recognize that the abstract and results section emphasize qualitative improvements without supporting quantitative evidence. To address this, we will revise the manuscript to include quantitative metrics such as PSNR, SSIM, and LPIPS computed on a test set of composite scenes, as well as results from a user study comparing our method to baselines. We will also incorporate ablation studies on the diffusion model training and the post-optimization step. These additions will provide the necessary support for the central claims. revision: yes

  2. Referee: [Method] Method (diffusion harmonization step): the model is applied independently to each rendered view with no explicit multi-view consistency mechanism (shared latent, geometric conditioning, or cross-view loss). The subsequent 3DGS post-optimization optimizes per-Gaussian appearance parameters without solving global illumination, leaving open whether harmonized outputs remain consistent enough to avoid new lighting or geometry artifacts when source/target lighting differs strongly.

    Authors: The harmonization is performed independently per view, as the diffusion model operates on 2D images. Multi-view consistency is achieved through the subsequent 3DGS post-optimization, which jointly optimizes the appearance parameters of the inserted object's Gaussians across all views. We will expand the method section to better explain this reliance on post-optimization and discuss its limitations regarding global illumination. While we did not implement explicit cross-view mechanisms in the diffusion stage to preserve the method's simplicity, we agree that additional analysis of consistency under strong lighting differences is warranted and will include relevant experiments or visualizations in the revision. revision: partial

  3. Referee: [Experiments / Evaluation] Evaluation: no analysis is provided of failure cases when rendered composites contain 3DGS-specific artifacts absent from the training pairs, nor of how well the post-optimization recovers coherence under large lighting mismatches.

    Authors: We acknowledge the lack of explicit failure case analysis in the current evaluation. In the revised manuscript, we will add a section on limitations and failure modes, including cases involving 3DGS-specific artifacts not present in the training data and scenarios with large lighting mismatches. This will include qualitative examples and discussion of when the post-optimization successfully recovers coherence and when it does not. revision: yes

Circularity Check

0 steps flagged

No circularity: method relies on external training data and post-processing without self-referential reductions

full rationale

The paper describes a pipeline that trains a diffusion model on an external heterogeneous dataset of inconsistent composite inputs and consistent outputs (synthetic, generated, and real data), renders inconsistent views from 3DGS composites, applies the model per-view, and consolidates via post-optimization. No equations, fitted parameters, or self-citations are presented that reduce any prediction or result to the inputs by construction. The central claim depends on generalization from training data and the effectiveness of post-optimization, which are independent of the paper's own outputs. This is a standard empirical ML approach with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that a diffusion model trained on mixed synthetic/generated/real inconsistent-to-consistent image pairs will generalize to 3DGS-rendered composites; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption A diffusion model trained on heterogeneous inconsistent-to-consistent image pairs will generalize to correct lighting in 3DGS object transfer renders
    The method depends on this generalization holding for the post-optimization step to succeed.

pith-pipeline@v0.9.1-grok · 5745 in / 1263 out tokens · 28142 ms · 2026-06-26T09:44:48.735014+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

300 extracted references · 91 canonical work pages · 16 internal anchors

  1. [1]

    arXiv preprint arXiv:2512.18314 , year=

    MatSpray: Fusing 2D Material World Knowledge on 3D Geometry , author=. arXiv preprint arXiv:2512.18314 , year=

    arXiv

  2. [2]

    2025 , month =

    Raisinghani, Naina , title =. 2025 , month =

    2025

  3. [3]

    arXiv preprint arXiv:2512.13157 , year=

    Intrinsic Image Fusion for Multi-View 3D Material Reconstruction , author=. arXiv preprint arXiv:2512.13157 , year=

    arXiv

  4. [4]

    Eurographics Symposium on Rendering , year =

    NeRF-Tex: Neural Reflectance Field Textures , author =. Eurographics Symposium on Rendering , year =

  5. [5]

    ACM Transactions on Graphics (Proc

    Kuznetsov, Alexandr and Mullia, Krishna and Xu, Zexiang and Hašan, Miloš and Ramamoorthi, Ravi , title =. ACM Transactions on Graphics (Proc. SIGGRAPH 2021) , volume =

    2021

  6. [6]

    Generative Adversarial Nets , url =

    Goodfellow, Ian and Pouget-Abadie, Jean and Mirza, Mehdi and Xu, Bing and Warde-Farley, David and Ozair, Sherjil and Courville, Aaron and Bengio, Yoshua , booktitle =. Generative Adversarial Nets , url =

