The reviewed record of science sign in
Pith

arxiv: 2607.05373 · v1 · pith:YTTALWDD · submitted 2026-07-06 · cs.CV

PixWorld: Unifying 3D Scene Generation and Reconstruction in Pixel Space

Sensen Gao , Zhaoqing Wang , Qihang Cao , Dongdong Yu , Changhu Wang , Jia-Wang Bian This is my paper

Reviewed by Pith2026-07-07 14:05 UTCglm-5.2pith:YTTALWDDopen to challenge →

classification cs.CV
keywords generationreconstructiondiffusionlatentpixworldspacealignsautoencoder
0
0 comments X

The pith

Pixel-space diffusion unifies 3D scene generation and reconstruction

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

The paper argues that 3D scene generation and reconstruction, long handled by separate paradigms, can be unified in a single model by running diffusion directly in pixel space rather than in a compressed latent space. The central mechanism is a two-stream diffusion transformer that partitions multi-view images into clean views (driving reconstruction) and noisy views (driving generation), both decoded into a pixel-aligned 3D Gaussian representation. Because the diffusion objective is supervised on rendered images via differentiable rendering, the training signal aligns directly with 3D scene fidelity rather than with targets in an intermediate latent encoding. The authors further introduce a geometry perception loss that uses a frozen 3D foundation model as a structural critic, penalizing geometric inconsistencies that 2D photometric losses miss. The paper claims this approach consistently outperforms latent-space generation methods and matches state-of-the-art reconstruction methods, suggesting that removing the VAE/RAE bottleneck is a viable path to higher-quality 3D scene modeling.

Core claim

The key finding is that a single pixel-space diffusion model, supervised through differentiable 3D Gaussian rendering and augmented with a geometry-aware feature loss from a pretrained 3D foundation model, can jointly handle 3D scene reconstruction and generation at quality matching or exceeding specialized methods. The unification works by partitioning multi-view inputs into clean and noisy subsets within one forward pass, so the same model reconstructs observed geometry and generates unobserved content without a latent autoencoder stage.

What carries the argument

Two mechanisms carry the argument. First, the pixel-space flow matching objective is defined on rendered multi-view images rather than latent codes, so gradients flow back through the differentiable 3D Gaussian renderer directly into the 3D representation. Second, the geometry perception loss extracts features from both rendered and ground-truth views using a frozen 3D foundation model (pi3), then minimizes the cosine distance between them, providing cross-view 3D structural supervision that 2D image-level losses cannot supply.

If this is right

  • If pixel-space diffusion scales, the VAE/RAE pretraining stage required by latent-space 3D methods becomes unnecessary, simplifying training pipelines and removing a source of information loss.
  • The clean/noisy view partitioning scheme suggests a continuum between reconstruction and generation, where the same model can interpolate between pure observation and pure synthesis based on how many views are clean.
  • The geometry perception loss pattern, using a frozen 3D foundation model as a differentiable structural critic, could be adopted by other 3D generation pipelines that currently rely only on photometric supervision.
  • If the approach generalizes beyond indoor/real-estate scenes, it could serve as a unified backbone for embodied AI applications requiring both scene understanding and scene completion.

Where Pith is reading between the lines

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

  • The 336x448 training resolution is a significant constraint; whether pixel-space diffusion can match latent-space efficiency at the higher resolutions needed for production-quality textures remains an open question that the paper acknowledges but does not resolve.
  • The geometry perception loss depends on the quality and coverage of the frozen 3D foundation model; if that model has systematic geometric blind spots (e.g., thin structures, reflective surfaces), those errors would propagate into PixWorld's learned geometry.
  • The reliance on pseudo-depth labels from DA3 for depth supervision introduces a dependency on an external model whose accuracy on diverse scenes directly affects PixWorld's geometric grounding.
  • The 15-second inference time at 100 NFE is competitive but the lower output resolution compared to video-diffusion baselines means the speed comparison is not strictly like-for-like; distilled video models may close the gap further.

Load-bearing premise

The approach depends on pixel-space diffusion at 336x448 resolution providing sufficient gradient signal through differentiable rendering to train a 1B-parameter model from scratch; if pixel-space diffusion does not scale to higher resolutions without prohibitive compute, the practical advantage over latent-space methods diminishes.

