pith. machine review for the scientific record. sign in

arxiv: 2604.08532 · v1 · submitted 2026-04-09 · 💻 cs.CV

Recognition: unknown

Self-Improving 4D Perception via Self-Distillation

Angjoo Kanazawa, Haiwen Feng, James M. Rehg, Nan Huang, Pengcheng Yu, Qianqian Wang, Weijia Zeng

Pith reviewed 2026-05-10 17:28 UTC · model grok-4.3

classification 💻 cs.CV
keywords self-distillation4D perceptionmulti-view reconstructionvideo depth estimationcamera estimationunlabeled videosself-improvementdynamic scenes
0
0 comments X

The pith

A self-distillation scheme with spatiotemporal asymmetry lets 4D perception models improve themselves on unlabeled videos

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

The paper introduces SelfEvo, a framework that continually refines pretrained multi-view reconstruction models using only unlabeled videos. It does this through a self-distillation process that creates asymmetry in how spatiotemporal context is presented to the model, generating its own training signal. This matters because ground-truth 3D and 4D labels are expensive to obtain and especially scarce for dynamic scenes, which currently limits scalability of learning-based methods. The approach is shown to deliver consistent gains across eight benchmarks and multiple base models, with the largest benefits appearing on moving scenes.

Core claim

SelfEvo enables self-improvement of 4D perception by using a self-distillation scheme based on spatiotemporal context asymmetry to supply a stable supervisory signal from unlabeled videos, producing measurable gains in video depth estimation and camera pose estimation without any external 3D or 4D annotations.

What carries the argument

Spatiotemporal context asymmetry in self-distillation, which presents the model with differing temporal and spatial views of the same unlabeled video sequence to create a non-collapsing training target.

If this is right

  • Video depth estimation improves by up to 36.5 percent relative to the starting model.
  • Camera pose estimation improves by up to 20.1 percent relative to the starting model.
  • Gains appear across eight diverse benchmarks and hold when the method is applied to different base models such as VGGT and π^{3}.
  • The largest benefits occur on dynamic scenes where motion makes reconstruction harder.

Where Pith is reading between the lines

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

  • Large unlabeled video datasets become more valuable for advancing 4D perception if the self-improvement loop holds.
  • The same asymmetry principle could be tested on other perception tasks that currently require expensive 3D labels.

Load-bearing premise

The asymmetry in spatiotemporal context during self-distillation supplies a stable supervisory signal that drives genuine improvement on unlabeled dynamic videos instead of collapse or simple preservation of the original performance.

What would settle it

Applying the full SelfEvo training loop to a fresh collection of unlabeled dynamic videos and measuring no gain, or an actual drop, in video depth and camera estimation accuracy relative to the original pretrained model.

Figures

Figures reproduced from arXiv: 2604.08532 by Angjoo Kanazawa, Haiwen Feng, James M. Rehg, Nan Huang, Pengcheng Yu, Qianqian Wang, Weijia Zeng.

Figure 1
Figure 1. Figure 1: We present SelfEvo, a self-improving framework for learning-based multi-view reconstruction via self-distillation, requiring no ground-truth annotations. Abstract. Large-scale multi-view reconstruction models have made re￾markable progress, but most existing approaches still rely on fully super￾vised training with ground-truth 3D/4D annotations. Such annotations are expensive and particularly scarce for dy… view at source ↗
Figure 2
Figure 2. Figure 2: We propose an annotation-free self-improving framework that continually post￾trains pretrained multi-view reconstruction models using unlabeled videos. Our method forms an online self-distillation loop where a richer-context teacher provides stop￾gradient pseudo targets to a student operating on reduced context, and is updated as an EMA of the student after each step. richer spatiotemporal context are typi… view at source ↗
Figure 3
Figure 3. Figure 3: Visual result on unseen-domain data, including animal motion, robotics and ego-centric. 4.4 Unseen-domain Generalization Our previous evaluation measures in-distribution adaptation and original-domain re￾tention, but it does not directly answer whether the gains obtained from self-improvement transfer beyond the adaptation domain. We therefore conduct an additional unseen￾domain generalization study under … view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results for camera and geometry on in the wild videos [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Context improves feedforward reconstruction quality. [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Checkpoint-wise depth evaluation on the held-out OmniGeo bench [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Checkpoint-wise depth evaluation on the held-out OmniVideo bench [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Checkpoint-wise camera evaluation on held-out OmniGeo and Om [PITH_FULL_IMAGE:figures/full_fig_p025_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results on additional training sources. [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
read the original abstract

