Pith. sign in

REVIEW 3 major objections 5 minor 93 references

2K Retrofit: Entropy-Guided Efficient Sparse Refinement for High-Resolution 3D Geometry Prediction

T0 review · 3 major / 5 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read Any frozen geometric foundation model can produce accurate 2K depth and point maps by sparsely refining only high-uncertainty pixels, without backbone changes or retraining.

desk verdict Practical sparse retrofit that makes frozen 3D foundation models usable at 2K; the entropy-proxy assumption is real but not fatal, and the empirical package is strong enough to take seriously. read the letter →

arxiv 2603.19964 v3 pith:MVGN2SM3 submitted 2026-03-20 cs.CV

Tianbao Zhang , Zhenyu Liang , Zhenbo Song , Nana Wang , Xiaomei Zhang , Xudong Cai , Zheng Zhu , Kejian Wu
show 2 more authors
Gang Wang Zhaoxin Fan
This is my paper · ORCID
classification cs.CV
keywords highresolutionsparserefinementfoundationmodelmonoculardepthestimationmulti-viewreconstructionentropy-guidedselection3Dgeometryprediction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

High-resolution depth and 3D geometry are needed for driving, robots, and AR, but foundation models become impractical at 2K because dense inference is too slow and memory-heavy. This paper claims the gap between a cheap low-resolution prediction and a true high-resolution result is concentrated in a sparse set of uncertain pixels. Freeze any backbone, run it at low resolution, use prediction entropy to select roughly the top tenth of pixels, and correct only those locations with a lightweight sparse network; a gated fusion then merges local fixes with the coarse map. On ARKitScenes, ScanNet++, and ETH3D the method matches or beats full high-resolution and patch-based baselines while running substantially faster. The result is that existing foundation models become usable at real-world sensor resolution without retraining.

What carries the argument

Entropy-guided sparse refinement: entropy computed on backbone head features before the final regression selects high-uncertainty pixels; a modified MinkowskiUNet sparse extractor predicts local corrections only at those sites; a gated ensembler fuses the residual with the coarse map according to confidence.

What would settle it

On held-out 2K scenes, measure how many high-error pixels the entropy top-10 percent actually covers; if that recall is far below the reported majority, or if random or edge selection matches entropy selection on final AbsRel and point-map error, the sparse-refinement premise fails.

Watch

Extended reading notes

Core claim

2K Retrofit establishes that accurate 2K geometric prediction is achievable from any frozen low-resolution geometric foundation model by pairing a fast coarse prediction with entropy-guided sparse refinement of only high-uncertainty pixels, matching or exceeding full high-resolution and patch-wise methods in fidelity at a fraction of the cost and without modifying or retraining the backbone.

Load-bearing premise

Prediction entropy from the backbone head features is a reliable enough stand-in for geometric error that refining only about the top 10 percent most uncertain pixels recovers most of the high-error locations that matter at 2K.

Editorial extensions

If this is right

  • Existing monocular depth and multi-view pointmap foundation models can be used at 2K resolution without architectural changes or retraining.
  • Compute and memory for high-resolution geometry fall by focusing work on a sparse uncertain subset rather than the full image or many overlapping patches.
  • Fine structures that pure low-resolution upsampling loses (boundaries, thin objects, occlusion edges) can be restored while keeping the backbone's global consistency.
  • The same adapter works across monocular depth and multi-view reconstruction backbones.
  • Foundation-model deployment for driving, robotics, and AR/MR at sensor-native high resolution becomes practical.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same pre-regression entropy signal may serve as a cheap uncertainty cue for other dense regression tasks at high resolution, such as surface normals or optical flow.
  • If error remains sparse at 4K and beyond, the retrofit pattern could scale further without ever training a native full-resolution backbone.
  • When errors become dense (textureless or reflective regions), a hybrid uncertainty score may be needed so the sparse selector does not miss large failure areas.
  • Synthetic 2K geometry sets of the kind built for training the adapter may become a reusable resource for fitting high-resolution correctors to other foundation models.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes 2K Retrofit, a plug-in framework that enables frozen geometric foundation models (e.g., Depth Anything V2, VGGT) to produce 2K-resolution depth or point maps without backbone modification or retraining. A low-resolution coarse prediction is upsampled; an entropy-based selector identifies high-uncertainty pixels from backbone head features; a sparse MinkowskiUNet predicts local residuals; and a gated fusion module combines coarse and refined estimates. The method is trained on a new 50k-image Omniverse synthetic dataset and evaluated on ARKitScenes, ScanNet++, and ETH3D for monocular depth and multi-view pointmap estimation, reporting accuracy and efficiency gains over patch-based and upsampling baselines.

