Pith. sign in

REVIEW 3 major objections 4 minor 51 references

A video diffusion model can render an animated mesh from tracking, world position, and normals, and its world-position channel beats depth by 1.26–1.55 dB in matched swaps.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-05 04:21 UTC pith:SX4CYP5E

load-bearing objection Useful 4D rendering interface, but the headline depth-vs-world-position ablation is confounded by channel count; otherwise a serious, honest empirical paper. the 3 major comments →

arxiv 2608.00094 v2 pith:SX4CYP5E submitted 2026-07-30 cs.CV

Video Models as Native 4D Renderers: World-Grounded Conditioning from Animated Mesh

Junhao Chen , Mingjin Chen , Henghaofan Zhang , Minglin Chen , Liaoyuan Fan , Boran Zhang , Saining Zhang , Mingze Sun
show 4 more authors
Hao Zhao Ruqi Huang Zhihao Li Yufei Wang
This is my paper
classification cs.CV
keywords 4D generative renderingvideo diffusionworld-position conditioningneural G-buffertracking mapsdepth vs world positionreference-guided synthesisanimated mesh rendering
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to show that a pretrained video diffusion model can act as a controllable 4D renderer: given an animated mesh, a target camera path, and a first-frame reference image, it can generate a video that honors both camera motion and scene-internal animation. To do this it proposes DAR, which rasterizes the mesh into a neural 4D G-buffer—persistent tracking, world position, and normals—and feeds those channels together with Plücker camera rays into a widened control adapter of a video diffusion backbone. The load-bearing experimental claim is that the world-position channel beats ordinary depth in the geometric slot: in a matched swap where only world position is replaced by depth, PSNR drops by 1.26–1.55 dB at every training checkpoint, and the full DAR condition outperforms off-the-shelf depth-conditioned rendering by 1.54 dB PSNR on the 68-case benchmark. If correct, the payoff is a practical way to re-render animated 3D scenes with a generative video model while keeping geometry and camera under explicit user control.

Core claim

DAR's central claim is that tracking plus world position, not depth, is the right visible 4D state code for reference-guided rendering of animated meshes. Tracking identifies which persistent surface element should receive appearance; world position states where that element currently is in scene coordinates; normals provide local shape; and Plücker rays separately encode the observer. Depth is a camera-dependent chart: it is non-injective over 3D points and changes when the same point is viewed from a translated camera, so it mixes camera and object motion in one number. The paper supports this with a controlled depth-swap ablation: keeping tracking and normals fixed and replacing only worl

What carries the argument

The carrying object is the neural 4D G-buffer, an image-format tensor computed from the animated mesh that stacks three aligned maps per pixel: tracking color q(s) marking persistent surface or part identity, scene-normalized world position Xbar_t(s), and normal nbar_t(s). The control tensor concatenates the 24 channels of temporally packed Plücker rays with these 9 geometry channels and is injected through a widened SimpleAdapter into the pretrained Wan2.2 transformer at the same early point as the original camera control. The argument that this is the right code is geometric: the pair (q, Xbar) determines both which surface element is visible and where it is in the 3D scene, independently

Load-bearing premise

The claim assumes the tracking color assigned to each surface point stays a unique, stable label across the whole animation and across all camera views; if tracking labels break under occlusion or large orbit, the identity component becomes noise and the advantage over depth could disappear.