Large-scale multi-view reconstruction models have made remarkable progress, but most existing approaches still rely on fully supervised training with ground-truth 3D/4D annotations. Such annotations are expensive and particularly scarce for dynamic scenes, limiting scalability. We propose SelfEvo, a self-improving framework that continually improves pretrained multi-view reconstruction models using unlabeled videos. SelfEvo introduces a self-distillation scheme using spatiotemporal context asymmetry, enabling self-improvement for learning-based 4D perception without external annotations. We systematically study design choices that make self-improvement effective, including loss signals, forms of asymmetry, and other training strategies. Across eight benchmarks spanning diverse datasets and domains, SelfEvo consistently improves pretrained baselines and generalizes across base models (e.g. VGGT and $\pi^3$), with significant gains on dynamic scenes. Overall, SelfEvo achieves up to 36.5% relative improvement in video depth estimation and 20.1% in camera estimation, without using any labeled data. Project Page: https://self-evo.github.io/.

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 / 3 minor

Summary. The manuscript introduces SelfEvo, a self-improving framework for pretrained multi-view 4D reconstruction models. It employs a self-distillation scheme based on spatiotemporal context asymmetry to enable continual improvement on unlabeled dynamic videos, without any external 3D/4D annotations. The authors systematically examine design choices (loss signals, asymmetry forms, training strategies) and report consistent gains across eight benchmarks on two base models (VGGT and π³), with relative improvements reaching 36.5% on video depth estimation and 20.1% on camera estimation, particularly on dynamic scenes.

Significance. If the empirical results hold under rigorous verification, the work has clear significance for scalable 4D perception: it directly tackles the annotation bottleneck for dynamic scenes by turning abundant unlabeled video into a self-improvement signal. The reported generalization across base models and the explicit design study that avoids collapse are strengths that could influence future self-supervised 4D pipelines.

major comments (2)
  1. [Experiments] Experiments section: the central claim of consistent, non-trivial self-improvement rests on the quantitative gains (36.5 % depth, 20.1 % camera). However, the reported numbers lack accompanying error bars, statistical significance tests, and per-scene breakdowns that would confirm the gains are robust rather than driven by a few easy sequences or post-hoc metric selection.
  2. [§3] §3 (Method), spatiotemporal asymmetry definition: the claim that the asymmetry supplies a strictly more informative signal than the model's own predictions on dynamic regions is load-bearing. The paper should explicitly demonstrate (via an additional controlled ablation) that removing the asymmetry collapses performance to the pretrained baseline or worse, rather than merely preserving it.
minor comments (3)
  1. [Abstract / §1] The abstract and introduction repeatedly cite “up to 36.5 %” and “20.1 %” relative improvement; these should be accompanied by the absolute metric values and the exact baseline numbers in the main text for immediate interpretability.
  2. [Figure 2] Figure 2 (framework diagram) would benefit from explicit arrows or labels indicating which frames are used as context versus target in the asymmetry construction.
  3. [§2] Related-work section should add a short paragraph contrasting SelfEvo with recent self-distillation methods in monocular depth and video reconstruction to clarify the novelty of the spatiotemporal asymmetry.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation, the recommendation of minor revision, and the constructive comments on empirical robustness and methodological validation. We address each major comment point by point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the central claim of consistent, non-trivial self-improvement rests on the quantitative gains (36.5 % depth, 20.1 % camera). However, the reported numbers lack accompanying error bars, statistical significance tests, and per-scene breakdowns that would confirm the gains are robust rather than driven by a few easy sequences or post-hoc metric selection.