What would settle it

If pixel-space diffusion at higher resolutions (needed for fine-grained texture) requires disproportionately more compute than latent-space alternatives, or if the geometry perception loss provides no measurable benefit beyond what multi-view photometric supervision already captures, the central claim of pixel-space superiority would weaken.

Figures

Figures reproduced from arXiv: 2607.05373 by Changhu Wang, Dongdong Yu, Jia-Wang Bian, Qihang Cao, Sensen Gao, Zhaoqing Wang.

Figure 1
Figure 1. Figure 1: PixWorld unifies 3D scene reconstruction and generation within a single model. Unlike [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of PixWorld. (a) PixWorld adopts a unified DiT-based framework that takes noisy and clean multi-view inputs, with optional text conditioning, and jointly predicts depth and 3DGS through shared transformer blocks. (b) A pixel-space flow matching loss is imposed on rendered multi-view images to directly optimize the underlying 3D representation. (c) A geometry perception loss further enforces struct… view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of PixWorld under different settings. PixWorld flexibly handles both 3D reconstruction and generation: when all input views are clean, it performs reconstruction; when clean and noisy views are arbitrarily mixed, it performs generation. We visualize the camera trajectory, where blue and red frustums denote clean input views and generated views, respectively [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of comparisons with baselines. The large view on top denotes the input view, while the two smaller views below show novel views generated by each method. loss supplies the 3D structural signal that 2D objectives cannot, validating it as a key component of PixWorld. 5 CONCLUSION We present PixWorld, an end-to-end pixel-space diffusion framework that unifies 3D scene generation and reconstructi… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation study on the Geometry Perception loss in PixWorld. Given a single input image, our model generates the subsequent 7 frames (8 frames in total); we visualize 4 representative frames here for clarity. Pose accuracy is quantitatively evaluated by comparing the estimated camera poses against the ground-truth poses. Compared to the variant without Geometry Perception (w/o Geom.), the full model achieve… view at source ↗
Figure 6
Figure 6. Figure 6: More visualizations of reconstruction and generation under varying view selections, [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: More visualizations of generated scenes. The first view is the input, and we show both [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
read the original abstract

3D reconstruction and generation are commonly tackled by separate paradigms: pixel-based regression for reconstruction, and latent diffusion for generation. Recent works attempt to unify them in latent space, but with notable drawbacks: the diffusion objective is defined on latent features rather than the underlying 3D representation, and both branches suffer from information loss introduced by latent encoding, while requiring a pretrained Variational Autoencoder (VAE) or Representation Autoencoder (RAE). In this paper, we reformulate these two tasks under a unified pixel-space diffusion paradigm and introduce PixWorld, a single model that jointly addresses 3D reconstruction and generation. By supervising diffusion directly on rendered images, PixWorld removes the above limitations and aligns optimization with 3D scene fidelity. Beyond photometric and perceptual supervision that operates at the 2D image level and lacks 3D geometric awareness, we further introduce a geometry perception loss that aligns rendered views with their ground truth in the geometry-aware feature space of a pretrained 3D foundation model, providing 3D structural supervision. PixWorld consistently outperforms prior latent-space generation methods and matches state-of-the-art reconstruction methods, demonstrating the superiority of a unified pixel-space approach.

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 / 7 minor

Summary. The paper introduces PixWorld, a unified pixel-space diffusion framework for 3D scene reconstruction and generation. The core idea is to perform flow matching directly on rendered images from a pixel-aligned 3D Gaussian Splatting (3DGS) representation, eliminating the intermediate VAE/RAE latent stage used by prior latent-space 3D generation methods. The model uses a two-stream DiT that processes clean (reconstruction) and noisy (generation) views jointly, supervised by a flow-matching loss on rendered images, a depth loss, and a novel geometry perception loss that aligns rendered and ground-truth views in the feature space of a frozen 3D foundation model (pi3). Experiments span four benchmarks (RealEstate10K, DL3DV-10K, WorldScore) across reconstruction (4/8-view) and generation (1/2-view) settings, with disaggregated per-configuration results in the appendix.