Significance. If the claims hold, the work offers a practical route to high-resolution geometric inference that preserves existing foundation-model weights and substantially reduces memory and latency relative to native 2K training or dense patch refinement. The modular design (entropy selector + sparse residual + gated fusion), the public-benchmark results under both zero-shot and fine-tuned settings, and the planned code release are concrete strengths. The contribution is primarily engineering and empirical rather than theoretical, but it addresses a clear deployment bottleneck in autonomous driving, robotics, and AR/MR.

major comments (3)
  1. §3.3 (i) and Fig. 4: The central efficiency claim rests on entropy H(p) computed from softmax-normalized backbone head features (before the final regression projection) being a reliable proxy for geometric error, so that refining only the top ~10% of pixels (α≈0.3) recovers ~80% of high-error locations. The manuscript never justifies treating intermediate regression features as a C-way categorical distribution, nor does it report high-error recall of the selector on real benchmarks or across backbones (Depth Anything, VGGT, Pi3, Depth Anything v3). Table 5a compares selectors only by final Acc/Comp/FPS; without a direct error-recall measurement under domain shift, the “any frozen backbone / no retraining” claim remains under-supported.
  2. Tables 1–2 and §4.1: All quantitative results are single-run point estimates with no error bars, multi-seed statistics, or variance across the free parameters (α, low-resolution input size). Given that α is a free threshold that trades accuracy for FPS (Table 5c) and that the synthetic training set is newly constructed, the reported SOTA margins (e.g., AbsRel reductions of ~30% vs. PRO) cannot be assessed for statistical reliability. At minimum, multi-seed means and standard deviations on ETH3D and one indoor benchmark are needed to substantiate the accuracy–efficiency frontier.
  3. §4.3–4.4 and Table 4: The comparison to “VGGT retrained at 2K” shows a large efficiency advantage but a small accuracy gap (Acc 0.935 vs 0.911). The manuscript does not specify the training protocol, data, or number of epochs used for the retrained baseline, nor whether the same synthetic 50k set was used. Without a controlled apples-to-apples protocol, the claim that sparse refinement is competitive with full high-resolution training is difficult to evaluate.
minor comments (5)
  1. Abstract and §1 repeatedly claim “the first framework”; related work already cites SparseRefine, PointRend, and patch-based high-res depth methods. Soften the priority language to “first model-agnostic sparse refinement for 2K geometric foundation models.”
  2. Fig. 1 caption and several places use “2K Retrofitefficiently” / “2K Retrofit (ours)” inconsistently; also “imrpove”, “denotedasI”, and missing spaces after periods appear throughout.
  3. Eq. (2)–(5): notation mixes Y, Geo, and Ĝeo; C is used both for output channels and for the number of softmax bins. Clarify whether C for entropy equals the regression output dimension or is an auxiliary head dimension.
  4. §4.1: the synthetic dataset is central yet only briefly described; a short appendix table of scene categories, depth range, and camera models would aid reproducibility.
  5. Table 1 header “Zero Shot Method” is ambiguous; the “No/Yes” column is clearer if labeled “Fine-tuned on target” or similar.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: accuracy claims rest on external benchmarks and a trained refinement module, not on quantities defined by the reported metrics.

full rationale

The paper’s derivation chain is a standard two-stage engineering pipeline: (i) freeze a low-resolution foundation model F and upsample its output to obtain a coarse Ŷ_HR; (ii) select a sparse pixel set P via entropy of intermediate head features, predict residuals with a trained sparse CNN R, and fuse with a gated MLP. None of Eqs. (1)–(5) define the evaluation metrics (AbsRel, RMSE, δ0.5, Acc/Comp) in terms of the method’s own outputs or fitted constants. Training uses a held-out synthetic Omniverse set; reported numbers come from independent public benchmarks (ARKitScenes, ScanNet++, ETH3D) that are not used to construct the entropy threshold or the gated weights. Citations to SparseRefine, PointRend, MinkowskiUNet, Depth Anything, and VGGT are external prior art, not self-authored uniqueness theorems that force the result. The entropy–error correlation (Fig. 4) is an empirical design justification, not a definitional identity that makes the accuracy numbers true by construction. Hyperparameter α and ablations (Table 5) are ordinary model selection, not fitted inputs renamed as predictions. Consequently there is no self-definitional loop, no fitted-input-as-prediction, and no load-bearing self-citation chain.