    Authors: We agree that error bars, significance testing, and per-scene analysis would strengthen the presentation of robustness. In the revised version we will report standard deviations computed over multiple independent runs with different random seeds, include statistical significance tests (e.g., paired t-tests) against the pretrained baselines, and add per-scene breakdowns for the primary benchmarks in the supplementary material. These additions will demonstrate that the reported relative gains are consistent rather than driven by outlier sequences. revision: yes

  2. Referee: [§3] §3 (Method), spatiotemporal asymmetry definition: the claim that the asymmetry supplies a strictly more informative signal than the model's own predictions on dynamic regions is load-bearing. The paper should explicitly demonstrate (via an additional controlled ablation) that removing the asymmetry collapses performance to the pretrained baseline or worse, rather than merely preserving it.

    Authors: We appreciate this request for a more direct demonstration. Our existing ablations already compare the full SelfEvo pipeline against a symmetric (no-asymmetry) variant and show that the latter yields no improvement. To address the referee's point explicitly, we will add a controlled ablation in the revised §3 and supplementary material that isolates the removal of spatiotemporal asymmetry while keeping all other components fixed; the results will show that performance reverts to (or falls below) the pretrained baseline, confirming that the asymmetry is necessary to generate the self-improvement signal on dynamic regions. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes an empirical self-distillation framework for improving pretrained 4D perception models on unlabeled videos, with gains measured on external benchmarks. No equations, derivations, or first-principles predictions are presented that reduce by construction to the method's own inputs or fitted parameters. The self-distillation signal is defined operationally via spatiotemporal asymmetry and evaluated through ablation studies and cross-model generalization, remaining independent of the reported performance metrics. No self-citation chains or uniqueness theorems are invoked as load-bearing justifications for the core claims.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no technical details on parameters, axioms, or new entities are provided.

pith-pipeline@v0.9.0 · 5502 in / 1198 out tokens · 73027 ms · 2026-05-10T17:28:07.893811+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

82 extracted references · 20 canonical work pages · 8 internal anchors

  1. [1]

    Communications of the ACM54(10), 105–112 (2011) 3

    Agarwal, S., Furukawa, Y., Snavely, N., Simon, I., Curless, B., Seitz, S.M., Szeliski, R.: Building rome in a day. Communications of the ACM54(10), 105–112 (2011) 3

    2011

  2. [2]

    AI, B.: Egocentric-10k (2025),https://huggingface.co/datasets/builddotai/ Egocentric-10K1, 7

    2025

  3. [3]

    In: Euro- pean conference on computer vision

    Bay, H., Tuytelaars, T., Van Gool, L.: Surf: Speeded up robust features. In: Euro- pean conference on computer vision. pp. 404–417. Springer (2006) 3

    2006

  4. [4]

    Advances in neural information processing systems32(2019) 3

    Bian, J., Li, Z., Wang, N., Zhan, H., Shen, C., Cheng, M.M., Reid, I.: Unsupervised scale-consistent depth and ego-motion learning from monocular video. Advances in neural information processing systems32(2019) 3

    2019

  5. [5]

    Bozic, A., Palafox, P., Thies, J., Dai, A., Niessner, M.: Transformerfusion: Monoc- ular rgb scene reconstruction using transformers. Proc. Neural Information Pro- cessing Systems (NeurIPS) (2021) 7, 8

    2021

  6. [6]

    In: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining

    Buciluˇ a, C., Caruana, R., Niculescu-Mizil, A.: Model compression. In: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining. pp. 535–541 (2006) 4 SelfEvo: Self-Improving 4D Perception via Self-Distillation 15

    2006

  7. [7]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Caron, M., Touvron, H., Misra, I., Jégou, H., Mairal, J., Bojanowski, P., Joulin, A.: Emerging properties in self-supervised vision transformers. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 9650–9660 (2021) 4

    2021

  8. [8]

    In: arXiv (2020) 3

    Chan, E., Monteiro, M., Kellnhofer, P., Wu, J., Wetzstein, G.: pi-gan: Periodic implicit generative adversarial networks for 3d-aware image synthesis. In: arXiv (2020) 3

    2020

  9. [9]

    Self-supervised flow matching for scalable multi-modal synthesis, 2026

    Chefer, H., Esser, P., Lorenz, D., Podell, D., Raja, V., Tong, V., Torralba, A., Rombach, R.: Self-supervised flow matching for scalable multi-modal synthesis. arXiv preprint arXiv:2603.06507 (2026) 4

    work page arXiv 2026

  10. [10]