What would settle it

Take a dynamic mesh benchmark and run the matched depth-swap ablation with tracking labels deliberately corrupted per frame, such as recoloring tracked surface IDs after occlusion; if a corrupted-tracking DAR no longer beats the depth-swap variant, the identity channel rather than world position is carrying the gain. A second check: on a public mesh-video corpus with ground-truth geometry, reproduce the positive 1.26–1.55 dB depth-swap gap at every checkpoint; a reversal would refute the claim.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Anything that already has an animated mesh, a camera path, and one reference image can be re-rendered with changed appearance by a video diffusion model, keeping pose and silhouette aligned.
  • The control signal is the representation: adding a camera-independent world-position channel beats adding depth even when tracking and normals are already present.
  • The method inherits the pretrained image-to-video prior, so it does not need per-scene optimization or an explicit PBR material authoring step for appearance transfer.
  • If the LoRA bottleneck is removed (full fine-tune), the same condition reaches higher fidelity (25.36 PSNR), suggesting the condition does not saturate with capacity.
  • Depth-based and camera-only baselines fail on different axes of the renderer interface, so existing video-backbone controls are not interchangeable with a joint camera-plus-geometry interface.

Where Pith is reading between the lines

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

  • By extension, the same ranking should hold for other camera-dependent geometric charts such as disparity, inverse depth, or ray-distance maps: the paper's identifiability argument predicts any camera chart will lose to scene coordinates when camera and object motion are both varied.
  • If tracking labels could be recovered from video rather than supplied by a mesh, the conditioning might transfer to real scenes; the paper's own limitation note about noisy geometry and confidence masks suggests this transfer will need label-noise handling, and the depth advantage may shrink.
  • A direct test the paper does not run is the motion-separation stress test proposed in its appendix: the same object animation with several cameras and the same camera with several object animations; the predicted failure mode is that depth-only control degrades on recombined motion while world-position conditioning stays stable.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes DAR, a reference-guided 4D renderer built on Wan2.2 video diffusion. It extends the camera-only Plücker-ray control by projecting an animated mesh into a neural 4D G-buffer (tracking, world position, normal) and injecting these 9 geometry channels through a widened adapter alongside the 24 camera channels. The central claim is that tracking+world position is a better visible 4D state code than camera-dependent depth for dual camera/object control. Evidence includes a 68-case synthetic benchmark, LoRA ablations, a full fine-tune, and an OOD probe. The paper reports +1.54 dB PSNR over off-the-shelf Wan2.2-Depth and +1.42 dB over a 'matched' depth-swap ablation at checkpoint 10k, with the advantage holding at every saved checkpoint.

Significance. If the central claim held, the paper would make a useful contribution: it identifies a representation-level design choice for controllable video generation and provides a dataset and controlled training protocol for studying mesh-conditioned rendering. Strengths include the uniform backbone/data/compute for all LoRA ablations, the checkpoint sweep, the deliberate separation of controlled ablations from diagnostic external baselines, the OOD generalization probe, and the honest statement of limitations (including a future user-study protocol and no synthetic preference numbers). The theoretical motivation is clear and the dataset infrastructure is valuable. However, the key empirical comparison that is supposed to isolate world-position versus depth is confounded by channel count, so the paper's central quantitative claim is not yet established.

