pith. sign in

arxiv: 2606.02552 · v1 · pith:NEUORS7Fnew · submitted 2026-06-01 · 💻 cs.CV · cs.AI

Modeling Depth Ambiguity: A Mixture-Density Representation for Flying-Point-Free Depth Estimation

Siyuan Bian , Congrong Xu , Jun Gao This is my paper

Pith reviewed 2026-06-28 15:30 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords depth estimationflying pointsmixture densityboundary reconstructiondepth ambiguitytransparent objectssky regions
0
0 comments X

The pith

Modeling depth per pixel as a mixture of hypotheses eliminates flying points by allowing separate surface alignments at boundaries.

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

Depth estimators produce flying points near boundaries because they force each pixel to a single depth value. When a pixel overlaps foreground and background, training pulls this value to an average that lies in empty space. MDA counters this by outputting a mixture of depth hypotheses with probabilities for every pixel. The final depth is then chosen from one hypothesis rather than computed as an average. This yields better boundaries and removes the artifacts across backbones with negligible added cost, and the mixture idea also handles transparency and sky regions.

Core claim

The paper establishes that replacing the single-depth output with a mixture-density representation allows the model to maintain multiple possible depths at ambiguous pixels. Different mixture components can align with different surfaces, so the decoded depth comes from an actual surface rather than the space between them. This directly addresses the source of flying points without changing the underlying network architecture.

What carries the argument

The MDA mixture-density representation, which outputs multiple depth hypotheses and their probabilities per pixel and decodes depth by selecting from these hypotheses.

If this is right

  • Boundary reconstruction improves substantially across different network backbones.
  • Flying-point artifacts are largely removed even under severe input blur.
  • The same framework predicts multiple depth layers for transparent objects.
  • A dedicated sky component separates unbounded regions from finite depths to produce clean skylines.
  • Runtime overhead stays negligible.

Where Pith is reading between the lines

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

  • Similar mixture representations could address depth ambiguities in other scenarios like reflections or occlusions.
  • Adopting MDA might require only swapping the final prediction layer in existing depth networks.
  • Dynamic determination of the number of mixture components per pixel could further improve flexibility.
  • Testing on datasets with more varied boundary conditions would strengthen evidence for the modeling choice being the main cause.

Load-bearing premise

That the dominant source of flying points is the single-depth modeling choice at ambiguous boundary pixels rather than other factors such as network capacity, loss design, or dataset statistics.

What would settle it

Compare flying-point counts at object boundaries between a standard single-depth network and an MDA network trained identically on the same dataset; a large reduction only in the MDA version would confirm the modeling change as the key factor.

Figures

Figures reproduced from arXiv: 2606.02552 by Congrong Xu, Jun Gao, Siyuan Bian.

Figure 1
Figure 1. Figure 1: Overview of our approach. Existing depth estimators model each pixel as a unimodal distribution, producing flying-point artifacts at boundaries. Our mixture-density model maintains multiple depth hypotheses, eliminating boundary artifacts, recovering layered depth behind transparent objects, and providing a clean skyline. ℓ1 and ℓ2 depth losses are negative log-likelihoods of Laplacian or Gaussian distribu… view at source ↗
Figure 2
Figure 2. Figure 2: Three forms of depth ambiguity. Ray 1 (Boundary): the pixel straddles a foreground edge and a background surface, producing two depth hypotheses whose relative mixture weights encode the model’s belief on which surface dominates. Ray 2 (Transparent object): the ray physically intersects multiple surfaces (e.g., the two sides of a glass cup and the background wall), and all of them are simultaneously valid … view at source ↗
Figure 3
Figure 3. Figure 3: Unimodal versus mixture-density depth at an object boundary. (a) A boundary pixel may mix foreground and background evidence along the camera ray. (b) A unimodal predictor must output one depth, often averaging the foreground (d1) and background (d2) hypotheses into an intermediate estimate (d3) that becomes a flying point. (c) Our mixture-density representation keeps multiple hypotheses and turns decoding… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative boundary comparison. Baseline methods (DA3, VGGT, PPD) leave visible flying points on the boundaries, while our approach always keeps the boundary clean. directly penalizes predicted points that fall away from both foreground and background surfaces [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: Boundary estimation Accuracy on NRGBD as a function of input blur s (Acc↓, mm). Our mixture model degrades less compared to baselines. Robustness to Input Blur. Our mixture-density rep￾resentation is especially useful when boundary evi￾dence is weakened by blur. We simulate degraded inputs by downsampling each frame by factor s with area averaging and bicubic upsampling it back to the model resolution; lar… view at source ↗
Figure 6
Figure 6. Figure 6: Per-component visualization with K=4 components. Top: the input image (leftmost) and the per-pixel mixture weight πk for each head (brighter pixels indicate where head k wins the argmax). Bottom: our final fused depth (leftmost) and each head’s mean depth Dˆ k. The four heads specialize spatially: each head is dominant in a different region, and the boundaries between regions concentrate at occlusion edges… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative boundary reconstruction under input blur. As s increases, baselines (DA3, PPD) accumulate thick bands of flying points at boundaries, while our model preserves clean boundary separation throughout. To isolate their effects, [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results on transparent objects and sky. Sky Estimation. To validate the dedicated sky component, we evaluate sky-segmentation quality on Sintel, reporting IoU against its semantic-segmentation ground truth; full quantitative results are deferred to §C.2.2. Figure 8b shows qualitative comparisons against a baseline identical to ours except for the missing sky component. Without the sky component… view at source ↗
Figure 9
Figure 9. Figure 9: Additional qualitative boundary reconstruction under input blur. As [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative multi-layer depth on the LayeredDepth real-world set [ [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Two representative frames from Sintel temple_3: the input RGB and the ground-truth sky mask (green = sky). Sky pixels dominate most of the frame — a sky-dominant configuration that does not appear in our synthetic training mix — which explains the near-zero IoU on this sequence reported in [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative comparison on sky regions. Without a dedicated sky component, the baseline [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: highlights two characteristic failure modes of our mixture representation. First ( [PITH_FULL_IMAGE:figures/full_fig_p023_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Additional qualitative boundary comparison across nine scenes drawn from 7Scenes, [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Per-component visualization of heads with [PITH_FULL_IMAGE:figures/full_fig_p025_15.png] view at source ↗
read the original abstract