    Advances in neural information pro- cessing systems33, 22243–22255 (2020) 4

    Chen, T., Kornblith, S., Swersky, K., Norouzi, M., Hinton, G.E.: Big self-supervised models are strong semi-supervised learners. Advances in neural information pro- cessing systems33, 22243–22255 (2020) 4

    2020

  11. [11]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Chen, Z., Qin, M., Yuan, T., Liu, Z., Zhao, H.: Long3r: Long sequence streaming 3d reconstruction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 5273–5284 (2025) 3

    2025

  12. [12]

    In: Proc

    Dai,A.,Chang,A.X.,Savva,M.,Halber,M.,Funkhouser,T.,Nießner,M.:Scannet: Richly-annotated 3d reconstructions of indoor scenes. In: Proc. Computer Vision and Pattern Recognition (CVPR), IEEE (2017) 1

    2017

  13. [13]

    arXiv preprint arXiv:2512.08930 (2025) 4

    Deng, Y., Peng, S., Zhang, J., Heal, K., Sun, T., Flynn, J., Marschner, S., Chai, L.: Selfi: Self improving reconstruction engine via 3d geometric feature alignment. arXiv preprint arXiv:2512.08930 (2025) 4

    work page arXiv 2025

  14. [14]

    In: Proceedings of the IEEE conference on com- puter vision and pattern recognition workshops

    DeTone, D., Malisiewicz, T., Rabinovich, A.: Superpoint: Self-supervised interest point detection and description. In: Proceedings of the IEEE conference on com- puter vision and pattern recognition workshops. pp. 224–236 (2018) 3

    2018

  15. [15]

    In: Proceedings of the Asian Conference on Computer Vision

    Doersch, C., Luc, P., Yang, Y., Gokay, D., Koppula, S., Gupta, A., Heyward, J., Rocco, I., Goroshin, R., Carreira, J., et al.: Bootstap: Bootstrapped training for tracking-any-point. In: Proceedings of the Asian Conference on Computer Vision. pp. 3257–3274 (2024) 4

    2024

  16. [16]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A.: An image is worth 16x16 words: Transformers for image recogni- tion at scale. arXiv preprint arXiv:2010.11929 (2020) 3

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  17. [17]

    Duisterhof, B., Zust, L., Weinzaepfel, P., Leroy, V., Cabon, Y., Revaud, J.: Mast3r- sfm: a fully-integrated solution for unconstrained structure-from-motion (2024), https://arxiv.org/abs/2409.191523

    work page arXiv 2024

  18. [18]

    In: Proceedings of the ieee/cvf conference on computer vision and pattern recognition

    Dusmanu, M., Rocco, I., Pajdla, T., Pollefeys, M., Sivic, J., Torii, A., Sattler, T.: D2-net: A trainable cnn for joint description and detection of local features. In: Proceedings of the ieee/cvf conference on computer vision and pattern recognition. pp. 8092–8101 (2019) 3

    2019

  19. [19]

    arXiv preprint arXiv:2210.07181 (2022) 3

    Fu, Y., Misra, I., Wang, X.: Mononerf: Learning generalizable nerfs from monocular videos without camera poses. arXiv preprint arXiv:2210.07181 (2022) 3

    work page arXiv 2022

  20. [20]

    Foundations and trends®in Computer Graphics and Vision9(1-2), 1–148 (2015) 3

    Furukawa, Y., Hernández, C., et al.: Multi-view stereo: A tutorial. Foundations and trends®in Computer Graphics and Vision9(1-2), 1–148 (2015) 3

    2015

  21. [21]

    The International Journal of Robotics Research32(11), 1231–1237 (2013) 7, 8

    Geiger, A., Lenz, P., Stiller, C., Urtasun, R.: Vision meets robotics: The kitti dataset. The International Journal of Robotics Research32(11), 1231–1237 (2013) 7, 8

    2013

  22. [22]

    In: Proceedings of the IEEE/CVF in- ternational conference on computer vision

    Godard, C., Mac Aodha, O., Firman, M., Brostow, G.J.: Digging into self- supervised monocular depth estimation. In: Proceedings of the IEEE/CVF in- ternational conference on computer vision. pp. 3828–3838 (2019) 3

