Pith. sign in

REVIEW 4 major objections 6 minor 69 references

Dense Metric Depth Completion from Sparse Direct Time-of-Flight Sensors

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A single model trained only on simulated dToF depth performs dense metric depth completion across six benchmarks and three real sensor families, running in tens of milliseconds.

desk verdict Masked joint attention is a solid, well-ablated contribution, but the '3 real dToF devices' zero-shot claim doesn't survive the actual evaluation protocol. read the letter →

arxiv 2608.04737 v1 pith:4NYOHYX6 submitted 2026-08-05 cs.CV cs.GR

Hakyeong Kim , Ruicheng Wang , Chengtang Yao , Jiaolong Yang , Min H. Kim This is my paper
classification cs.CVcs.GR
keywords depthcompletiondirecttime-of-flightsparsemetricestimationzero-shotgeneralizationvisiontransformermaskedjointattentionsensorsimulation
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

The paper sets out to show that dense metric depth can be completed from the extremely sparse, noisy, and low-resolution outputs of direct time-of-flight sensors using a single model trained entirely on synthetic data. The authors argue that prior depth-completion methods either tie themselves to one sensor layout or pay heavy inference costs with diffusion or iterative refinement, whereas their dual-branch transformer with masked joint attention fuses RGB and sparse depth at the encoder level and keeps a lightweight decoder. If the claim holds, VR/XR, robotics, and 3D perception would no longer need per-device depth pipelines or paired real-sensor training data: one synthetic-trained network would transfer zero-shot across flash, sub-VGA flash, and rotating LiDAR-style dToF devices, with accuracy and speed that beat existing approaches.

What carries the argument

The load-bearing object is the masked joint attention block, a modification of the standard self-attention layer. In each block, image tokens and depth tokens are concatenated and queries, keys, and values are computed jointly; a directional mask $G = \begin{bmatrix}1&1\\0&1\end{bmatrix}$ permits depth-to-image attention but suppresses image-to-depth attention. This lets sparse, accurate dToF measurements steer the image features while preventing unreliable RGB cues from corrupting the geometric measurements. The companion mechanism is the dToF simulation pipeline: it procedurally generates flash patterns with 64 to 10K random points, sub-VGA flash patterns with downsampled grids and erosion-dilation edge degradation, and rotating LiDAR patterns with line-structured scans and randomized angular parameters, then adds noise, jitter, outliers, and occlusion inpainting, so training on synthetic RGB-D datasets transfers to real sensors.

What would settle it

Collect paired RGB and dToF depth from a real dToF device whose sampling pattern, resolution, or noise profile was not used in training, such as a new flash array with a different fill factor or a LiDAR with a different vertical field of view, and measure relative error; if the error is much larger than on the simulated benchmarks, or the model fails on a common artifact like multipath interference, the simulation-fidelity assumption carrying the paper's claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that a depth-guided dual-branch Vision Transformer encoder, combined with a simulation pipeline that replicates flash, sub-VGA flash, and rotating dToF sensing, yields a single model that generalizes zero-shot from synthetic training to real sparse depth inputs. In masked joint attention, image and depth tokens are concatenated, queries, keys, and values are computed jointly, and a directional mask allows depth tokens to guide image features while blocking image-to-depth information flow; this preserves the depth representation and produces depth-aware image features. A lightweight multi-scale decoder then predicts normalized dense depth and a validity mask, with metric scale recovered by de-normalizing with the same log-scale parameters used in preprocessing. The paper reports that this model, trained on millions of synthetic frames from indoor, outdoor, driving, and object domains, outperforms state-of-the-art depth completion methods on six benchmarks and three real dToF devices while being roughly an order of magnitude faster and lighter.

Load-bearing premise

The load-bearing premise is that the synthetic dToF simulation reproduces the real sparsity, noise, and hardware artifacts of flash, sub-VGA flash, and rotating sensors well enough that a model trained only on simulated depth transfers to real devices; if the simulation misses a systematic real-sensor effect, the zero-shot generalization claim weakens.