major comments (3)
  1. [§6.1, Table 2; Appendix A.3] The matched depth-swap ablation is not channel-matched. DAR uses tracking+world position+normal (9 geometry channels; 24+9=33 input channels), while the depth-swap row uses tracking+normal+depth (7 geometry channels; 24+7=31). The +1.42 dB PSNR / +0.034 SSIM at ck10k, and the +1.26–+1.55 dB sweep in Table 6, are therefore not attributable solely to replacing depth by world position; they may reflect the two additional input channels changing adapter capacity or providing extra correlated signal. Since Appendix A.1 (Eq. 12) concedes that depth plus calibrated rays recovers world position exactly, the empirical test of the inductive-bias claim must be unconfounded. Please add a channel-matched depth variant (e.g., depth replicated or expanded to 3 channels, or depth plus two derived maps) at 24+9=33 channels and report the same checkpoint sweep.
  2. [Abstract; §6.1, Table 2] The headline improvement of +1.54 dB over Wan2.2-Depth is an off-the-shelf comparison: Wan2.2-Depth is not fine-tuned on DAR-4D, while DAR is trained on the benchmark. This conflates representation choice with training. The controlled comparison is the depth-swap ablation, which is confounded as noted above. Please either fine-tune a depth-conditioned Wan2.2 variant under the same LoRA recipe (matched in data, compute, and channel count) or report the off-the-shelf number only as an untrained reference point and remove it from the abstract's central claim.
  3. [§3.3, §3.4, §7; Appendix A.4] The theoretical identifiability argument assumes q(s) is a unique, persistent label across the whole animation and all camera views. The paper does not provide a quantitative analysis of tracking-label stability; its own conclusion notes that videos beyond 200 frames can flicker under large camera changes and long occlusions, and the benchmark includes clips up to 249 frames. Because tracking identity is half of the proposed 4D code, the authors should report a tracking-stability measure (e.g., label consistency across views/occlusion, per-clip endpoint metrics for the longest clips) and state how the benchmark's longest clips behave. This also bears on the OOD-34 result, where the full 9-channel DAR underperforms depth-only conditioning.
minor comments (4)
  1. [§5.3] Typo: 'all 684d_vis cases' should be 'all 68 4d_vis cases'.
  2. [§6.1, Table 2] The heading 'Single-channel' is misleading because depth is 1 channel while tracking, normal, and world position are 3 channels each. Consider renaming to 'Single-family' or make the channel counts more prominent, since the later comparison of single-channel rows with DAR is not channel-controlled.
  3. [Appendix A.5] The user-study protocol is described as future work rather than conducted. This should be stated in the main text limitations, not only in an appendix; a renderer paper should make clear that no human evaluation is included.
  4. [General] No code or dataset release link is indicated. Releasing the DAR-4D splits, projection renderer, and training scripts would substantially strengthen reproducibility, especially since the benchmark is not publicly observable.

Circularity Check

0 steps flagged

No circularity: the depth-vs-world-position claim rests on a controlled empirical ablation, not on definitions; the main concern is a channel-count confound, which is a correctness risk rather than a circular reduction.

full rationale

The paper's formal argument in Sec. 3.4/Prop. 1 is definitional: Definition 1 defines a visible 4D state code as the pair (tracking, world position), and Prop. 1 restates that depth is camera-dependent while world position is camera-independent. This is a tautological framing of what the authors choose to call a 4D state code, but the central quantitative claim is not derived from that definition. The reported PSNR/SSIM/LPIPS differences come from a training ablation (Table 2 and Table 6), and the paper explicitly disclaims that the theory forces the numbers (Sec. 6.1: 'The theory does not require world position alone to dominate every low-capacity LoRA row'). Appendix A.1 honestly concedes (Eq. 12) that depth plus calibrated Plücker rays can recover world position exactly, so the paper's own argument is about inductive bias, not information loss. No fitted parameter is renamed as a prediction, and no uniqueness theorem from the authors' prior work is invoked to forbid alternatives. The dataset is built on the authors' OVOW pipeline [Chen et al. 2026g], but that self-citation is not load-bearing for the central condition-format claim; the external Wan2.2-Depth baseline and DAS/VideoFrom3D diagnostics provide independent comparisons. The most substantive issue is that the 'matched depth-swap' ablation is not channel-matched: Table A.3 lists DAR as 24+9=33 channels and the depth-swap variant as 24+7=31, so the +1.42 dB advantage could partly reflect adapter capacity rather than world-position semantics. That is a methodological confound and a correctness risk, not a circularity, because the empirical result is not equivalent to its inputs by construction. Appendix A.5 also withholds synthetic user-study numbers, further indicating that the authors are not passing fitted or invented evidence off as a prediction. Overall, no circular derivation chain is present.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

No new physical entities are postulated. The neural 4D G-buffer is an image-format representation, not an entity with independent falsifiable consequences. The central unstated burden is the stability of tracking labels and the unspecified world-position normalization; both are needed for the condition to work as claimed.