Assumptions & free parameters 2 free parameters · 3 assumptions · 2 invented entities

The central claim rests on a small set of modeling choices and one free threshold rather than on new physical entities or deep mathematical axioms. The main load-bearing assumptions are that low-resolution foundation-model predictions already capture global geometry well enough to serve as a base, and that entropy of pre-projection features is a cheap, reliable error proxy. Free parameters are limited to the entropy threshold and ordinary training hyper-parameters.

free parameters (2)
  • entropy threshold α = 0.3
    Controls which fraction of pixels receive sparse refinement; set to 0.3 after ablation (Table 5c) and directly affects both accuracy and FPS.
  • low-resolution input size = ~256 px longer side
    Longer side downsampled to ~256 px before the frozen backbone; chosen for speed/accuracy trade-off rather than derived.
assumptions (3)
  • domain assumption Prediction entropy computed from backbone head features (before final regression) is a reliable proxy for geometric error at high resolution.
    Stated and illustrated in §3.3 and Fig. 4; underpins the entire sparse-selection strategy.
  • domain assumption A frozen low-resolution foundation model already supplies globally consistent geometry that only needs local residual correction.
    Core premise of the two-stage pipeline in §3.1–3.2.
  • domain assumption Sparse convolutions (MinkowskiUNet) can extract useful high-resolution features from an irregular set of selected pixels.
    Justifies the choice of sparse feature extractor in §3.3(ii).
invented entities (2)
  • 2K Retrofit pipeline (entropy selector + sparse residual + gated ensembler)
    purpose: Modular add-on that converts any frozen geometric foundation model into a 2K predictor without retraining.
    The composite system is the paper’s main technical product; its components are adaptations of known modules rather than new physical entities.
  • 50k-image Omniverse synthetic 2K geometry dataset
    purpose: Provides high-resolution supervision for training the sparse refinement branch.
    Newly generated training resource; generation details are only sketched, so external verification is currently limited.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 2K Retrofit: Entropy-Guided Efficient Sparse Refinement for High-Resolution 3D Geometry Prediction." pith.science (2026). https://pith.science/paper/MVGN2SM3

@misc{pith2026260319964,
  author       = {Pith},
  title        = {Pith review of: 2K Retrofit: Entropy-Guided Efficient Sparse Refinement for High-Resolution 3D Geometry Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MVGN2SM3}},
  note         = {Machine review of arXiv:2603.19964}
}
read the original abstract

High-resolution geometric prediction is essential for robust perception in autonomous driving, robotics, and AR/MR, but current foundation models are fundamentally limited by their scalability to real-world, high-resolution scenarios. Direct inference on 2K images with these models incurs prohibitive computational and memory demands, making practical deployment challenging. To tackle the issue, we present 2K Retrofit, a novel framework that enables efficient 2K-resolution inference for any geometric foundation model, without modifying or retraining the backbone. Our approach leverages fast coarse predictions and an entropy-based sparse refinement to selectively enhance high-uncertainty regions, achieving precise and high-fidelity 2K outputs with minimal overhead. Extensive experiments on widely used benchmark demonstrate that 2K Retrofit consistently achieves state-of-the-art accuracy and speed, bridging the gap between research advances and scalable deployment in high-resolution 3D vision applications. Code will be released upon acceptance.

Figures

Figures reproduced from arXiv: 2603.19964 by the authors.

Figure 1
Figure 1. 2K Retrofit efficiently enhances high-resolution geometric prediction through an entropy-guided sparse refinement strategy. Given 2K-resolution RGB inputs from diverse scenes (single-object, indoor, and outdoor), our framework first performs fast coarse geometric inference using a frozen low-resolution foundation model, then selec￾tively refines only high-uncertainty pixels at full resolution. This design preserves … view at source ↗
Figure 2
Figure 2. Differences of 2K Retrofit with exising high-resolution geometry [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Framework of 2K Retrofit. Given an HR input image IHR, we first down￾sample it and feed the result to a frozen 3D foundation model, upsampling its output to obtain a coarse Geo ˆ HR. An entropy-based selector then identifies a sparse set of high-uncertainty HR pixels, for which a lightweight sparse extractor predicts localize refinements ∆Geo. A gated ensembler merges the coarse map with ∆Geo, yielding high-fidelity… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Analysis of the correlation between prediction error and entropy-based un￾certainty, which confirms a strong correspondence between the sparse pixels that we selected and the error map. 3.3 Sparse High-Resolution Geometric Refinement While low-resolution predictions de…
Figure 5
Figure 5. Figure 5: Qualitative comparisons of 2K Retrofit with state-of-the-art methods on the ETH3D dataset. The top two rows show results on the depth estimation task, while the bottom two rows present results on the point map estimation task. 2K Retrofit consistently produces sharper …
Figure 6
Figure 6. Figure 6: Generalization and Robustness Analysis. 4.6 Ablation Study We analyze alternative designs for key components of our method, focusing on point map estimation. Detailed breakdowns of accuracy and efficiency improve￾ments are provided below. Impact of Entropy Selectors. W…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

93 extracted references · 21 linked inside Pith

  1. [1]

    Information fusion76, 243–297 (2021)