    2019

  23. [23]

    arXiv preprint arXiv:2512.04012 (2025) 3

    Han, J., Hong, S., Jung, J., Jang, W., An, H., Wang, Q., Kim, S., Feng, C.: Emergent outlier view rejection in visual geometry grounded transformers. arXiv preprint arXiv:2512.04012 (2025) 11 16 N. Huang et al

    work page arXiv 2025

  24. [24]

    Hartley, R.: Multiple view geometry in computer vision, vol. 665. Cambridge uni- versity press (2003) 3

    2003

  25. [25]

    Distilling the Knowledge in a Neural Network

    Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015) 4

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  26. [26]

    Rayzer: A Self- supervised Large View Synthesis Model.arXiv preprint arXiv:2505.00702, 2025

    Jiang, H., Tan, H., Wang, P., Jin, H., Zhao, Y., Bi, S., Zhang, K., Luan, F., Sunkavalli, K., Huang, Q., et al.: Rayzer: A self-supervised large view synthesis model. arXiv preprint arXiv:2505.00702 (2025) 3

    work page arXiv 2025

  27. [27]

    In: ECCV (2018) 3

    Kanazawa, A., Tulsiani, S., Efros, A.A., Malik, J.: Learning category-specific mesh reconstruction from image collections. In: ECCV (2018) 3

    2018

  28. [28]

    Keetha, N., Müller, N., Schönberger, J., Porzi, L., Zhang, Y., Fischer, T., Knapitsch, A., Zauss, D., Weber, E., Antunes, N., Luiten, J., Lopez-Antequera, M., Bulò, S.R., Richardt, C., Ramanan, D., Scherer, S., Kontschieder, P.: MapA- nything: Universal feed-forward metric 3D reconstruction (2025), arXiv preprint arXiv:2509.13414 1

    work page internal anchor Pith review arXiv 2025

  29. [29]

    Khazatsky, A., Pertsch, K., Nair, S., Balakrishna, A., Dasari, S., Karamcheti, S., Nasiriany, S., Srirama, M.K., Chen, L.Y., Ellis, K., Fagan, P.D., Hejna, J., Itkina, M., Lepert, M., Ma, Y.J., Miller, P.T., Wu, J., Belkhale, S., Dass, S., Ha, H., Jain, A., Lee, A., Lee, Y., Memmel, M., Park, S., Radosavovic, I., Wang, K., Zhan, A., Black, K., Chi, C., Ha...

    2024

  30. [30]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Lai, Z., Liu, S., Efros, A.A., Wang, X.: Video autoencoder: self-supervised disen- tanglement of static 3d structure and motion. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 9730–9740 (2021) 3

    2021

  31. [31]

    In: European Conference on Computer Vision

    Leroy, V., Cabon, Y., Revaud, J.: Grounding image matching in 3d with mast3r. In: European Conference on Computer Vision. pp. 71–91. Springer (2024) 3

    2024

  32. [32]

    In: Advances in Neural Information Processing Systems (NeurIPS) (2020) 3

    Lin, C.H., Wang, C., Lucey, S.: Sdf-srn: Learning signed distance 3d object re- construction from static images. In: Advances in Neural Information Processing Systems (NeurIPS) (2020) 3

    2020

  33. [33]

    Depth Anything 3: Recovering the Visual Space from Any Views

    Lin, H., Chen, S., Liew, J.H., Chen, D.Y., Li, Z., Shi, G., Feng, J., Kang, B.: Depth anything 3: Recovering the visual space from any views. arXiv preprint arXiv:2511.10647 (2025) 2, 3

    work page internal anchor Pith review arXiv 2025

  34. [34]

    In: Proceedings of the IEEE/CVF international conference on com- puter vision

    Lindenberger, P., Sarlin, P.E., Pollefeys, M.: Lightglue: Local feature matching at light speed. In: Proceedings of the IEEE/CVF international conference on com- puter vision. pp. 17627–17638 (2023) 3

    2023

  35. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Liu, Y., Liu, Y., Jiang, C., Lyu, K., Wan, W., Shen, H., Liang, B., Fu, Z., Wang, H., Yi, L.: Hoi4d: A 4d egocentric dataset for category-level human-object interaction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 21013–21022 (June 2022) 3, 9, 10