Despite advances in depth estimation, flying points remain a persistent failure mode: near object boundaries, depth estimators often predict spurious 3D points in the empty space between foreground and background surfaces. We trace this artifact to a standard modeling choice: assigning each pixel a single depth hypothesis. At boundaries, a pixel can straddle a foreground and a background surface, so its true depth is ambiguous between the two. A model that predicts a single depth cannot keep both possibilities, so training instead pulls the prediction toward an intermediate depth that lies on neither surface. We address this with MDA, a mixture-density representation that lets the model predict multiple depth hypotheses and their associated probabilities for each pixel. Near boundaries, different hypotheses can align with different surfaces, and the decoded depth is selected from one of these hypotheses rather than placed in the empty space between them. Across different backbones, MDA substantially improves boundary reconstruction and largely removes flying-point artifacts even under severe input blur, while adding negligible runtime overhead. The same mixture-density framework naturally extends to transparent objects, where it predicts multiple depth layers at transparent pixels, and to sky regions, where a dedicated component separates the unbounded sky from finite-depth regions, producing flying-point-free skylines. Project Page: https://biansy000.github.io/mda-site/.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper claims that flying-point artifacts in monocular depth estimation arise from the standard single-depth hypothesis per pixel, which forces intermediate predictions at ambiguous boundary pixels; it introduces MDA, a mixture-density representation allowing multiple depth hypotheses and probabilities per pixel so that decoded depths align with surfaces rather than empty space. The approach is asserted to substantially improve boundary reconstruction and remove flying points across backbones even under severe blur, with negligible overhead, and to extend naturally to multi-layer depths for transparent objects and sky separation.

Significance. If the central attribution holds after proper controls, the work supplies a clean representational fix for a persistent artifact rather than an architectural or loss tweak, with practical value for 3D reconstruction pipelines and extensibility to transparency and unbounded regions. The negligible runtime claim, if verified, strengthens deployability.

major comments (2)
  1. [Experiments] Experiments section: the claim that gains are driven by the mixture representation (rather than output dimensionality, loss formulation, or training dynamics) requires explicit controls. Baselines must be re-trained with output heads matched in channel count to MDA's multiple hypotheses plus probabilities; without this, the attribution to single-depth modeling as the dominant cause of flying points cannot be isolated.
  2. [Abstract and §4] Abstract and §4 (results): qualitative statements of 'substantially improves boundary reconstruction' and 'largely removes flying-point artifacts' must be backed by quantitative tables reporting boundary-specific metrics (e.g., edge F-score, flying-point count) and ablations across backbones; the current absence of such numbers leaves the cross-backbone applicability unverified.
minor comments (2)
  1. [Method] Method section: clarify the exact decoding procedure from mixture components to final depth map (e.g., whether argmax probability or expectation is used) and how it differs from standard regression losses.
  2. [Figures] Figure captions: add quantitative annotations (e.g., flying-point counts or boundary error) to qualitative result figures to make visual claims directly verifiable.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The points raised highlight opportunities to strengthen the experimental validation of MDA's benefits. We address each major comment below and will incorporate the suggested controls and metrics in the revised manuscript.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the claim that gains are driven by the mixture representation (rather than output dimensionality, loss formulation, or training dynamics) requires explicit controls. Baselines must be re-trained with output heads matched in channel count to MDA's multiple hypotheses plus probabilities; without this, the attribution to single-depth modeling as the dominant cause of flying points cannot be isolated.