    Abdar, M., Pourpanah, F., Hussain, S., Rezazadegan, D., Liu, L., Ghavamzadeh, M., Fieguth, P., Cao, X., Khosravi, A., Acharya, U.R., et al.: A review of un- certainty quantification in deep learning: Techniques, applications and challenges. Information fusion76, 243–297 (2021)

    2021

  2. [3]

    arXiv preprint arXiv:2302.12288 (2023)

    Bhat,S.F.,Birkl,R.,Wofk,D.,Wonka,P.,Müller,M.:Zoedepth:Zero-shottransfer by combining relative and metric depth. arXiv preprint arXiv:2302.12288 (2023)

    arXiv 2023

  3. [4]

    IEEE transactions on pattern analysis and machine intelligence44(12), 9802–9813 (2021)

    Bian,J.W.,Zhan,H.,Wang,N.,Chin,T.J.,Shen,C.,Reid,I.:Auto-rectifynetwork for unsupervised indoor depth estimation. IEEE transactions on pattern analysis and machine intelligence44(12), 9802–9813 (2021)

    2021

  4. [5]

    arXiv preprint arXiv:2410.02073 (2024)

    Bochkovskii, A., Delaunoy, A., Germain, H., Santos, M., Zhou, Y., Richter, S.R., Koltun, V.: Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073 (2024)

    arXiv 2024

  5. [6]

    arXiv preprint arXiv:2210.09461 (2022)

    Bolya, D., Fu, C.Y., Dai, X., Zhang, P., Feichtenhofer, C., Hoffman, J.: Token merging: Your vit but faster. arXiv preprint arXiv:2210.09461 (2022)

    arXiv 2022

  6. [7]

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

    Bolya, D., Hoffman, J.: Token merging for fast stable diffusion. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4599– 4603 (2023)

    2023

  7. [8]

    arXiv preprint arXiv:2208.02541 (2022)

    Cao, C., Ren, X., Fu, Y.: Mvsformer: Multi-view stereo by learning robust image features and temperature-based depth. arXiv preprint arXiv:2208.02541 (2022)

    arXiv 2022

  8. [9]

    arXiv preprint arXiv:2401.11673 (2024)

    Cao,C.,Ren,X.,Fu,Y.:Mvsformer++:Revealingthedevilintransformer’sdetails for multi-view stereo. arXiv preprint arXiv:2401.11673 (2024)

    arXiv 2024

Show all 93 references
  1. [10]

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

    Choy, C., Gwak, J., Savarese, S.: 4d spatio-temporal convnets: Minkowski convolu- tional neural networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 3075–3084 (2019)

    2019

  2. [11]

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

    Eigen, D., Fergus, R.: Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture. In: Proceedings of the IEEE in- ternational conference on computer vision. pp. 2650–2658 (2015)

    2015

  3. [12]

    Advances in neural information processing systems27 (2014) 16 F

    Eigen, D., Puhrsch, C., Fergus, R.: Depth map prediction from a single image using a multi-scale deep network. Advances in neural information processing systems27 (2014) 16 F. Author et al

    2014

  4. [13]

    Advances in neural information processing systems27 (2014)

    Eigen, D., Puhrsch, C., Fergus, R.: Depth map prediction from a single image using a multi-scale deep network. Advances in neural information processing systems27 (2014)

    2014

  5. [14]

    arXiv preprint arXiv:2507.16290 (2025)

    Fang, X., Gao, J., Wang, Z., Chen, Z., Ren, X., Lyu, J., Ren, Q., Yang, Z., Yang, X., Yan, Y., et al.: Dens3r: A foundation model for 3d geometry prediction. arXiv preprint arXiv:2507.16290 (2025)

    2025 arXiv

  6. [15]

    Advances in Neural Information Processing Systems35, 3403–3416 (2022)

    Fu, Q., Xu, Q., Ong, Y.S., Tao, W.: Geo-neus: Geometry-consistent neural implicit surfaces learning for multi-view reconstruction. Advances in Neural Information Processing Systems35, 3403–3416 (2022)

    2022

  7. [16]

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

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

    2015

  8. [17]

    In: Proceedings of the IEEE international conference on computer vision

    Galliani, S., Lasinger, K., Schindler, K.: Massively parallel multiview stereopsis by surface normal diffusion. In: Proceedings of the IEEE international conference on computer vision. pp. 873–881 (2015)

    2015

  9. [18]

    arXiv preprint arXiv:2205.03892 (2022)

    Gao, P., Ma, T., Li, H., Lin, Z., Dai, J., Qiao, Y.: Convmae: Masked convolution meets masked autoencoders. arXiv preprint arXiv:2205.03892 (2022)

    2022 arXiv

  10. [19]