Significance. The central claim — that pixel-space diffusion with differentiable 3DGS rendering can unify reconstruction and generation while matching or exceeding latent-space methods — is well-motivated and supported by a substantial experimental effort. The geometry perception loss (Sec. 3.3) is a sensible contribution that addresses a real gap in 2D photometric supervision. The ablation (Tab. 5) shows a clear 1.13 dB PSNR drop without it. The paper provides a detailed architecture breakdown (Tab. 6), inference-speed comparison (Tab. 9), and per-configuration disaggregation (Tabs. 7–8), which strengthen reproducibility. The claim of removing the VAE/RAE bottleneck is architecturally sound and the differentiable rendering pipeline is internally consistent.

major comments (2)
  1. §4.2 and Appendix E: The evaluation protocol for NVS metrics (PSNR/SSIM/LPIPS) does not specify how comparisons are handled when methods have different native output resolutions. Appendix E states that PixWorld runs at a lower output resolution (336×448) than video-diffusion baselines. The paper does not state whether all outputs are resized to a common resolution before metric computation, and if so, in which direction (up vs. down). This matters for the generation comparisons in Tables 2–3, where some margins are modest (e.g., +0.75 dB over Gen3R on DL3DV 1-view in Table 2; LVSM edges PixWorld on PSNR at 23.61 vs. 23.54 on RE10K 2-view in Table 3). The reconstruction comparison (Table 1) is less affected since baselines there are pixel-aligned Gaussian methods likely operating at similar resolutions. The concern is not that results are wrong, but that an unspecified protocol could conf
  2. Abstract and §4.4: The claim that PixWorld 'consistently outperforms prior latent-space generation methods' is strong but partially undercut by the resolution mismatch noted above and by the fact that LVSM (a deterministic regression model, not a latent-space diffusion method) is competitive on PSNR/SSIM in the 2-view setting (Table 3). The paper should either (a) narrow the claim to specify that it outperforms latent-space *diffusion* methods (excluding LVSM from the comparison group for this claim), or (b) clarify that 'consistently' refers to the aggregate across NVS, generation quality, and camera control metrics rather than PSNR alone. As stated, 'consistently outperforms' is not strictly supported by Table 3 where LVSM wins on PSNR and SSIM for RE10K 2-view.
minor comments (7)
  1. §3.2, Eq. (6): The LPIPS term is defined as L_lpips = 1[t > t_th] * (1/N) * sum LPIPS(bar_I_n, I_n), summing over all N views including clean ones. However, the text says 'perceptual supervision is unreliable when the noisy input is too close to pure noise,' which explains gating by t > t_th for noisy views but does not explain why clean views (which have no noise) are also gated. Please clarify whether the indicator is intended to apply to all views or only noisy ones.
  2. §4.1: The loss weights (lambda_depth=1.0, lambda_lpips=lambda_geo=0.1, t_th=0.3) are stated but no sensitivity analysis is provided beyond the single ablation on the geometry loss (Tab. 5). A brief note on robustness to these choices, or at least acknowledgment that they were selected by informal search, would strengthen the paper.
  3. Table 5: The ablation is conducted on a 10K-sequence subset for 30K steps, which is a reduced setting compared to the full training (67K scenes, 200K steps). While the controlled comparison is valid, the paper should note whether the relative effect of the geometry loss is expected to scale similarly at full training, or whether the ablation setting was chosen for compute reasons.
  4. §3.3, Eq. (9): The geometry perception loss uses cosine distance in the feature space of a frozen 3D foundation model. The choice of pi3 is mentioned in §4.1, and VGGT is listed as an alternative in §3.3. It would help to briefly note whether results with VGGT were tried and how the choice of critic affects results.
  5. Appendix E, Table 9: The inference-speed comparison notes that PixWorld runs at lower resolution than baselines, which reduces per-step compute. Stating the exact resolution difference (e.g., 'baselines operate at X×Y') would improve transparency.
  6. Figure 2: The '×?' in 'Shared DiT Block × ?' should be replaced with the actual count (24, per Appendix B).
  7. The paper uses '1-view' and '2-view' terminology throughout. On first use in §4.2, a brief parenthetical clarifying that '1-view' means single-image conditioning and '2-view' means two-image conditioning would improve readability for readers unfamiliar with the convention.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. We address both major comments below.

read point-by-point responses

  1. Referee: §4.2 and Appendix E: The evaluation protocol for NVS metrics (PSNR/SSIM/LPIPS) does not specify how comparisons are handled when methods have different native output resolutions. Appendix E states that PixWorld runs at a lower output resolution (336×448) than video-diffusion baselines. The paper does not state whether all outputs are resized to a common resolution before metric computation, and if so, in which direction (up vs. down).