    Authors: We agree that isolating the contribution of the mixture-density representation requires controlling for output dimensionality. In the revision we will retrain the single-hypothesis baselines with expanded output heads that produce the same number of depth channels as MDA (while retaining their original loss and training procedure). The resulting comparisons will be added to the experiments section to better support the attribution of flying-point reduction to the mixture modeling itself. revision: yes

  2. Referee: [Abstract and §4] Abstract and §4 (results): qualitative statements of 'substantially improves boundary reconstruction' and 'largely removes flying-point artifacts' must be backed by quantitative tables reporting boundary-specific metrics (e.g., edge F-score, flying-point count) and ablations across backbones; the current absence of such numbers leaves the cross-backbone applicability unverified.

    Authors: We acknowledge that the manuscript would benefit from explicit quantitative support for the boundary and flying-point claims. We will add a dedicated table (and corresponding ablations) in Section 4 that reports edge F-score, flying-point counts, and related boundary metrics for MDA and the baselines across the evaluated backbones. These numbers will also be referenced in the abstract to substantiate the qualitative statements. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes MDA, a mixture-density representation for depth estimation, tracing flying points to single-depth modeling at ambiguous boundary pixels and claiming empirical improvements across backbones. No equations, derivations, or load-bearing steps are shown that reduce any prediction or result to a fitted input or self-citation by construction. The central contribution is a representational change with reported empirical gains; no self-definitional, fitted-input, or uniqueness-imported patterns appear in the abstract or described claims. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies insufficient detail to enumerate concrete free parameters, axioms, or invented entities; the core modeling assumption (mixture per pixel) is treated as a standard density-estimation technique whose validity is not independently verified here.

pith-pipeline@v0.9.1-grok · 5760 in / 1045 out tokens · 37590 ms · 2026-06-28T15:30:01.703235+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

37 extracted references · 5 linked inside Pith

  1. [1]

    Goldman, Matthias Nießner, and Justus Thies

    Dejan Azinovi ´c, Ricardo Martin-Brualla, Dan B. Goldman, Matthias Nießner, and Justus Thies. Neural RGB-D surface reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022

    2022

  2. [2]

    Estimating or propagating gradients through stochastic neurons for conditional computation.arXiv preprint arXiv:1308.3432, 2013

    Yoshua Bengio, Nicholas Léonard, and Aaron Courville. Estimating or propagating gradients through stochastic neurons for conditional computation.arXiv preprint arXiv:1308.3432, 2013

    Pith/arXiv arXiv 2013

  3. [3]

    Accurately computing the log-sum-exp and softmax functions.IMA Journal of Numerical Analysis, 41(4):2311–2330, 2021

    Pierre Blanchard, Desmond J Higham, and Nicholas J Higham. Accurately computing the log-sum-exp and softmax functions.IMA Journal of Numerical Analysis, 41(4):2311–2330, 2021

    2021

  4. [4]

    Butler, Jonas Wulff, Garrett B

    Daniel J. Butler, Jonas Wulff, Garrett B. Stanley, and Michael J. Black. A naturalistic open source movie for optical flow evaluation. InEuropean Conference on Computer Vision, 2012

    2012

  5. [5]

    Virtual KITTI 2.arXiv preprint arXiv:2001.10773, 2020

    Yohann Cabon, Naila Murray, and Martin Humenberger. Virtual KITTI 2.arXiv preprint arXiv:2001.10773, 2020

    Pith/arXiv arXiv 2001

  6. [6]

    Depth map prediction from a single image using a multi-scale deep network.Advances in neural information processing systems, 27, 2014

    David Eigen, Christian Puhrsch, and Rob Fergus. Depth map prediction from a single image using a multi-scale deep network.Advances in neural information processing systems, 27, 2014

    2014

  7. [7]

    Vision meets robotics: The KITTI dataset.International Journal of Robotics Research (IJRR), 32(11):1231–1237, 2013

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The KITTI dataset.International Journal of Robotics Research (IJRR), 32(11):1231–1237, 2013. 10

    2013

  8. [8]