    In: Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture

    Gondimalla, A., Chesnut, N., Thottethodi, M., Vijaykumar, T.: Sparten: A sparse tensor accelerator for convolutional neural networks. In: Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture. pp. 151–165 (2019)

    2019

  11. [21]

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

    Gu, X., Fan, Z., Zhu, S., Dai, Z., Tan, F., Tan, P.: Cascade cost volume for high-resolution multi-view stereo and stereo matching. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2495–2504 (2020)

    2020

  12. [22]

    Guo, M.H., Lu, C.Z., Hou, Q., Liu, Z., Cheng, M.M., Hu, S.M.: Segnext: Rethink- ing convolutional attention design for semantic segmentation. arxiv. 2022. arXiv preprint arXiv:2209.08575 (2022)

    2022 arXiv

  13. [23]

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

    Han, S.H., Park, M.G., Yoon, J.H., Kang, J.M., Park, Y.J., Jeon, H.G.: High- fidelity 3d human digitization from single 2k resolution images. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12869–12879 (2023)

    2023

  14. [24]

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

    He, K., Chen, X., Xie, S., Li, Y., Dollár, P., Girshick, R.: Masked autoencoders are scalable vision learners. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 16000–16009 (2022)

    2022

  15. [25]

    Proceedings of Machine Learning and Systems5, 428–441 (2023)

    Hong, K., Yu, Z., Dai, G., Yang, X., Lian, Y., Xu, N., Wang, Y.: Exploiting hard- ware utilization and adaptive dataflow for efficient sparse convolution in 3d point clouds. Proceedings of Machine Learning and Systems5, 428–441 (2023)

    2023

  16. [26]

    Advances in Neural Information Processing Systems35, 19997–20010 (2022)

    Huang, L., You, S., Zheng, M., Wang, F., Qian, C., Yamasaki, T.: Green hierarchi- cal vision transformer for masked image modeling. Advances in Neural Information Processing Systems35, 19997–20010 (2022)

    2022

  17. [27]

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

    Huang, P.H., Matzen, K., Kopf, J., Ahuja, N., Huang, J.B.: Deepmvs: Learning multi-view stereopsis. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2821–2830 (2018)

    2018

  18. [28]

    In: 2019 16th International Con- ference on Machine Vision Applications (MVA)

    Huang, Y.H., Proesmans, M., Georgoulis, S., Van Gool, L.: Uncertainty based model selection for fast semantic segmentation. In: 2019 16th International Con- ference on Machine Vision Applications (MVA). pp. 1–6. IEEE (2019) Abbreviated paper title 17

    2019

  19. [29]

    In: 2019 16th International Con- ference on Machine Vision Applications (MVA)

    Huang, Y.H., Proesmans, M., Georgoulis, S., Van Gool, L.: Uncertainty based model selection for fast semantic segmentation. In: 2019 16th International Con- ference on Machine Vision Applications (MVA). pp. 1–6. IEEE (2019)

    2019

  20. [30]

    In: European conference on computer vision

    Hui, T.W., Loy, C.C., Tang, X.: Depth map super-resolution by deep multi-scale guidance. In: European conference on computer vision. pp. 353–369. Springer (2016)

    2016

  21. [31]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Izquierdo, S., Sayed, M., Firman, M., Garcia-Hernando, G., Turmukhambetov, D., Civera, J., Mac Aodha, O., Brostow, G., Watson, J.: Mvsanywhere: Zero-shot multi-view stereo. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 11493–11504 (2025)

    2025

  22. [32]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Ke, B., Obukhov, A., Huang, S., Metzger, N., Daudt, R.C., Schindler, K.: Re- purposing diffusion-based image generators for monocular depth estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 9492–9502 (2024)

    2024

  23. [33]

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

    Kirillov,A.,Wu,Y.,He,K.,Girshick,R.:Pointrend:Imagesegmentationasrender- ing. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 9799–9808 (2020)

    2020

  24. [34]

    URL https://arxiv

    Kong, Z., Dong, P., Ma, X., Meng, X., Sun, M., Niu, W., Shen, X., Yuan, G., Ren, B., Qin, M., et al.: Spvit: enabling faster vision transformers via soft token pruning (2022). URL https://arxiv. org/abs/2112.13890 (2021)

    2022 arXiv

  25. [35]

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

    Kwon, B., Kim, M.: One look is enough: Seamless patchwise refinement for zero- shot monocular depth estimation on high-resolution images. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 8077–8087 (2025)

    2025

  26. [36]

    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)

    2024

  27. [37]

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

    Li, Z., Snavely, N.: Megadepth: Learning single-view depth prediction from internet photos. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2041–2050 (2018)

  28. [38]

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

    Li, Z., Bhat, S.F., Wonka, P.: Patchfusion: An end-to-end tile-based framework for high-resolution monocular metric depth estimation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10016– 10025 (2024)