    Authors: The referee is correct that the current manuscript does not specify the resolution protocol for NVS metric computation. We will add this to §4.2. Concretely: all metrics are computed at the ground-truth resolution of each test scene. For RealEstate10K and DL3DV-10K, the ground-truth target views are at 336×448 (the native resolution of both datasets' test clips as we process them). PixWorld renders directly at this resolution. For video-diffusion baselines that produce higher-resolution outputs, we bilinearly downsample their generated frames to 336×448 before computing PSNR/SSIM/LPIPS against the ground-truth targets. This means baselines are evaluated at a resolution equal to or lower than their native output, which if anything favors them (downsampling can slightly improve PSNR by suppressing high-frequency noise). We will state this explicitly in the revised §4.2 and note in Appendix E that the resolution difference does not advantage PixWorld. We agree this was an omission that should be corrected for reproducibility. revision: yes

  2. Referee: Abstract and §4.4: The claim that PixWorld 'consistently outperforms prior latent-space generation methods' is strong but partially undercut by the resolution mismatch noted above and by the fact that LVSM (a deterministic regression model, not a latent-space diffusion method) is competitive on PSNR/SSIM in the 2-view setting (Table 3). The paper should either (a) narrow the claim to specify that it outperforms latent-space diffusion methods (excluding LVSM from the comparison group for this claim), or (b) clarify that 'consistently' refers to the aggregate across NVS, generation quality, and camera control metrics rather than PSNR alone.

    Authors: The referee raises a fair point. We note that LVSM is a deterministic regression model, not a latent-space generation method, so it already falls outside the comparison group for the claim as literally stated ('prior latent-space generation methods'). However, we agree the wording is ambiguous because LVSM appears in the same tables and a reader could reasonably interpret 'consistently outperforms' as applying to all baselines shown. We will adopt the referee's option (b) and revise the abstract and §4.4 to clarify that 'consistently' refers to the aggregate across NVS, generation quality, and camera control metrics — not PSNR alone. We will also explicitly note in §4.4 that LVSM, as a deterministic regression model rather than a generative method, is competitive on raw PSNR/SSIM in the 2-view setting (gap <0.07 dB on RealEstate10K) but trails PixWorld on LPIPS, all generation-quality metrics, and camera control AUC. This makes the scope of the claim precise without overstating it. revision: yes

Circularity Check

0 steps flagged

No significant circularity: the derivation is self-contained with standard objectives, external frozen critics, and no fitted-parameter-as-prediction.

full rationale

The paper's derivation chain is self-contained. The flow matching objective (Eq. 2) is a standard rectified-flow formulation from JiT (Li & He, 2025), an external citation. The 3DGS rendering uses differentiable rasterization from Kerbl et al. (2023), external. The geometry perception loss (Eq. 9) uses a frozen pretrained 3D foundation model (π3, Wang et al. 2025d) as a structural critic — gradients are stopped on the reference branch, so Ψ is not fitted to the target. Depth supervision uses DA3 (Lin et al. 2025), also external and frozen. The clean/noisy partitioning (Eq. 3-4) is a training strategy, not a fitted parameter renamed as a prediction. The loss weights (λ_depth=1.0, λ_lpips=λ_geo=0.1, t_th=0.3) are hand-tuned hyperparameters, but the paper does not claim they are predicted or derived from first principles — they are stated as design choices. No 'prediction' or 'first-principles result' reduces to its inputs by construction. The ablation (Tab. 5) toggles the geometry perception loss under identical training settings, showing independent degradation when removed. While some authors appear on related prior work (e.g., OneWorld, Gao et al. 2026), these citations are contextual literature references, not load-bearing uniqueness theorems invoked to forbid alternatives. The central claim — that pixel-space diffusion supervising 3DGS through differentiable rendering outperforms latent-space alternatives — is an empirical claim validated against external benchmarks (RE10K, DL3DV, WorldScore), not a tautological reduction.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities, particles, or postulated objects. The 3D Gaussian representation, 3DGS renderer, flow matching objective, and foundation model critic are all existing external components. The free parameters are standard loss weights and training hyperparameters, not fitted constants in a derivation. The axioms are domain assumptions about the suitability of pixel-space diffusion and frozen-model features for 3D supervision, not ad hoc postulates.