    2022

  36. [36]

    In: Proceedings of the seventh IEEE international conference on computer vision

    Lowe, D.G.: Object recognition from local scale-invariant features. In: Proceedings of the seventh IEEE international conference on computer vision. vol. 2, pp. 1150–

  37. [37]

    Ieee (1999) 3 SelfEvo: Self-Improving 4D Perception via Self-Distillation 17

    1999

  38. [38]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Lu, J., Huang, T., Li, P., Dou, Z., Lin, C., Cui, Z., Dong, Z., Yeung, S.K., Wang, W., Liu, Y.: Align3r: Aligned monocular depth estimation for dynamic videos. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 22820–22830 (2025) 3

    2025

  39. [39]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Mahjourian, R., Wicke, M., Angelova, A.: Unsupervised learning of depth and ego- motion from monocular video using 3d geometric constraints. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 5667–5675 (2018) 3

    2018

  40. [40]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Murai, R., Dexheimer, E., Davison, A.J.: Mast3r-slam: Real-time dense slam with 3d reconstruction priors. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 16695–16705 (2025) 3

    2025

  41. [41]

    In: IEEE Conf

    Mustikovela, S.K., Jampani, V., De Mello, S., Liu, S., Iqbal, U., Rother, C., Kautz, J.: Self-supervised viewpoint learning from image collections. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) (June 2020) 3

    2020

  42. [42]

    In: The IEEE Interna- tional Conference on Computer Vision (ICCV) (Nov 2019) 3

    Nguyen-Phuoc, T., Li, C., Theis, L., Richardt, C., Yang, Y.L.: Hologan: Unsuper- vised learning of 3d representations from natural images. In: The IEEE Interna- tional Conference on Computer Vision (ICCV) (Nov 2019) 3

    2019

  43. [43]

    DINOv2: Learning Robust Visual Features without Supervision

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023) 4, 10

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Palazzolo, E., Behley, J., Lottes, P., Giguere, P., Stachniss, C.: Refusion: 3d recon- struction in dynamic environments for rgb-d cameras exploiting residuals. In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 7855–7862. IEEE (2019) 7, 8, 12

    2019

  45. [45]

    In: CVPR (2025) 1, 7, 8

    Rockwell, C., Tung, J., Lin, T.Y., Liu, M.Y., Fouhey, D.F., Lin, C.H.: Dynamic camera poses and where to find them. In: CVPR (2025) 1, 7, 8

    2025

  46. [46]

    CVPR (2023) 3

    Sajjadi,M.S.M.,Mahendran,A.,Kipf,T.,Pot,E.,Duckworth,D.,Lučić,M.,Greff, K.: RUST: Latent Neural Scene Representations from Unposed Imagery. CVPR (2023) 3

    2023

  47. [47]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Sarlin, P.E., DeTone, D., Malisiewicz, T., Rabinovich, A.: Superglue: Learning feature matching with graph neural networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4938–4947 (2020) 3

    2020

  48. [48]

    In: European conference on computer vision

    Schönberger, J.L., Zheng, E., Frahm, J.M., Pollefeys, M.: Pixelwise view selection for unstructured multi-view stereo. In: European conference on computer vision. pp. 501–518. Springer (2016) 3

    2016

  49. [49]

    In: Conference on Computer Vision and Pattern Recognition (CVPR) (2016) 3

    Schönberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Conference on Computer Vision and Pattern Recognition (CVPR) (2016) 3

    2016

  50. [50]

    In: European Conference on Computer Vision (ECCV) (2016) 3

    Schönberger, J.L., Zheng, E., Pollefeys, M., Frahm, J.M.: Pixelwise view selection for unstructured multi-view stereo. In: European Conference on Computer Vision (ECCV) (2016) 3

    2016

  51. [51]

    Advances in neural information processing systems 33, 20154–20166 (2020) 3

    Schwarz, K., Liao, Y., Niemeyer, M., Geiger, A.: Graf: Generative radiance fields for 3d-aware image synthesis. Advances in neural information processing systems 33, 20154–20166 (2020) 3

    2020

  52. [52]