    DeepMVS: Learning multi-view stereopsis

    Po-Han Huang, Kevin Matzen, Johannes Kopf, Narendra Ahuja, and Jia-Bin Huang. DeepMVS: Learning multi-view stereopsis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

    2018

  9. [9]

    DynamicStereo: Consistent dynamic depth from stereo videos

    Nikita Karaev, Ignacio Rocco, Benjamin Graham, Natalia Neverova, Andrea Vedaldi, and Christian Rupprecht. DynamicStereo: Consistent dynamic depth from stereo videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

    2023

  10. [10]

    Mapanything: Universal feed-forward metric 3d reconstruction.arXiv preprint arXiv:2509.13414, 2025

    Nikhil Keetha, Norman Müller, Johannes Schönberger, Lorenzo Porzi, Yuchen Zhang, Tobias Fischer, Arno Knapitsch, Duncan Zauss, Ethan Weber, Nelson Antunes, et al. Mapanything: Universal feed-forward metric 3d reconstruction.arXiv preprint arXiv:2509.13414, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    Stream3r: Scalable sequential 3d reconstruction with causal transformer.arXiv preprint arXiv:2508.10893, 2025

    Yushi Lan, Yihang Luo, Fangzhou Hong, Shangchen Zhou, Honghua Chen, Zhaoyang Lyu, Shuai Yang, Bo Dai, Chen Change Loy, and Xingang Pan. Stream3r: Scalable sequential 3d reconstruction with causal transformer.arXiv preprint arXiv:2508.10893, 2025

    arXiv 2025

  12. [12]

    Grounding image matching in 3d with mast3r

    Vincent Leroy, Yohann Cabon, and Jérôme Revaud. Grounding image matching in 3d with mast3r. InEuropean conference on computer vision, pages 71–91. Springer, 2024

    2024

  13. [13]

    Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang

    Haotong Lin, Sili Chen, Jun Hao 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, 2025

    Pith/arXiv arXiv 2025

  14. [14]

    Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision

    Lu Ling, Yichen Sheng, Zhi Tu, Wentian Zhao, Cheng Xin, Kun Wan, Lantao Yu, Qianyu Guo, Zixun Yu, Yawen Lu, et al. Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22160–22169, 2024

    2024

  15. [15]

    ReFusion: 3d reconstruction in dynamic environments for RGB-D cameras exploiting residuals

    Emanuele Palazzolo, Jens Behley, Philipp Lottes, Philippe Giguère, and Cyrill Stachniss. ReFusion: 3d reconstruction in dynamic environments for RGB-D cameras exploiting residuals. InIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019

    2019

  16. [16]

    Aria digital twin: A new benchmark dataset for egocentric 3d machine perception

    Xiaqing Pan, Nicholas Charron, Yongqian Yang, Scott Peters, Thomas Whelan, Chen Kong, Omkar Parkhi, Richard Newcombe, and Yuheng Carl Ren. Aria digital twin: A new benchmark dataset for egocentric 3d machine perception. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 20133–20143, 2023

    2023

  17. [17]

    Open challenges in deep stereo: the booster dataset

    Pierluigi Zama Ramirez, Fabio Tosi, Matteo Poggi, Samuele Salti, Stefano Mattoccia, and Luigi Di Stefano. Open challenges in deep stereo: the booster dataset. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21168–21178, 2022

    2022

  18. [18]

    René Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, and Vladlen Koltun. Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer.IEEE transactions on pattern analysis and machine intelligence, 44(3):1623–1637, 2020

    2020

  19. [19]

    Vision transformers for dense prediction

    René Ranftl, Alexey Bochkovskiy, and Vladlen Koltun. Vision transformers for dense prediction. InProceedings of the IEEE/CVF international conference on computer vision, pages 12179– 12188, 2021

    2021

  20. [20]

    Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding

    Mike Roberts, Jason Ramapuram, Anurag Ranjan, Atulit Kumar, Miguel Angel Bautista, Nathan Paczan, Russ Webb, and Joshua M Susskind. Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding. InProceedings of the IEEE/CVF international conference on computer vision, pages 10912–10922, 2021

    2021

  21. [21]

    Clear grasp: 3d shape estimation of transparent objects for manipulation

    Shreeyak Sajjan, Matthew Moore, Mike Pan, Ganesh Nagaraja, Johnny Lee, Andy Zeng, and Shuran Song. Clear grasp: 3d shape estimation of transparent objects for manipulation. In2020 IEEE international conference on robotics and automation (ICRA), pages 3634–3642. IEEE, 2020

    2020

  22. [22]