free parameters (7)
  • λ_depth = 1.0
    Weight for depth supervision loss, hand-set.
  • λ_lpips = 0.1
    Weight for LPIPS perceptual loss, hand-set.
  • λ_geo = 0.1
    Weight for geometry perception loss, hand-set.
  • t_th = 0.3
    Threshold gating perceptual and geometric losses, hand-set.
  • Training resolution = 336×448
    Fixed training resolution, chosen by compute budget.
  • N (views per scene) = 4–8
    Range of multi-view samples per scene, hand-set.
  • t_th for LPIPS/geo gating = 0.3
    Same as t_th above; activates perceptual/geometric supervision only when noise level is sufficiently low.
axioms (4)
  • domain assumption Pixel-space diffusion can scale to multi-view 3D scene modeling without a latent compression stage.
    Section 3.1 extends 2D pixel-space diffusion to posed multi-view; the assumption that this is tractable is load-bearing for the entire framework.
  • domain assumption Differentiable rendering of 3DGS provides useful gradient signal for diffusion training.
    Section 3.2 defines the rendering loss; the assumption that gradients through the rasterizer are informative is unstated but central.
  • domain assumption The feature space of a frozen 3D foundation model (π3) encodes cross-view 3D structure useful as supervision.
    Section 3.3 introduces L_geo; the claim that π3 features capture geometric structure beyond 2D appearance is motivated but not independently verified.
  • domain assumption Pseudo-depth from DA3 is accurate enough to supervise depth prediction.
    Section 3.2, Eq. 7 uses DA3 depth as pseudo-ground-truth; the accuracy of this pseudo-label is assumed.

pith-pipeline@v1.1.0-glm · 28209 in / 2951 out tokens · 291622 ms · 2026-07-07T14:05:43.431816+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

43 extracted references · 43 canonical work pages · 25 internal anchors

  1. [1]

    VFM-VAE: Vision Foundation Models Can Be Good Tokenizers for Latent Diffusion Models

    Tianci Bi, Xiaoyi Zhang, Yan Lu, and Nanning Zheng. Vision foundation models can be good tokenizers for latent diffusion models.arXiv preprint arXiv:2510.18457,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Aligning visual foundation encoders to tokenizers for diffusion models

    Bowei Chen, Sai Bi, Hao Tan, He Zhang, Tianyuan Zhang, Zhengqi Li, Yuanjun Xiong, Jianming Zhang, and Kai Zhang. Aligning visual foundation encoders to tokenizers for diffusion models. arXiv preprint arXiv:2509.25162, 2025a. Jiuhai Chen, Zhiyang Xu, Xichen Pan, Yushi Hu, Can Qin, Tom Goldstein, Lifu Huang, Tianyi Zhou, Saining Xie, Silvio Savarese, et al....

    work page arXiv

  3. [3]

    Fantasyworld: Geometry-consistent world modeling via unified video and 3d prediction.arXiv preprint arXiv:2509.21657,

    Yixiang Dai, Fan Jiang, Chiyu Wang, Mu Xu, and Yonggang Qi. Fantasyworld: Geometry-consistent world modeling via unified video and 3d prediction.arXiv preprint arXiv:2509.21657,

    work page arXiv

  4. [4]

    Worldscore: A unified evaluation benchmark for world generation.arXiv preprint arXiv:2504.00983,

    Haoyi Duan, Hong-Xing Yu, Sirui Chen, Li Fei-Fei, and Jiajun Wu. Worldscore: A unified evaluation benchmark for world generation.arXiv preprint arXiv:2504.00983,

    work page arXiv

  5. [5]

    CAT3D: Create Anything in 3D with Multi-View Diffusion Models

    Ruiqi Gao, Aleksander Holynski, Philipp Henzler, Arthur Brussee, Ricardo Martin-Brualla, Pratul Srinivasan, Jonathan T Barron, and Ben Poole. Cat3d: Create anything in 3d with multi-view diffusion models.arXiv preprint arXiv:2405.10314,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    Oneworld: Taming scene generation with 3d unified representation autoencoder.arXiv preprint arXiv:2603.16099,

    Sensen Gao, Zhaoqing Wang, Qihang Cao, Dongdong Yu, Changhu Wang, Tongliang Liu, Mingming Gong, and Jiawang Bian. Oneworld: Taming scene generation with 3d unified representation autoencoder.arXiv preprint arXiv:2603.16099,

    work page arXiv

  7. [7]