    Self-Distillation Enables Continual Learning

    Shenfeld, I., Damani, M., Hübotter, J., Agrawal, P.: Self-distillation enables con- tinual learning. arXiv preprint arXiv:2601.19897 (2026) 4

    work page internal anchor Pith review arXiv 2026

  53. [53]

    In: ACM siggraph 2006 papers, pp

    Snavely, N., Seitz, S.M., Szeliski, R.: Photo tourism: exploring photo collections in 3d. In: ACM siggraph 2006 papers, pp. 835–846 (2006) 3

    2006

  54. [54]

    Expanding the capabilities of reinforcement learning via text feedback.arXiv preprint arXiv:2602.02482, 2026

    Song, Y., Chen, L., Tajwar, F., Munos, R., Pathak, D., Bagnell, J.A., Singh, A., Zanette, A.: Expanding the capabilities of reinforcement learning via text feedback. arXiv preprint arXiv:2602.02482 (2026) 4 18 N. Huang et al

    work page arXiv 2026

  55. [55]

    Mv-dust3r+: Single-stage scene reconstruction from sparse views in 2 seconds

    Tang,Z.,Fan,Y.,Wang,D.,Xu,H.,Ranjan,R.,Schwing,A.,Yan,Z.:Mv-dust3r+: Single-stage scene reconstruction from sparse views in 2 seconds. arXiv preprint arXiv:2412.06974 (2024) 3

    work page arXiv 2024

  56. [56]

    Advances in neural information processing systems30(2017) 4

    Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems30(2017) 4

    2017

  57. [57]

    In: The Thirty- ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track (2025) 2, 6

    Tesch, J., Becherini, G., Achar, P., Yiannakidis, A., Kocabas, M., Patel, P., Black, M.J.: BEDLAM2.0: Synthetic humans and cameras in motion. In: The Thirty- ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track (2025) 2, 6

    2025

  58. [58]

    In: International conference on machine learning

    Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., Jégou, H.: Training data-efficient image transformers & distillation through attention. In: International conference on machine learning. pp. 10347–10357. PMLR (2021) 4

    2021

  59. [59]

    Advances in neural information pro- cessing systems30(2017) 3

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. Advances in neural information pro- cessing systems30(2017) 3

    2017

  60. [60]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wang, J., Chen, M., Karaev, N., Vedaldi, A., Rupprecht, C., Novotny, D.: Vggt: Visual geometry grounded transformer. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5294–5306 (2025) 2, 3, 4, 6, 7, 8, 10

    2025

  61. [61]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wang, Q., Zhang, Y., Holynski, A., Efros, A.A., Kanazawa, A.: Continuous 3d perception model with persistent state. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 10510–10522 (2025) 3, 7

    2025

  62. [62]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wang, S., Leroy, V., Cabon, Y., Chidlovskii, B., Revaud, J.: Dust3r: Geometric 3d vision made easy. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20697–20709 (2024) 2, 3, 7

    2024

  63. [63]

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

    Wang, Y., Zhou, J., Zhu, H., Chang, W., Zhou, Y., Li, Z., Chen, J., Pang, J., Shen, C., He, T.: pi-3: Permutation-equivariant visual geometry learning. arXiv preprint arXiv:2507.13347 (2025) 2, 3, 4, 6, 7, 8

    work page internal anchor Pith review arXiv 2025

  64. [64]

    In: ICCV (2023) 3

    Weinzaepfel, P., Lucas, T., Leroy, V., Cabon, Y., Arora, V., Brégier, R., Csurka, G., Antsfeld, L., Chidlovskii, B., Revaud, J.: CroCo v2: Improved Cross-view Com- pletion Pre-training for Stereo Matching and Optical Flow. In: ICCV (2023) 3

    2023

  65. [65]

    In: NeurIPS (2022) 3

    Weinzaepfel, Philippe and Leroy, Vincent and Lucas, Thomas and Brégier, Romain and Cabon, Yohann and Arora, Vaibhav and Antsfeld, Leonid and Chidlovskii, Boris and Csurka, Gabriela and Revaud Jérôme: CroCo: Self-Supervised Pre- training for 3D Vision Tasks by Cross-View Completion. In: NeurIPS (2022) 3

    2022

  66. [66]