    2024

  29. [39]

    In: European Confer- ence on Computer Vision

    Li, Z., Bhat, S.F., Wonka, P.: Patchrefiner: Leveraging synthetic data for real- domain high-resolution monocular metric depth estimation. In: European Confer- ence on Computer Vision. pp. 250–267. Springer (2024)

    2024

  30. [40]

    arXiv preprint arXiv:2509.05296 (2025)

    Li, Z., Zhou, J., Wang, Y., Guo, H., Chang, W., Zhou, Y., Zhu, H., Chen, J., Shen, C., He, T.: Wint3r: Window-based streaming reconstruction with camera token pool. arXiv preprint arXiv:2509.05296 (2025)

    2025 arXiv

  31. [41]

    arXiv preprint arXiv:2202.07800 (2022)

    Liang, Y., Ge, C., Tong, Z., Song, Y., Wang, J., Xie, P.: Not all patches are what you need: Expediting vision transformers via token reorganizations. arXiv preprint arXiv:2202.07800 (2022)

    2022 arXiv

  32. [42]

    arXiv preprint arXiv:2511.10647 (2025)

    Lin, H., Chen, S., Liew, J., 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)

    2025 arXiv

  33. [43]

    In: Proceedings of the Computer Vision and Pattern Recognition Con- ference

    Lin, H., Peng, S., Chen, J., Peng, S., Sun, J., Liu, M., Bao, H., Feng, J., Zhou, X., Kang, B.: Prompting depth anything for 4k resolution accurate metric depth estimation. In: Proceedings of the Computer Vision and Pattern Recognition Con- ference. pp. 17070–17080 (2025)

    2025

  34. [44]

    In: MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture

    Lin, Y., Zhang, Z., Tang, H., Wang, H., Han, S.: Pointacc: Efficient point cloud accelerator. In: MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture. pp. 449–461 (2021) 18 F. Author et al

    2021

  35. [45]

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

    Ling, L., Sheng, Y., Tu, Z., Zhao, W., Xin, C., Wan, K., Yu, L., Guo, Q., Yu, Z., Lu, Y., et al.: Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 22160–22169 (2024)

    2024

  36. [46]

    In: 2021 International conference on 3D vision (3DV)

    Lipson, L., Teed, Z., Deng, J.: Raft-stereo: Multilevel recurrent field transforms for stereo matching. In: 2021 International conference on 3D vision (3DV). pp. 218–227. IEEE (2021)

    2021

  37. [47]

    Advances in neural information processing systems35, 6735–6748 (2022)

    Liu, J., Chen, Y., Ye, X., Tian, Z., Tan, X., Qi, X.: Spatial pruned sparse convo- lution for efficient 3d object detection. Advances in neural information processing systems35, 6735–6748 (2022)

    2022

  38. [48]

    In: European Conference on Computer Vision

    Liu, Z., Zhang, Z., Khaki, S., Yang, S., Tang, H., Xu, C., Keutzer, K., Han, S.: Sparse refinement for efficient high-resolution semantic segmentation. In: European Conference on Computer Vision. pp. 108–127. Springer (2024)

    2024

  39. [49]

    In: Seminal graphics: pioneering efforts that shaped the field, pp

    Lorensen, W.E., Cline, H.E.: Marching cubes: A high resolution 3d surface con- struction algorithm. In: Seminal graphics: pioneering efforts that shaped the field, pp. 347–353 (1998)

    1998

  40. [50]

    arXiv preprint arXiv:1608.03983 (2016)

    Loshchilov, I., Hutter, F.: Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983 (2016)

    2016 arXiv

  41. [51]

    arXiv preprint arXiv:1711.05101 (2017)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    2017 arXiv

  42. [52]

    arXiv preprint arXiv:2303.01494 (2023)

    Ma, X., Zhou, Y., Wang, H., Qin, C., Sun, B., Liu, C., Fu, Y.: Image as set of points. arXiv preprint arXiv:2303.01494 (2023)

    2023 arXiv

  43. [53]

    In: Eu- ropean Conference on Computer Vision

    Ma, Z., Teed, Z., Deng, J.: Multiview stereo with cascaded epipolar raft. In: Eu- ropean Conference on Computer Vision. pp. 734–750. Springer (2022)

    2022

  44. [54]

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

    Metzger, N., Daudt, R.C., Schindler, K.: Guided depth super-resolution by deep anisotropic diffusion. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18237–18246 (2023)

    2023

  45. [56]

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

    Miangoleh, S.M.H., Dille, S., Mai, L., Paris, S., Aksoy, Y.: Boosting monocular depth estimation models to high-resolution via content-adaptive multi-resolution merging. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9685–9694 (2021)

    2021

  46. [57]