    Vist3a: Text-to-3d by stitching a multi-view reconstruction network to a video generator

    Hyojun Go, Dominik Narnhofer, Goutam Bhat, Prune Truong, Federico Tombari, and Konrad Schindler. Vist3a: Text-to-3d by stitching a multi-view reconstruction network to a video generator. arXiv preprint arXiv:2510.13454, 2025a. Hyojun Go, Byeongjun Park, Jiho Jang, Jin-Young Kim, Soonwoo Kwon, and Changick Kim. Splatflow: Multi-view rectified flow model fo...

    work page arXiv

  8. [8]

    LRM: Large Reconstruction Model for Single Image to 3D

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. Lrm: Large reconstruction model for single image to 3d.arXiv preprint arXiv:2311.04400,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Gen3r: 3d scene generation meets feed-forward reconstruction.arXiv preprint arXiv:2601.04090,

    Jiaxin Huang, Yuanbo Yang, Bangbang Yang, Lin Ma, Yuewen Ma, and Yiyi Liao. Gen3r: 3d scene generation meets feed-forward reconstruction.arXiv preprint arXiv:2601.04090,

    work page arXiv

  10. [10]

    Repurposing geometric foundation models for multi-view diffusion

    Wooseok Jang, Seonghu Jeon, Jisang Han, Jinhyeok Choi, Minkyung Kwon, Seungryong Kim, Saining Xie, and Sainan Liu. Repurposing geometric foundation models for multi-view diffusion. arXiv preprint arXiv:2603.22275,

    work page arXiv

  11. [11]

    Auto-Encoding Variational Bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    3d and 4d world modeling: A survey.arXiv preprint arXiv:2509.07996,

    Lingdong Kong, Wesley Yang, Jianbiao Mei, Youquan Liu, Ao Liang, Dekai Zhu, Dongyue Lu, Wei Yin, Xiaotao Hu, Mingkai Jia, et al. 3d and 4d world modeling: A survey.arXiv preprint arXiv:2509.07996,

    work page arXiv

  13. [13]

    Back to Basics: Let Denoising Generative Models Denoise

    Ruilong Li, Brent Yi, Junchen Liu, Hang Gao, Yi Ma, and Angjoo Kanazawa. Cameras as relative positional encoding.Advances in Neural Information Processing Systems, 2025a. Tianhong Li and Kaiming He. Back to basics: Let denoising generative models denoise.arXiv preprint arXiv:2511.13720,

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Flashworld: High-quality 3d scene generation within seconds.arXiv preprint arXiv:2510.13678, 2025b

    Xinyang Li, Tengfei Wang, Zixiao Gu, Shengchuan Zhang, Chunchao Guo, and Liujuan Cao. Flashworld: High-quality 3d scene generation within seconds.arXiv preprint arXiv:2510.13678, 2025b. Xiuyu Li, Yijiang Liu, Long Lian, Huanrui Yang, Zhen Dong, Daniel Kang, Shanghang Zhang, and Kurt Keutzer. Q-diffusion: Quantizing diffusion models. InProceedings of the I...

    work page arXiv

  15. [15]

    Depth Anything 3: Recovering the Visual Space from Any Views

    12 Haotong Lin, Sili Chen, Junhao Liew, Donny Y Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang. Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647,

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model

    Fangfu Liu, Wenqiang Sun, Hanyang Wang, Yikai Wang, Haowen Sun, Junliang Ye, Jun Zhang, and Yueqi Duan. Reconx: Reconstruct any scene from sparse views with video diffusion model.arXiv preprint arXiv:2408.16767,

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    SyncDreamer: Generating Multiview-consistent Images from a Single-view Image

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Generating multiview-consistent images from a single-view image.arXiv preprint arXiv:2309.03453,

    work page internal anchor Pith review Pith/arXiv arXiv

  18. [18]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101,

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [19]

    DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generation

    Zehong Ma, Longhui Wei, Shuai Wang, Shiliang Zhang, and Qi Tian. Deco: Frequency-decoupled pixel diffusion for end-to-end image generation.arXiv preprint arXiv:2511.19365,

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis.arXiv preprint arXiv:2307.01952,

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    DreamFusion: Text-to-3D using 2D Diffusion

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988,

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    ZeroNVS: Zero-Shot 360-Degree View Synthesis from a Single Image