  7. [7]

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

    A style-based generator architecture for generative adversarial networks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  8. [8]

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

    Analyzing and improving the image quality of stylegan , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  9. [9]

    arXiv preprint arXiv:1506.03365 , year=

    Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop , author=. arXiv preprint arXiv:1506.03365 , year=

    Pith/arXiv arXiv

  10. [10]

    Computer Graphics Forum , volume=

    State-of-the-Art in the Architecture, Methods and Applications of StyleGAN , author=. Computer Graphics Forum , volume=. 2022 , organization=

    2022

  11. [11]

    Proceedings of the IEEE international conference on computer vision , pages=

    Mask r-cnn , author=. Proceedings of the IEEE international conference on computer vision , pages=

  12. [12]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2017 , publisher=

    2017

  13. [13]

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

    Exploring intermediate representation for monocular vehicle pose estimation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  14. [14]

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

    Efficient geometry-aware 3D generative adversarial networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  15. [15]

    European conference on computer vision , pages=

    Nerf: Representing scenes as neural radiance fields for view synthesis , author=. European conference on computer vision , pages=. 2020 , organization=

    2020

  16. [16]

    Scientific Visualization: Advanced Concepts , title =

    Max, Nelson and Chen, Min , year =. Scientific Visualization: Advanced Concepts , title =

  17. [17]

    Proceedings of the IEEE international conference on computer vision , pages=

    Arbitrary style transfer in real-time with adaptive instance normalization , author=. Proceedings of the IEEE international conference on computer vision , pages=

  18. [18]

    ImageNet: A large-scale hierarchical image database , year=

    Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Kai Li and Li Fei-Fei , booktitle=. ImageNet: A large-scale hierarchical image database , year=

  19. [19]

    arXiv preprint arXiv:2202.00273 , year=

    Stylegan-xl: Scaling stylegan to large diverse datasets , author=. arXiv preprint arXiv:2202.00273 , year=

    arXiv

  20. [20]

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

    A Structured Dictionary Perspective on Implicit Neural Representations , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  21. [21]

    Owen , year = 2013, title =

    Art B. Owen , year = 2013, title =

    2013

  22. [22]

    International conference on machine learning , pages=

    Which training methods for GANs do actually converge? , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  23. [23]

    arXiv preprint arXiv:1710.10196 , year=

    Progressive growing of gans for improved quality, stability, and variation , author=. arXiv preprint arXiv:1710.10196 , year=

    Pith/arXiv arXiv

  24. [24]

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

    pi-gan: Periodic implicit generative adversarial networks for 3d-aware image synthesis , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  25. [25]

    Advances in Neural Information Processing Systems , volume=

    Implicit neural representations with periodic activation functions , author=. Advances in Neural Information Processing Systems , volume=

  26. [26]

    arXiv preprint arXiv:2110.08985 , year=

    Stylenerf: A style-based 3d-aware generator for high-resolution image synthesis , author=. arXiv preprint arXiv:2110.08985 , year=

    arXiv

  27. [27]

    Advances in Neural Information Processing Systems , volume=

    Graf: Generative radiance fields for 3d-aware image synthesis , author=. Advances in Neural Information Processing Systems , volume=

  28. [28]

    arXiv preprint arXiv:2206.10535 , year=

    EpiGRAF: Rethinking training of 3D GANs , author=. arXiv preprint arXiv:2206.10535 , year=

    arXiv

  29. [29]

    Advances in neural information processing systems , volume=

    Improved training of wasserstein gans , author=. Advances in neural information processing systems , volume=

  30. [30]

    International conference on machine learning , pages=

    Wasserstein generative adversarial networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  31. [31]

    International Conference on Machine Learning , pages=

    On the spectral bias of neural networks , author=. International Conference on Machine Learning , pages=. 2019 , organization=

    2019

  32. [32]

    Advances in Neural Information Processing Systems , volume=

    Ganspace: Discovering interpretable gan controls , author=. Advances in Neural Information Processing Systems , volume=

  33. [33]

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

    Image2stylegan: How to embed images into the stylegan latent space? , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  34. [34]

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

    Image2stylegan++: How to edit the embedded images? , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  35. [35]