    In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Niemeyer, M., Mescheder, L., Oechsle, M., Geiger, A.: Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision. In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 3504–3515 (2020)

    2020

  47. [58]

    In: Proceedings of the AAAI Conference on Artificial In- telligence

    Ouyang, W., Song, X., Feng, B., Xu, Z.: Octocc: High-resolution 3d occupancy prediction with octree. In: Proceedings of the AAAI Conference on Artificial In- telligence. vol. 38, pp. 4369–4377 (2024)

    2024

  48. [59]

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

    Pan, B., Lin, W., Fang, X., Huang, C., Zhou, B., Lu, C.: Recurrent residual module for fast inference in videos. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 1536–1545 (2018)

    2018

  49. [60]

    arXiv preprint arXiv:2102.07887 (2021)

    Pan, B., Panda, R., Fosco, C., Lin, C.C., Andonian, A., Meng, Y., Saenko, K., Oliva, A., Feris, R.: Va-red2: Video adaptive redundancy reduction. arXiv preprint arXiv:2102.07887 (2021)

    2021 arXiv

  50. [61]

    Advances in neural information processing systems34, 24898–24911 (2021) Abbreviated paper title 19

    Pan, B., Panda, R., Jiang, Y., Wang, Z., Feris, R., Oliva, A.: Ia-red 2: Interpretability-aware redundancy reduction for vision transformers. Advances in neural information processing systems34, 24898–24911 (2021) Abbreviated paper title 19

    2021

  51. [62]

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

    Park, J.H., Jeong, C., Lee, J., Jeon, H.G.: Depth prompting for sensor-agnostic depth estimation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9859–9869 (2024)

    2024

  52. [63]

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

    Peng, R., Wang, R., Wang, Z., Lai, Y., Wang, R.: Rethinking depth estimation for multi-view stereo: A unified representation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8645–8654 (2022)

    2022

  53. [64]

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

    Qi, C.R., Su, H., Mo, K., Guibas, L.J.: Pointnet: Deep learning on point sets for 3d classification and segmentation. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 652–660 (2017)

    2017

  54. [65]

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

    Ranftl, R., Bochkovskiy, A., Koltun, V.: Vision transformers for dense prediction. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 12179–12188 (2021)

    2021

  55. [66]

    IEEE transactions on pattern analysis and machine intelligence44(3), 1623–1637 (2020)

    Ranftl, R., Lasinger, K., Hafner, D., Schindler, K., Koltun, V.: Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE transactions on pattern analysis and machine intelligence44(3), 1623–1637 (2020)

    2020

  56. [67]

    Advances in neural infor- mation processing systems34, 13937–13949 (2021)

    Rao, Y., Zhao, W., Liu, B., Lu, J., Zhou, J., Hsieh, C.J.: Dynamicvit: Efficient vision transformers with dynamic token sparsification. Advances in neural infor- mation processing systems34, 13937–13949 (2021)

    2021

  57. [68]

    In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition

    Ren, M., Pokrovsky, A., Yang, B., Urtasun, R.: Sbnet: Sparse blocks network for fast inference. In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. pp. 8711–8720 (2018)

    2018

  58. [69]

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

    Rey-Area, M., Yuan, M., Richardt, C.: 360monodepth: High-resolution 360deg monocular depth estimation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3762–3772 (2022)

    2022

  59. [70]

    In: European conference on computer vision

    Riegler, G., Rüther, M., Bischof, H.: Atgv-net: Accurate depth super-resolution. In: European conference on computer vision. pp. 268–284. Springer (2016)

    2016

  60. [71]

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

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  61. [72]

    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)

    2016

  62. [73]

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

    Schops, T., Schonberger, J.L., Galliani, S., Sattler, T., Schindler, K., Pollefeys, M., Geiger, A.: A multi-view stereo benchmark with high-resolution images and multi-camera videos. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 3260–32...

    2017

  63. [74]

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

    Spencer, J., Russell, C., Hadfield, S., Bowden, R.: Kick back & relax: Learning to reconstruct the world by watching slowtv. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 15768–15779 (2023)

    2023

  64. [75]

    arXiv preprint arXiv:1904.04514 (2019)

    Sun, K., Zhao, Y., Jiang, B., Cheng, T., Xiao, B., Liu, D., Mu, Y., Wang, X., Liu, W., Wang, J.: High-resolution representations for labeling pixels and regions. arXiv preprint arXiv:1904.04514 (2019)

    1904 arXiv

  65. [76]

    Proceedings of Machine Learning and Systems4, 302–315 (2022)

    Tang, H., Liu, Z., Li, X., Lin, Y., Han, S.: Torchsparse: Efficient point cloud infer- ence engine. Proceedings of Machine Learning and Systems4, 302–315 (2022)

    2022

  66. [77]

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

    Tang, J., Tian, F.P., An, B., Li, J., Tan, P.: Bilateral propagation network for depth completion. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9763–9772 (2024)

    2024

  67. [78]

    Author et al

    Tang, S., Chen, J., Wang, D., Tang, C., Zhang, F., Fan, Y., Chandra, V., Furukawa, Y.,Ranjan,R.:Mvdiffusion++:Adensehigh-resolutionmulti-viewdiffusionmodel 20 F. Author et al. for single or sparse-view 3d object reconstruction. In: European Conference on Computer Vision. pp. 1...