    Kyle Sargent, Zizhang Li, Tanmay Shah, Charles Herrmann, Hong-Xing Yu, Yunzhi Zhang, Eric Ryan Chan, Dmitry Lagun, Li Fei-Fei, Deqing Sun, et al. Zeronvs: Zero-shot 360-degree view synthesis from a single real image.arXiv preprint arXiv:2310.17994,

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    Post-training quantization on diffusion models

    Yuzhang Shang, Zhihang Yuan, Bin Xie, Bingzhe Wu, and Yan Yan. Post-training quantization on diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1972–1981,

    work page 1972

  24. [24]

    GLU Variants Improve Transformer

    Noam Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202,

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  25. [25]

    Latent diffusion model without variational autoencoder.arXiv preprint arXiv:2510.15301,

    Minglei Shi, Haolin Wang, Wenzhao Zheng, Ziyang Yuan, Xiaoshi Wu, Xintao Wang, Pengfei Wan, Jie Zhou, and Jiwen Lu. Latent diffusion model without variational autoencoder.arXiv preprint arXiv:2510.15301,

    work page arXiv

  26. [26]

    MVDream: Multi-view Diffusion for 3D Generation

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d generation.arXiv preprint arXiv:2308.16512,

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    DimensionX: Create Any 3D and 4D Scenes from a Single Image with Controllable Video Diffusion

    13 Wenqiang Sun, Shuo Chen, Fangfu Liu, Zilong Chen, Yueqi Duan, Jun Zhang, and Yikai Wang. Dimensionx: Create any 3d and 4d scenes from a single image with controllable video diffusion. arXiv preprint arXiv:2411.04928,

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    Vggt-world: Transforming vggt into an autoregressive geometry world model.arXiv preprint arXiv:2603.12655,

    Xiangyu Sun, Shijie Wang, Fengyi Zhang, Lin Liu, Caiyan Jia, Ziying Song, Zi Huang, and Yadan Luo. Vggt-world: Transforming vggt into an autoregressive geometry world model.arXiv preprint arXiv:2603.12655,

    work page arXiv

  29. [29]

    DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation

    Jiaxiang Tang, Jiawei Ren, Hang Zhou, Ziwei Liu, and Gang Zeng. Dreamgaussian: Generative gaussian splatting for efficient 3d content creation.arXiv preprint arXiv:2309.16653,

    work page internal anchor Pith review Pith/arXiv arXiv

  30. [30]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video generative models. arXiv preprint arXiv:2503.20314,

    work page internal anchor Pith review Pith/arXiv arXiv

  31. [31]

    MMGen: Unified Multi-modal Image Generation and Understanding in One Go

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. InIEEE Conf. Comput. Vis. Pattern Recog., pp. 5294–5306, 2025a. Jiepeng Wang, Zhaoqing Wang, Hao Pan, Yuan Liu, Dongdong Yu, Changhu Wang, and Wenping Wang. Mmgen: Unified multi-modal image generation and understa...

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    $\pi^3$: Permutation-Equivariant Visual Geometry Learning

    Yifan Wang, Jianjun Zhou, Haoyi Zhu, Wenzheng Chang, Yang Zhou, Zizun Li, Junyi Chen, Jiangmiao Pang, Chunhua Shen, and Tong He. π3: Permutation-equivariant visual geometry learning.arXiv preprint arXiv:2507.13347, 2025d. Zhaoqing Wang, Xiaobo Xia, Zhuolin Bie, Jinlin Liu, Dongdong Yu, Jia-Wang Bian, and Changhu Wang. Taming camera-controlled video genera...

    work page internal anchor Pith review Pith/arXiv arXiv

  33. [33]

    Geometry Forcing: Marrying Video Diffusion and 3D Representation for Consistent World Modeling

    Haoyu Wu, Diankun Wu, Tianyu He, Junliang Guo, Yang Ye, Yueqi Duan, and Jiang Bian. Geometry forcing: Marrying video diffusion and 3d representation for consistent world modeling.arXiv preprint arXiv:2507.07982,

    work page internal anchor Pith review Pith/arXiv arXiv

  34. [34]

    World Action Models are Zero-shot Policies

    Botao Ye, Boqi Chen, Haofei Xu, Daniel Barath, and Marc Pollefeys. Yonosplat: You only need one model for feedforward 3d gaussian splatting. InInternational Conference on Learning Representations (ICLR), 2026a. Seonghyeon Ye, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, You Liang Tan, Chuning Zhu, Jiannan Xiang, et al...