Editorial extensions

If this is right

  • A single synthetic-trained network can replace sensor-specific depth-completion pipelines, because the same weights handle 8x8 dToF patches, 500-point random samples, and full LiDAR line scans.
  • The reported 34 ms and 0.44 GB inference cost make the approach practical for mobile or real-time use, where diffusion-based and iterative-refinement baselines are too slow.
  • Accuracy at extreme sparsity, such as 100 points, suggests that even very low-cost, low-resolution dToF modules could support dense metric depth in consumer devices.
  • The validity-mask output gives a per-pixel reliability signal that downstream systems, such as occlusion handling or sensor fusion, can use directly.
  • Scaling the backbone from small to large improves all reported benchmarks, so the architecture's benefit is tied to model capacity rather than to a particular size.

Reading between the lines

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

  • An implication the paper leaves implicit is that the same simulation-plus-fusion recipe is a natural template for other sparse-to-dense sensors, such as event cameras, single-photon LiDAR, or radar, provided their sampling and noise statistics can be procedurally modeled.
  • If simulation fidelity is the true cause of zero-shot transfer, then retraining under perturbed simulated noise or sparsity distributions should shift benchmark errors in a predictable direction; the paper does not run this experiment.
  • The directional mask's one-way information flow could apply to any fusion task where a sparse, trustworthy modality must guide a dense, ambiguous one, not only depth completion.
  • A practical test beyond the paper's benchmarks is deployment on a dToF device built after the paper's simulation parameters, since the claim predicts no retraining should be needed.
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

4 major / 6 minor

Summary. This paper presents a depth completion framework that takes an RGB image and a sparse direct time-of-flight (dToF) depth map as input and predicts a dense metric depth map and a validity mask. The method uses a dual-branch Vision Transformer encoder with a masked joint attention mechanism that lets depth tokens guide image tokens while preventing image tokens from overwriting depth tokens. The authors also propose a synthetic dToF simulation pipeline covering flash, sub-VGA flash, and rotating LiDAR-style sensors, and a training regime with depth-weighted L1, global and local scale-invariant losses in 3D, and a mask loss. They evaluate the model zero-shot on six datasets (KITTI-DC, ZJUL5, DDAD, DIODE, ETH3D, iBims-1) and claim strong generalization to three real dToF devices, with better average accuracy and speed than several state-of-the-art depth completion methods. The paper includes ablations on the fusion architecture, loss functions, backbone scale, and sparsity/robustness.

Significance. If the central claims hold, the paper offers a practical and efficient solution for dense depth completion from sparse dToF measurements, an important problem for VR/XR, robotics, and mobile perception. The proposed masked joint attention architecture is a clean and plausible way to fuse RGB and sparse depth while preserving depth-token integrity, and the ablation study supports its benefit. The synthetic simulation pipeline is a valuable resource for training without real paired data, and the authors plan to release code and models, which would aid reproducibility. The main contributions—fast inference, strong zero-shot behavior, and a sensor-agnostic design—are well aligned with the needs of the field. However, the headline claims about 'entirely synthetic training' and evaluation on 'three real dToF devices' are overstated relative to the actual protocol, and the rotating-sensor simulation does not match the real sensor used for evaluation. These issues are correctable but should be addressed before the paper is accepted.