    Scene coordinate regression forests for camera relocalization in RGB-D images

    Jamie Shotton, Ben Glocker, Christopher Zach, Shahram Izadi, Antonio Criminisi, and Andrew Fitzgibbon. Scene coordinate regression forests for camera relocalization in RGB-D images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2013. 11

    2013

  23. [23]

    Smd-nets: Stereo mixture density networks

    Fabio Tosi, Yiyi Liao, Carolin Schmitt, and Andreas Geiger. Smd-nets: Stereo mixture density networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8942–8952, 2021

    2021

  24. [24]

    Vggt: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

    2025

  25. [25]

    Continuous 3d perception model with persistent state

    Qianqian Wang, Yifei Zhang, Aleksander Holynski, Alexei A Efros, and Angjoo Kanazawa. Continuous 3d perception model with persistent state. pages 10510–10522, 2025

    2025

  26. [26]

    Dust3r: Geometric 3d vision made easy

    Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, and Jerome Revaud. Dust3r: Geometric 3d vision made easy. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 20697–20709, 2024

    2024

  27. [27]

    Tartanair: A dataset to push the limits of visual slam

    Wenshan Wang, Delong Zhu, Xiangwei Wang, Yaoyu Hu, Yuheng Qiu, Chen Wang, Yafei Hu, Ashish Kapoor, and Sebastian Scherer. Tartanair: A dataset to push the limits of visual slam. In2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4909–4916. IEEE, 2020

    2020

  28. [28]

    π3: Permutation-equivariant visual geometry learning.arXiv preprint arXiv:2507.13347, 2025

    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, 2025

    Pith/arXiv arXiv 2025

  29. [29]

    Wang, and Jeong Joon Park

    Zichen Wang, Ang Cao, Liam J. Wang, and Jeong Joon Park. MoE3D: A mixture-of-experts module for 3D reconstruction.arXiv preprint arXiv:2601.05208, 2026

    arXiv 2026

  30. [30]

    Seeing and seeing through the glass: Real and synthetic data for multi-layer depth estimation

    Hongyu Wen, Yiming Zuo, Venkat Subramanian, Patrick Chen, and Jia Deng. Seeing and seeing through the glass: Real and synthetic data for multi-layer depth estimation. pages 6715–6725, 2025

    2025

  31. [31]

    Pixel-perfect depth with semantics- prompted diffusion transformers.arXiv preprint arXiv:2510.07316, 2025

    Gangwei Xu, Haotong Lin, Hongcheng Luo, Xianqi Wang, Jingfeng Yao, Lianghui Zhu, Yuechuan Pu, Cheng Chi, Haiyang Sun, Bing Wang, et al. Pixel-perfect depth with semantics- prompted diffusion transformers.arXiv preprint arXiv:2510.07316, 2025

    arXiv 2025

  32. [32]

    Pixel-perfect visual geometry estimation.arXiv preprint arXiv:2601.05246, 2026

    Gangwei Xu, Haotong Lin, Hongcheng Luo, Haiyang Sun, Bing Wang, Guang Chen, Sida Peng, Hangjun Ye, and Xin Yang. Pixel-perfect visual geometry estimation.arXiv preprint arXiv:2601.05246, 2026

    arXiv 2026

  33. [33]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10371–10381, 2024

    2024

  34. [34]

    Depth anything v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2. volume 37, pages 21875–21911, 2024

    2024

  35. [35]

    UnrealStereo: Controlling hazardous factors to analyze stereo vision

    Yi Zhang, Weichao Qiu, Qi Chen, Xiaolin Hu, and Alan Yuille. UnrealStereo: Controlling hazardous factors to analyze stereo vision. InInternational Conference on 3D Vision (3DV), 2018

    2018

  36. [36]

    Harley, Bokui Shen, Gordon Wetzstein, and Leonidas J

    Yang Zheng, Adam W. Harley, Bokui Shen, Gordon Wetzstein, and Leonidas J. Guibas. PointOdyssey: A large-scale synthetic dataset for long-term point tracking. InProceedings of the IEEE/CVF International Conference on Computer Vision, 2023

    2023

  37. [37]

    DA3-nested

    Yang Zhou, Yifan Wang, Jianjun Zhou, Wenzheng Chang, Haoyu Guo, Zizun Li, Kaijing Ma, Xinyue Li, Yating Wang, Haoyi Zhu, et al. Omniworld: A multi-domain and multi-modal dataset for 4d world modeling.arXiv preprint arXiv:2509.12201, 2025. 12 Supplementary Material Modeling Depth Ambiguity: A Mixture-Density Representation for Flying-Point-Free Depth Estim...

    arXiv 2025