free parameters (5)
  • World-position normalization scheme = Not specified
    DAR uses bar-X as scene-normalized world position; the exact normalization bounds are not given, so the central condition cannot be reproduced exactly and may vary across scenes.
  • LoRA rank = 256
    Chosen by hand; controls adapter capacity in all LoRA runs; the full-fine-tune variant removes this bottleneck.
  • Guidance scale and denoising steps = 3.0, 50
    Inference hyperparameters; likely affect PSNR and LPIPS, though they are fixed across compared variants.
  • Training steps and checkpoints = 10,000 steps, checkpoint every 2,000
    The main ranking is at checkpoint 10k; the checkpoint sweep in Table 6 supports stability.
  • Geometry channel count N = 9 (3 tracking, 3 world position, 3 normal)
    The central design choice; no search over N is reported.
axioms (5)
  • domain assumption Tracking color q(s) provides a persistent and unique label for each surface element at the selected granularity throughout the animation.
    Sec. 3.3 and 3.4; if tracking labels split, merge, or change under occlusion, appearance transport via q fails.
  • domain assumption Scene normalization of world position is injective inside the scene bounds and is camera-independent.
    Appendix A.1 relies on this for the identifiability argument; the exact normalization is not specified.
  • domain assumption Rasterized G-buffer maps align with the video latent temporal grid after VAE resampling.
    Eq. (5) in Sec. 3.2 requires P_vae to correctly align per-pixel mesh conditions with the diffusion latent grid.
  • domain assumption The pretrained Wan2.2-Fun-5B-Control-Camera backbone can be fine-tuned to accept the widened adapter without destroying the image-to-video prior.
    The entire method depends on this transfer; the paper provides only the benchmark results as evidence.
  • domain assumption The first-frame reference plus binary mask is a sufficient appearance anchor.
    Sec. 3.5 reuses the Wan2.2 inpainting interface without ablating this choice.

pith-pipeline@v1.3.0-alltime-deepseek · 18222 in / 16007 out tokens · 166014 ms · 2026-08-05T04:21:23.089544+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Video Models as Native 4D Renderers: World-Grounded Conditioning from Animated Mesh." pith.science (2026). https://pith.science/paper/SX4CYP5E

@misc{pith2026260800094,
  author       = {Pith},
  title        = {Pith review of: Video Models as Native 4D Renderers: World-Grounded Conditioning from Animated Mesh},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SX4CYP5E}},
  note         = {Machine review of arXiv:2608.00094}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Pretrained video diffusion models can act as renderers when the desired scene state is already specified by an animated mesh, a camera trajectory, and a reference image. This 4D generative rendering setting raises a representation question: what image-format condition lets a video backbone obey both camera motion and scene-internal animation? We propose DAR, a reference-guided renderer that extends Wan2.2 camera control from Pl\"ucker rays alone to a joint camera-plus-geometry interface. DAR projects a neural 4D G-buffer (tracking, world position, and normal) from the animated mesh and injects it through a widened control adapter while preserving the pretrained image-to-video prior. The central design choice is the pair of tracking and world position. Tracking identifies the persistent surface element that should carry appearance; world position gives its current scene-coordinate state; normal supplies local shape. Depth plus calibrated rays can recover 3D in principle, but depth is a camera-dependent chart in which camera and object motion are mixed. On the 68-case DAR-4D benchmark, LoRA DAR reaches PSNR 23.22, SSIM 0.895, and LPIPS 0.134, improving over off-the-shelf Wan2.2-Depth by 1.54 dB PSNR; a full fine-tune reaches PSNR 25.36 and SSIM 0.917. Matched ablations show that replacing world position by depth reduces PSNR by 1.26--1.55 dB at every checkpoint, supporting tracking+world-position correspondence as a practical 4D rendering condition.

Figures

Figures reproduced from arXiv: 2608.00094 by Boran Zhang, Hao Zhao, Henghaofan Zhang, Junhao Chen, Liaoyuan Fan, Mingjin Chen, Minglin Chen, Mingze Sun, Ruqi Huang, Saining Zhang, Yufei Wang, Zhihao Li.