major comments (4)
  1. [Section 4.2; Abstract] The claim of zero-shot generalization across '3 real dToF devices' is not supported by the reported protocol. Section 4.2 lists only KITTI-DC and ZJUL5 under 'real-sensor evaluation'; DDAD is described under 'simulated benchmarks' with 500 points randomly sampled from ground-truth depth maps, which discards the native Luminar-H2 scan geometry. The sentence 'Notably, DDAD is captured using Luminar-H2, adding another real dToF sensor under evaluation' does not justify counting DDAD as a third real-device test, because the evaluation does not use the sensor's native measurements. Please rephrase the claim to count two real sensors, or add an evaluation using the native Luminar-H2 pattern.
  2. [Section 3.5 vs. Section 4.2] The rotating dToF simulation is parameterized for the Velodyne VLP-16 and VLP-32 families, but the real rotating-sensor benchmark, KITTI-DC, is captured with a Velodyne HDL-64E, which has 64 beams and a different vertical field of view. This is a distribution shift between the simulated family and the evaluated real sensor, so the reported synthetic-to-real transfer for rotating sensors is an extrapolation beyond the simulation. The authors should either add a simulation of the HDL-64E pattern or explicitly acknowledge this mismatch and discuss why VLP-family simulation transfers to a 64-beam device.
  3. [Abstract; Section 1; Section 4.1] The statement 'trained entirely on synthetic data' (and 'trained solely on synthetic data' in the contributions) is contradicted by Section 4.1, which states that the encoder is initialized with ViT-Small weights pretrained using DINOv2, and DINOv2 is pretrained on real images. The depth-completion training uses only synthetic paired RGB-depth data, but the model is not trained entirely on synthetic data. Please rephrase to something like 'trained only on synthetic paired RGB-depth data, initialized from a real-image-pretrained backbone,' or provide evidence that the DINOv2 initialization does not contribute to the zero-shot result.
  4. [Section 4.2; PromptDA comparison] The comparison with PromptDA is weakened by the adaptation described in Section 4.2: because PromptDA does not natively support sparse depth, the authors complete missing regions with flood-fill after nearest-neighbor downsampling. This gives PromptDA a dense, interpolated depth input rather than the true sparse dToF input, which may alter its behavior in ways that are not quantified. Please justify this adaptation or compare with a version of PromptDA that ingests sparse depth more directly, and report the sensitivity of the comparison to the filling strategy.
minor comments (6)
  1. [Abstract vs. Section 1] The abstract states 'Our code and models are open-sourced', while the main text says 'Our code and models will be open-sourced'; please make these statements consistent.
  2. [Section 4.2] The sentence 'As these datasets provide real sensor measurements, we follow the common depth completion protocol that generates sparser inputs via downsampling' is confusing because it appears in the 'simulated benchmarks' paragraph; please clarify which datasets provide native sensor measurements and what exactly is simulated.
  3. [Equation (2)] The 2x2 block mask G is described in the text, but its extension to multi-head attention and to varying sequence lengths is not spelled out; please clarify the dimensions and the exact masking operation on the attention logits.
  4. [Table 1] The claim of outperforming state-of-the-art methods in both accuracy and efficiency should be qualified, because on KITTI-DC, OMNI-DC achieves a lower Rel (1.48 vs. 2.00) while being slower; the statement holds on average and for most datasets but not universally.
  5. [Table 1, ZJUL5] The ZJUL5 relative error of 11.13% is the best among the compared methods but is high in absolute terms; the discussion should acknowledge the remaining difficulty of extremely sparse, low-resolution dToF inputs on this benchmark.
  6. [Section 3.2] The phrase 'tokens from the image [QI, KI, VI] and depth branches [QZ, KZ, VZ] are first concatenated' is imprecise: in standard attention, queries, keys, and values are computed from the concatenated token sequence, not concatenated before projection. Please rephrase for clarity.

Circularity Check

1 steps flagged · score 4.0 of 10

Masked-joint-attention architecture is independently ablated, but the headline '3 real dToF devices' count reduces partly to a simulated random-sampling benchmark (DDAD); MoGe self-citations are supportive, not load-bearing.

  1. other [Section 4.2 (Evaluation Setting), in tension with the abstract's '3 real dToF devices' claim]
    "For real-sensor evaluation, we use KITTI-DC [32], captured with a Velodyne LiDAR, and ZJUL5 [46], obtained from the lightweight VL53L5CX sensor. ... For simulated benchmarks, we randomly sample 500 sparse depth points from the ground-truth depth maps. ... Notably, DDAD is captured using Luminar-H2, adding another real dToF sensor under evaluation."

    The abstract claims zero-shot generalization across '3 real dToF devices.' Under the paper's own Section 4.2 protocol, the third device (DDAD's Luminar-H2) is evaluated by 'randomly sample 500 sparse depth points from the ground-truth depth maps,' which discards the native Luminar-H2 scan geometry and coincides with the flash-dToF simulation protocol of Section 3.5 ('we randomly sample 64–10K points from the ground-truth depth map'). The 'third real device' in the headline is therefore constituted by the paper's own simulated input-generation procedure rather than by real sensor measurements: the device count reduces by construction to two real-sensor evaluations plus one simulated random-sampling benchmark.