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

    Editing in style: Uncovering the local semantics of gans , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  36. [36]

    arXiv preprint arXiv:2107.07437 , year=

    Stylefusion: A generative model for disentangling spatial segments , author=. arXiv preprint arXiv:2107.07437 , year=

    arXiv

  37. [37]

    International Conference on Machine Learning , pages=

    Learning transferable visual models from natural language supervision , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  38. [38]

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

    Labels4free: Unsupervised segmentation using stylegan , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  39. [39]

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

    Styleclip: Text-driven manipulation of stylegan imagery , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  40. [40]

    ACM Transactions on Graphics (TOG) , volume=

    Stylegan-nada: Clip-guided domain adaptation of image generators , author=. ACM Transactions on Graphics (TOG) , volume=. 2022 , publisher=

    2022

  41. [41]

    Advances in Neural Information Processing Systems , volume=

    Training generative adversarial networks with limited data , author=. Advances in Neural Information Processing Systems , volume=

  42. [42]

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

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  43. [43]

    arXiv preprint arXiv:2010.07492 , year=

    Nerf++: Analyzing and improving neural radiance fields , author=. arXiv preprint arXiv:2010.07492 , year=

    arXiv 2010

  44. [44]

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

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  45. [45]

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

    Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  46. [46]

    arXiv preprint arXiv:2112.03907 , year=

    Ref-nerf: Structured view-dependent appearance for neural radiance fields , author=. arXiv preprint arXiv:2112.03907 , year=

    arXiv

  47. [47]

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

    Stargan v2: Diverse image synthesis for multiple domains , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  48. [48]

    The International Journal of Robotics Research , volume=

    Vision meets robotics: The kitti dataset , author=. The International Journal of Robotics Research , volume=. 2013 , publisher=

    2013

  49. [49]

    Beyond PASCAL: A benchmark for 3D object detection in the wild , year=

    Xiang, Yu and Mottaghi, Roozbeh and Savarese, Silvio , booktitle=. Beyond PASCAL: A benchmark for 3D object detection in the wild , year=

  50. [50]

    2016 , publisher=

    Physically based rendering: From theory to implementation , author=. 2016 , publisher=

    2016

  51. [51]

    1980 , issue_date =

    Whitted, Turner , title =. 1980 , issue_date =. doi:10.1145/358876.358882 , journal =

    work page doi:10.1145/358876.358882 1980

  52. [52]

    Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume=

    Image-based rendering of cars using semantic labels and approximate reflection flow , author=. Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume=

  53. [53]

    Advances in neural information processing systems , volume=

    Gans trained by a two time-scale update rule converge to a local nash equilibrium , author=. Advances in neural information processing systems , volume=

  54. [54]

    2003 , publisher=

    Multiple view geometry in computer vision , author=. 2003 , publisher=

    2003

  55. [55]

    2001 , isbn =

    Buehler, Chris and Bosse, Michael and McMillan, Leonard and Gortler, Steven and Cohen, Michael , title =. 2001 , isbn =. doi:10.1145/383259.383309 , booktitle =

    work page doi:10.1145/383259.383309 2001

  56. [56]

    , title =

    Kajiya, James T. , title =. SIGGRAPH Comput. Graph. , month =. 1986 , issue_date =. doi:10.1145/15886.15902 , abstract =

    work page doi:10.1145/15886.15902 1986

  57. [57]

    Eurographics workshop on Rendering techniques , pages=

    Global illumination using photon maps , author=. Eurographics workshop on Rendering techniques , pages=. 1996 , organization=

    1996

  58. [58]

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

    Nerd: Neural reflectance decomposition from image collections , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  59. [59]

    ACM Transactions on Graphics (TOG) , volume=

    Nerfactor: Neural factorization of shape and reflectance under an unknown illumination , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

    2021

  60. [60]

    arXiv preprint arXiv:2201.05989 , year=

    Instant neural graphics primitives with a multiresolution hash encoding , author=. arXiv preprint arXiv:2201.05989 , year=

    arXiv

  61. [61]

    ACM Transactions on Graphics (TOG) , volume=

    Pivotal tuning for latent-based editing of real images , author=. ACM Transactions on Graphics (TOG) , volume=. 2022 , publisher=

    2022

  62. [62]

    Advances in Neural Information Processing Systems , volume=

    Neural-pil: Neural pre-integrated lighting for reflectance decomposition , author=. Advances in Neural Information Processing Systems , volume=