Figure 1
Figure 1. Figure 1: DAR is a reference-guided 4D renderer: with the animated mesh and camera fixed, changing only the first-frame reference image re-renders the scene’s appearance. The same animated car mesh, camera trajectory, and 4D geometry are held fixed; each column uses a different reference style and each row is a normalized animation time. DAR alters paint style, color, and vehicle appearance while keeping the camera,… view at source ↗
Figure 2
Figure 2. Figure 2: DAR pipeline. An animated 4D mesh, target camera trajectory, reference image, and text prompt are converted into aligned video-diffusion conditions. Plücker rays encode the observer; a mesh projection renderer produces tracking, world-position, and normal maps; a widened SimpleAdapter injects the combined camera-plus-geometry control into the pretrained Wan2.2 video DiT. 3.3 Neural 4D G-buffer For each vis… view at source ↗
Figure 3
Figure 3. Figure 3: 4D rotation gallery. DAR outputs under a horizontal orbit (±60◦ ) around six Blender meshes. Each row shows one scene; columns show the first-frame reference and DAR outputs at 𝑡=0, 𝑇 3 , 2𝑇 3 , 𝑇 −1. DAR follows the target camera trajectory while preserving the animated mesh state and reference appearance across humans, animals, vehicles, mechanical objects, and indoor scenes. 7 Conclusion We introduced 4… view at source ↗
Figure 4
Figure 4. Figure 4: DAR controls both camera trajectory and in-scene 4D mesh animation for generative rendering. Each row shows one OVOW Blender asset under a ±60◦ horizontal orbit. Except for the reference column, all panels show the final frame 𝑡=𝑇 , where the target view is farthest from the first-frame reference. Columns show reference, mesh-derived 9-channel geometry, Wan2.2-Camera, Wan2.2-Depth, Blender GT, and DAR. Wan… view at source ↗
Figure 5
Figure 5. Figure 5: Executed baseline suite across animation time. Three OVOW assets are shown at five normalized time points (𝑡=0/25%/50%/75%/𝑇 ) across Reference, Blender GT, Wan2.2-Camera, Wan2.2-Depth, DAS, VideoFrom3D, and DAR. All videos are sampled by normalized animation time. DAR more consistently preserves both target camera trajectory and animated mesh state. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

Review history (2 revisions) →

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

51 extracted references · 36 canonical work pages · 3 internal anchors

  1. [1]

    Greff, Klaus and Belletti, Francois and Beyer, Lucas and Doersch, Carl and Du, Yilun and Duckworth, Daniel and Fleet, David J and Gnanapragasam, Dan and Golemo, Florian and Herrmann, Charles and Kipf, Thomas and Kundu, Abhijit and others , booktitle=

    work page

  2. [2]

    Xu, Zhongcong and Zhang, Jianfeng and Liew, Jun Hao and Yan, Hanshu and Liu, Jia-Wei and Zhang, Chenxu and Feng, Jiashi and Shou, Mike Zheng , booktitle=

    work page

  3. [3]

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

    Li, Yang and Takehara, Hikari and Taketomi, Takafumi and Zheng, Bo and Nie. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  4. [4]

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

    ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  5. [5]

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

    Texture Generation on 3D Meshes with Point-UV Diffusion , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  6. [6]

    arXiv preprint arXiv:2503.20314 , year=

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

    Pith/arXiv arXiv

  7. [7]

    arXiv preprint arXiv:2604.02329 , year=

    Generative World Renderer , author=. arXiv preprint arXiv:2604.02329 , year=

    arXiv

  8. [8]

    Engine-Native Editable 3D World Reconstruction with Objects and Lighting

    Engine-Native Editable 3D World Reconstruction with Objects and Lighting , author=. arXiv preprint arXiv:2607.20889 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