    work page internal anchor Pith review Pith/arXiv arXiv

  35. [35]

    PixelDiT: Pixel Diffusion Transformers for Image Generation

    Hong-Xing Yu, Haoyi Duan, Charles Herrmann, William T Freeman, and Jiajun Wu. Wonderworld: Interactive 3d scene generation from a single image. InIEEE Conf. Comput. Vis. Pattern Recog., pp. 5916–5926, 2025a. Yongsheng Yu, Wei Xiong, Weili Nie, Yichen Sheng, Shiqiu Liu, and Jiebo Luo. Pixeldit: Pixel diffusion transformers for image generation.arXiv prepri...

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [36]

    GenXD: Generating Any 3D and 4D Scenes

    Yuyang Zhao, Chung-Ching Lin, Kevin Lin, Zhiwen Yan, Linjie Li, Zhengyuan Yang, Jianfeng Wang, Gim Hee Lee, and Lijuan Wang. Genxd: Generating any 3d and 4d scenes.arXiv preprint arXiv:2411.02319,

    work page internal anchor Pith review Pith/arXiv arXiv

  37. [37]

    Diffusion Transformers with Representation Autoencoders

    Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with represen- tation autoencoders.arXiv preprint arXiv:2510.11690,

    work page internal anchor Pith review Pith/arXiv arXiv

  38. [38]

    Drivinggen: A comprehensive benchmark for generative video world models in autonomous driving.arXiv preprint arXiv:2601.01528,

    Yang Zhou, Hao Shao, Letian Wang, Zhuofan Zong, Hongsheng Li, and Steven L Waslander. Drivinggen: A comprehensive benchmark for generative video world models in autonomous driving.arXiv preprint arXiv:2601.01528,

    work page arXiv

  39. [39]

    17.12 0.622 0.394 0.959 0.961 0.569 0.530 0.664 0.521 0.338 FlashWorld (Li et al., 2025b) 16.26 0.613 0.417 0.951 0.9540.6170.544 0.849 0.766 0.553 Gen3R (Huang et al.,

    work page arXiv

  40. [40]

    Finally, Fig

    15.940.507 0.487 0.948 0.944 0.551 0.532 0.571 0.389 0.133 PixWorld (Ours) 16.63 0.533 0.443 0.956 0.958 0.637 0.571 0.797 0.709 0.493 with the rendered appearance, indicating that the joint depth and 3D-Gaussian prediction in PixWorld captures scene geometry faithfully even under heavy extrapolation from a single conditioning image. Finally, Fig. 5 compl...

    work page 2023

  41. [41]

    Best inbold; second best underlined

    under 2-view Interpolation and 2-view Extrapolation. Best inbold; second best underlined . Novel View Synthesis Generation Quality Camera Control Method PSNR↑SSIM↑LPIPS↓I2V Subj.↑I2V BG↑I.Q.↑Aes.Q.↑AUC@30↑AUC@15↑AUC@5↑ RealEstate10K– 2-view Interpolation LVSM (Jin et al., 2025)24.26 0.832 0.2000.969 0.963 0.613 0.518 0.871 0.806 0.638 GF (Wu et al.,

    work page 2025

  42. [42]

    20.24 0.695 0.307 0.9760.970 0.556 0.542 0.677 0.521 0.220 PixWorld (Ours) 23.63 0.819 0.206 0.978 0.974 0.631 0.564 0.883 0.819 0.651 DL3DV-10K– 2-view Interpolation LVSM (Jin et al., 2025)19.37 0.594 0.3280.918 0.917 0.538 0.511 0.758 0.630 0.403 GF (Wu et al.,

    work page 2025

  43. [43]

    PixWorld takes a step toward unifying 3D scene reconstruction and generation in pixel space, but several directions remain open

    49 100 882 FlashWorld (Li et al., 2025b) 24 410 PixWorld (Ours) 8 100 15 F RESPONSIBLECONSIDERATIONS F.1 LIMITATIONS. PixWorld takes a step toward unifying 3D scene reconstruction and generation in pixel space, but several directions remain open. Our experiments focus on widely used scene-level datasets such as RealEstate10K and DL3DV-10K; further evaluat...

    work page 2023