  63. [63]

    arXiv preprint arXiv:2205.15768 , year=

    SAMURAI: Shape And Material from Unconstrained Real-world Arbitrary Image collections , author=. arXiv preprint arXiv:2205.15768 , year=

    arXiv

  64. [64]

    ACM Transactions on Graphics (Proceedings SIGGRAPH) , volume =

    Pandey, Rohit and Orts-Escolano, Sergio and LeGendre, Chloe and Haene, Christian and Bouaziz, Sofien and Rhemann, Christoph and Debevec, Paul and Fanello, Sean , title =. ACM Transactions on Graphics (Proceedings SIGGRAPH) , volume =. 2021 , doi =

    2021

  65. [65]

    arXiv preprint arXiv:2201.04873 , year=

    VoLux-GAN: A Generative Model for 3D Face Synthesis with HDRI Relighting , author=. arXiv preprint arXiv:2201.04873 , year=

    arXiv

  66. [66]

    Advances in Neural Information Processing Systems , volume=

    A shading-guided generative implicit model for shape-accurate 3d-aware image synthesis , author=. Advances in Neural Information Processing Systems , volume=

  67. [67]

    Proceedings of the 27th annual conference on Computer graphics and interactive techniques , pages=

    Acquiring the reflectance field of a human face , author=. Proceedings of the 27th annual conference on Computer graphics and interactive techniques , pages=

  68. [68]

    arXiv preprint arXiv:2209.10510 , year=

    Learning to Relight Portrait Images via a Virtual Light Stage and Synthetic-to-Real Adaptation , author=. arXiv preprint arXiv:2209.10510 , year=

    arXiv

  69. [69]

    and Rusinkiewicz, Szymon and Heide, Felix , title =

    Wang, Zhibo and Yu, Xin and Lu, Ming and Wang, Quan and Qian, Chen and Xu, Feng , title =. 2020 , issue_date =. doi:10.1145/3414685.3417824 , journal =

    work page doi:10.1145/3414685.3417824 2020

  70. [70]

    Proceedings of the IEEE international conference on computer vision , pages=

    Least squares generative adversarial networks , author=. Proceedings of the IEEE international conference on computer vision , pages=

  71. [71]

    ACM Transactions on Graphics (SIGGRAPH Asia) , publisher =

    Thomas Leimk\"uhler and George Drettakis , title =. ACM Transactions on Graphics (SIGGRAPH Asia) , publisher =. 2021 , doi =

    2021

  72. [72]

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

    The unreasonable effectiveness of deep features as a perceptual metric , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  73. [73]

    Deep Relightable Textures: Volumetric Performance Capture with Neural Rendering , year =

    Meka, Abhimitra and Pandey, Rohit and H\". Deep Relightable Textures: Volumetric Performance Capture with Neural Rendering , year =. doi:10.1145/3414685.3417814 , journal =

    work page doi:10.1145/3414685.3417814

  74. [74]

    SIGGRAPH 2012 , year=

    The Light Stages and Their Applications to Photoreal Digital Actors , author=. SIGGRAPH 2012 , year=

    2012

  75. [75]

    2019 , issue_date =

    Guo, Kaiwen and Lincoln, Peter and Davidson, Philip and Busch, Jay and Yu, Xueming and Whalen, Matt and Harvey, Geoff and Orts-Escolano, Sergio and Pandey, Rohit and Dourgarian, Jason and Tang, Danhang and Tkach, Anastasia and Kowdle, Adarsh and Cooper, Emily and Dou, Mingsong and Fanello, Sean and Fyffe, Graham and Rhemann, Christoph and Taylor, Jonathan...

    work page doi:10.1145/3355089.3356571 2019

  76. [76]

    2010 , publisher=

    High dynamic range imaging: acquisition, display, and image-based lighting , author=. 2010 , publisher=

    2010

  77. [77]

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

    Giraffe: Representing scenes as compositional generative neural feature fields , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  78. [78]

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

    GIRAFFE HD: A High-Resolution 3D-aware Generative Model , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  79. [79]

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

    A large-scale car dataset for fine-grained categorization and verification , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  80. [80]

    Tero Karras and Miika Aittala and Samuli Laine and Erik H\"ark\"onen and Janne Hellsten and Jaakko Lehtinen and Timo Aila , title =. Proc. NeurIPS , year =

Showing first 80 references.