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

    Generative Rendering: Controllable 4D-Guided Video Generation with 2D Diffusion Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  10. [10]

    One Video, One World: Turning Monocular Video into Physical 4D Scenes

    One Video, One World: Turning Monocular Video into Physical 4D Scenes , author=. arXiv preprint arXiv:2606.31388 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    The Fourteenth International Conference on Learning Representations (ICLR) , year=

    DanceTogether: Generating Interactive Multi-Person Video without Identity Drifting , author=. The Fourteenth International Conference on Learning Representations (ICLR) , year=

    work page

  12. [12]

    arXiv preprint arXiv:2503.15147 , year=

    PBR-Inspired Controllable Diffusion for Image Generation , author=. arXiv preprint arXiv:2503.15147 , year=

    work page arXiv

  13. [13]

    arXiv preprint arXiv:2509.07978 , year=

    One View, Many Worlds: Single-Image to 3D Object Meets Generative Domain Randomization for One-Shot 6D Pose Estimation , author=. arXiv preprint arXiv:2509.07978 , year=

    Pith/arXiv arXiv

  14. [14]

    Zeng, Xianfang and Chen, Xin and Qi, Zhongqi and Liu, Wen and Zhao, Zibo and Wang, Zhibin and Fu, Bin and Liu, Yong and Yu, Gang , booktitle=

    work page

  15. [15]

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

    Animator-Centric Skeleton Generation on Objects with Fine-Grained Details , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  16. [16]

    IEEE transactions on image processing , volume=

    Image quality assessment: from error visibility to structural similarity , author=. IEEE transactions on image processing , volume=

  17. [17]

    arXiv preprint arXiv:2605.27351 , year=

    Feedforward 3D Editing Learns from Semantic-Part Transformation , author=. arXiv preprint arXiv:2605.27351 , year=

    Pith/arXiv arXiv

  18. [18]

    Deitke, Matt and Liu, Ruoshi and Wallingford, Matthew and Ngo, Huong and Michel, Oscar and Kusupati, Aditya and Fan, Alan and Laforte, Christian and Voleti, Vikram and Gadre, Samir Yitzhak and others , booktitle=

    work page

  19. [19]

    ACM Transactions on Graphics (TOG) , volume=

    Functional maps: a flexible representation of maps between shapes , author=. ACM Transactions on Graphics (TOG) , volume=

    work page

  20. [20]

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

    HVG-3D: Bridging Real and Simulation Domains for 3D-Conditional Hand-Object Interaction Video Synthesis , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  21. [21]

    PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation

    PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation , author=. arXiv preprint arXiv:2607.01883 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    arXiv preprint arXiv:2506.01103 , year=

    DeepVerse: 4D Autoregressive Video Generation as a World Model , author=. arXiv preprint arXiv:2506.01103 , year=

    Pith/arXiv arXiv

  23. [23]

    arXiv preprint arXiv:2604.09415 , year=

    Physinone: Visual physics learning and reasoning in one suite , author=. arXiv preprint arXiv:2604.09415 , year=

    Pith/arXiv arXiv

  24. [24]

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

    GenesisTex: Adapting Image Denoising Diffusion to Texture Space , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  25. [25]

    SIGGRAPH Asia 2025 Conference Papers , year=

    VideoFrom3D: 3D Scene Video Generation via Complementary Image and Video Diffusion Models , author=. SIGGRAPH Asia 2025 Conference Papers , year=

    work page 2025

  26. [26]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  27. [27]

    arXiv preprint arXiv:2503.09733 , year=

    I2V3D: Controllable image-to-video generation with 3D guidance , author=. arXiv preprint arXiv:2503.09733 , year=

    Pith/arXiv arXiv

  28. [28]