    2024

  68. [79]

    In: 2017 IEEE/RSJ international conference on intelligent robots and sys- tems (IROS)

    Tobin, J., Fong, R., Ray, A., Schneider, J., Zaremba, W., Abbeel, P.: Domain randomization for transferring deep neural networks from simulation to the real world. In: 2017 IEEE/RSJ international conference on intelligent robots and sys- tems (IROS). pp. 23–30. IEEE (2017)

    2017

  69. [80]

    IEEE Transactions on pattern analysis and machine intelligence 13(4), 376–380 (2002)

    Umeyama, S.: Least-squares estimation of transformation parameters between two point patterns. IEEE Transactions on pattern analysis and machine intelligence 13(4), 376–380 (2002)

    2002

  70. [81]

    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)

    2025

  71. [82]

    arXiv preprint arXiv:1912.09678 (2019)

    Wang, Q., Zheng, S., Yan, Q., Deng, F., Zhao, K., Chu, X.: Irs: A large naturalistic indoor robotics stereo dataset to train deep models for disparity and surface normal estimation. arXiv preprint arXiv:1912.09678 (2019)

    1912 arXiv

  72. [83]

    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)

    2024

  73. [84]

    arXiv preprint arXiv:2507.13347 (2025)

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

    2025 arXiv

  74. [85]

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

    Wei, Y., Liu, S., Rao, Y., Zhao, W., Lu, J., Zhou, J.: Nerfingmvs: Guided opti- mization of neural radiance fields for indoor multi-view stereo. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 5610–5619 (2021)

    2021

  75. [86]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wen, B., Trepte, M., Aribido, J., Kautz, J., Gallo, O., Birchfield, S.: Foundation- stereo: Zero-shot stereo matching. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5249–5260 (2025)

    2025

  76. [87]

    arXiv preprint arXiv:2011.11865 (2020)

    Xian, C., Qian, K., Zhang, Z., Wang, C.C.: Multi-scale progressive fusion learning for depth map super-resolution. arXiv preprint arXiv:2011.11865 (2020)

    2011 arXiv

  77. [88]

    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)

    2024

  78. [89]

    Advances in Neural Information Processing Systems37, 21875–21911 (2024)

    Yang, L., Kang, B., Huang, Z., Zhao, Z., Xu, X., Feng, J., Zhao, H.: Depth anything v2. Advances in Neural Information Processing Systems37, 21875–21911 (2024)

    2024

  79. [90]

    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)

    2018

  80. [91]

    In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition

    Yao, Y., Luo, Z., Li, S., Zhang, J., Ren, Y., Zhou, L., Fang, T., Quan, L.: Blended- mvs: A large-scale dataset for generalized multi-view stereo networks. In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1790–1799 (2020)

    2020

  81. [92]

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

    Yeshwanth, C., Liu, Y.C., Nießner, M., Dai, A.: Scannet++: A high-fidelity dataset of 3d indoor scenes. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12–22 (2023)

    2023

  82. [93]

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

    Zhang, G., Lu, X., Tan, J., Li, J., Zhang, Z., Li, Q., Hu, X.: Refinemask: Towards high-quality instance segmentation with fine-grained features. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6861– 6869 (2021)

    2021

  83. [94]

    In: 2020 IEEE International Symposium on High Perfor- mance Computer Architecture (HPCA)

    Zhang, Z., Wang, H., Han, S., Dally, W.J.: Sparch: Efficient architecture for sparse matrix multiplication. In: 2020 IEEE International Symposium on High Perfor- mance Computer Architecture (HPCA). pp. 261–274. IEEE (2020) Abbreviated paper title 21

    2020

  84. [95]

    ACM Computing Surveys55(14s), 1–36 (2023)

    Zhong, Z., Liu, X., Jiang, J., Zhao, D., Ji, X.: Guided depth map super-resolution: A survey. ACM Computing Surveys55(14s), 1–36 (2023)

    2023

  85. [96]

    arXiv preprint arXiv:2507.11539 (2025)

    Zhuo, D., Zheng, W., Guo, J., Wu, Y., Zhou, J., Lu, J.: Streaming 4d visual geometry transformer. arXiv preprint arXiv:2507.11539 (2025)

    2025 arXiv

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.