    In: 2013 Interna- tional Conference on 3D Vision-3DV 2013

    Wu, C.: Towards linear-time incremental structure from motion. In: 2013 Interna- tional Conference on 3D Vision-3DV 2013. pp. 127–134. IEEE (2013) 3

    2013

  67. [67]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Xie, Q., Luong, M.T., Hovy, E., Le, Q.V.: Self-training with noisy student improves imagenet classification. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10687–10698 (2020) 4

    2020

  68. [68]

    Advances in neural information processing systems29(2016) 3

    Yan, X., Yang, J., Yumer, E., Guo, Y., Lee, H.: Perspective transformer nets: Learning single-view 3d object reconstruction without 3d supervision. Advances in neural information processing systems29(2016) 3

    2016

  69. [69]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Yang, J., Sax, A., Liang, K.J., Henaff, M., Tang, H., Cao, A., Chai, J., Meier, F., Feiszli, M.: Fast3r: Towards 3d reconstruction of 1000+ images in one forward pass. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 21924–21935 (2025) 3

    2025

  70. [70]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yang, L., Kang, B., Huang, Z., Xu, X., Feng, J., Zhao, H.: Depth anything: Un- leashing the power of large-scale unlabeled data. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10371–10381 (2024) 4 SelfEvo: Self-Improving 4D Perception via Self-Distillation 19

    2024

  71. [71]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yang,N.,Stumberg,L.v.,Wang,R.,Cremers,D.:D3vo:Deepdepth,deepposeand deep uncertainty for monocular visual odometry. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1281–1292 (2020) 3

    2020

  72. [72]

    In: Proceedings of the European conference on computer vision (ECCV)

    Yao, Y., Luo, Z., Li, S., Fang, T., Quan, L.: Mvsnet: Depth inference for unstruc- tured multi-view stereo. In: Proceedings of the European conference on computer vision (ECCV). pp. 767–783 (2018) 3

    2018

  73. [73]

    In: European conference on computer vision

    Yi, K.M., Trulls, E., Lepetit, V., Fua, P.: Lift: Learned invariant feature transform. In: European conference on computer vision. pp. 467–483. Springer (2016) 3

    2016

  74. [74]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Yun, S., Han, D., Oh, S.J., Chun, S., Choe, J., Yoo, Y.: Cutmix: Regularization strategy to train strong classifiers with localizable features. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6023–6032 (2019) 4

    2019

  75. [75]

    arXiv preprint arXiv:2410.03825 (2024)

    Zhang, J., Herrmann, C., Hur, J., Jampani, V., Darrell, T., Cole, F., Sun, D., Yang, M.H.: Monst3r: A simple approach for estimating geometry in the presence of motion. arXiv preprint arXiv:2410.03825 (2024) 3, 7

    work page arXiv 2024

  76. [76]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Song, J., Gao, A., Chen, J., Bao, C., Ma, K.: Be your own teacher: Improve the performance of convolutional neural networks via self distillation. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3713–3722 (2019) 4

    2019

  77. [77]

    Embarrassingly simple self-distillation improves code generation.arXiv preprint arXiv:2604.01193, 2026a

    Zhang, R., Bai, R.H., Zheng, H., Jaitly, N., Collobert, R., Zhang, Y.: Em- barrassingly simple self-distillation improves code generation. arXiv preprint arXiv:2604.01193 (2026) 4

    work page arXiv 2026

  78. [78]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Zhang, S., Wang, J., Xu, Y., Xue, N., Rupprecht, C., Zhou, X., Shen, Y., Wet- zstein, G.: Flare: Feed-forward geometry, appearance and camera estimation from uncalibrated sparse views. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 21936–21947 (2025) 3

    2025

  79. [79]

    arXiv preprint arXiv:2512.10950 (2025) 3

    Zhao, Q., Tan, H., Wang, Q., Bi, S., Zhang, K., Sunkavalli, K., Tulsiani, S., Jiang, H.: E-rayzer: Self-supervised 3d reconstruction as spatial visual pre-training. arXiv preprint arXiv:2512.10950 (2025) 3

    work page arXiv 2025

  80. [80]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zhou, T., Brown, M., Snavely, N., Lowe, D.G.: Unsupervised learning of depth and ego-motion from video. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1851–1858 (2017) 3

    2017

Showing first 80 references.