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

    SceneTex: High-Quality Texture Synthesis for Indoor Scenes via Diffusion Priors , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  29. [29]

    arXiv preprint arXiv:2601.00393 , year=

    NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videos , author=. arXiv preprint arXiv:2601.00393 , year=

    arXiv

  30. [30]

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

    CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  31. [31]

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

    Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  32. [32]

    Weng, Fangsheng and Chen, Junhao and Li, Xiang and Qin, Jie and Guo, Hanzhong and Shao, Chunhao and Han, Xiaoguang , booktitle=

    work page

  33. [33]

    arXiv preprint arXiv:2602.01661 , year=

    From Frames to Sequences: Temporally Consistent Human-Centric Dense Prediction , author=. arXiv preprint arXiv:2602.01661 , year=

    arXiv

  34. [34]

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

    Recammaster: Camera-controlled generative rendering from a single video , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  35. [35]

    Wen, Kairun and Huang, Yuzhi and Chen, Runyu and Zheng, Hui and Lin, Yunlong and Pan, Panwang and Li, Chenxin and Cong, Wenyan and Zhang, Jian and Lu, Junbin and others , journal=

    work page

  36. [36]

    arXiv preprint arXiv:2412.15050 , year=

    Uni-Renderer: Unifying Rendering and Inverse Rendering Via Dual Stream Diffusion , author=. arXiv preprint arXiv:2412.15050 , year=

    Pith/arXiv arXiv

  37. [37]

    arXiv preprint arXiv:2503.14492 , year=

    Cosmos-transfer1: Conditional world generation with adaptive multimodal control , author=. arXiv preprint arXiv:2503.14492 , year=

    Pith/arXiv arXiv

  38. [38]

    Ling, Lu and Sheng, Yichen and Tu, Zhi and Zhao, Wentian and Xin, Cheng and Wan, Kun and Yu, Lantao and Guo, Qianyu and Yu, Zixun and Lu, Yawen and others , booktitle=

    work page

  39. [39]

    arXiv preprint arXiv:2501.18590 , year=

    DiffusionRenderer: Neural Inverse and Forward Rendering with Video Diffusion Models , author=. arXiv preprint arXiv:2501.18590 , year=

    Pith/arXiv arXiv

  40. [40]

    Machine Vision and Applications , volume=

    Ultraman: ultra-fast and high-resolution texture generation for 3D human reconstruction from a single image , author=. Machine Vision and Applications , volume=

    work page

  41. [41]

    Seedance 1.0: Exploring the Boundaries of Video Generation Models , author=

    work page

  42. [42]

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

    DRiVE: Diffusion-based Rigging Empowers Generation of Versatile and Expressive Characters , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  43. [43]

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

    RoomPainter: View-Integrated Diffusion for Consistent Indoor Scene Texturing , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  44. [44]

    Proceedings of the 31st International Conference on Computational Linguistics (COLING) , pages=

    Idea23d: Collaborative lmm agents enable 3d model generation from interleaved multimodal inputs , author=. Proceedings of the 31st International Conference on Computational Linguistics (COLING) , pages=

    work page

  45. [45]

    arXiv preprint arXiv:2501.03847 , year=

    Diffusion as Shader: 3D-aware Video Diffusion for Versatile Video Generation Control , author=. arXiv preprint arXiv:2501.03847 , year=

    Pith/arXiv arXiv

  46. [46]

    Guler, Riza Alp and Neverova, Natalia and Kokkinos, Iasonas , booktitle=

    work page

  47. [47]

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

    Vace: All-in-one video creation and editing , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  48. [48]

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

    LottieGPT: Tokenizing Vector Animation for Autoregressive Generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    work page

  49. [49]

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

    Text2Tex: Text-driven Texture Synthesis via Diffusion Models , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    work page

  50. [50]

    Karaev, Nikita and Rocco, Ignacio and Graham, Benjamin and Neverova, Natalia and Vedaldi, Andrea and Rupprecht, Christian , booktitle=

  51. [51]

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

    Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    work page