full rationale

This paper's core technical contribution — the depth-guided dual-branch encoder with masked joint attention — is tested in a self-contained way: Section 4.5 and Table 3/V ablate masked joint attention against decoder-level depth prompting and unmasked joint attention under identical training, and real-sensor results on KITTI-DC and ZJUL5 plus external benchmarks provide independent comparison points. No parameter is fitted to the evaluation sets; preprocessing normalization (Eq. 1) and its inverse (Eq. 3) are an invertible input-dependent re-scaling, and the losses (Eqs. 4–7) supervise against independent ground truth, so no 'prediction' is forced by a fitted input. The self-citations to MoGe/MoGe-2 (ROE scale solver in Eq. 6, the MoGe-style mask loss in Eq. 7, the MoGe-2 multi-scale decoder, and the MoGe training weighting) come from overlapping authors, but they are supporting engineering tools: the ablation in Table II shows the ROE-based global/local losses add only a small gain (3.53 to 3.46 Rel), and the central masked-joint-attention result is not inherited from those citations, so they are not load-bearing circularity. The one genuine reduction is in the headline '3 real dToF devices' claim: Section 4.2 places DDAD under 'simulated benchmarks' with random 500-point sampling, discarding the native Luminar-H2 scan pattern, yet still counts DDAD as 'adding another real dToF sensor under evaluation.' The third device in the headline is thus constructed from the same random-sampling protocol used to synthesize training inputs (Section 3.5), not from real sensor measurements — a claim-protocol reduction that should weigh on correctness/overclaiming, even though the architectural result is unaffected. Additional fidelity risk, not circularity: the rotating-sensor simulation covers Velodyne VLP-16/32 while KITTI-DC uses a Velodyne HDL-64E, so the synthetic-to-real evidence rests on one flash sensor and one rotating sensor outside its simulated family.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical entities. The main accepted costs are the simulation fidelity assumption and the use of a pretrained encoder initialized on real data, which is not disclosed in the abstract's 'trained entirely on synthetic data' phrasing.

free parameters (4)
  • Flash dToF point count range = 64-10K points
    Flash dToF simulation samples 64-10K points per image, following real SPAD array sizes; the range is chosen by hand from sensor specs.
  • Rotating sensor Gaussian noise sigma = up to 0.001
    Noise level for rotating dToF simulation, chosen to mimic Velodyne VLP-16/32 characteristics.
  • Outlier injection ratio = 0.2-1.0%
    Random outlier points added to simulate multipath and hardware errors; ratio chosen to match typical failure rates.
  • Loss weights = equal (1:1:1:1)
    All four loss terms weighted equally, following prior work [37] design choice.
assumptions (3)
  • domain assumption Depth completion from sparse dToF measurements is a well-posed learning problem; the simulation captures the real distribution adequately for transfer.
    Section 3.5 assumes synthetic training transfers to real sensors; this is the core assumption behind zero-shot generalization.
  • domain assumption DINOv2 pretrained features, trained on real images, provide a suitable initialization for the dual-branch encoder including the depth branch.
    Section 3.2 initializes both branches with DINOv2 weights; the depth branch input is a normalized three-channel tensor, not RGB, so the transfer is assumed.
  • standard math The ROE solver (from MoGe [37]) computes the global scale factor correctly for the scale-invariant loss.
    Section 3.4 relies on ROE solver for Lg; this is a published algorithm, not re-derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dense Metric Depth Completion from Sparse Direct Time-of-Flight Sensors." pith.science (2026). https://pith.science/paper/4NYOHYX6

@misc{pith2026260804737,
  author       = {Pith},
  title        = {Pith review of: Dense Metric Depth Completion from Sparse Direct Time-of-Flight Sensors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4NYOHYX6}},
  note         = {Machine review of arXiv:2608.04737}
}
read the original abstract

Direct Time-of-Flight (dToF) sensors provide highly accurate metric depth and are more robust than indirect ToF systems in challenging real-world conditions. However, their high manufacturing cost and limited photodiode array size produce depth maps that are extremely sparse, low-resolution, and noisy, making them unsuitable for VR/XR, robotics, and 3D perception tasks that require dense metric depth. Existing monocular and depth completion methods struggle to handle the unique sampling patterns and hardware artifacts of dToF devices, and their performance often deteriorates significantly under severe sparsity or noise. We present a generalizable framework for dense metric depth completion from sparse dToF measurements, capable of operating across diverse sensor types, sparsity levels, and noise conditions. Our model employs a depth-guided dual-branch Vision Transformer encoder that processes RGB images and sparse dToF measurements separately, while a masked joint attention module allows depth tokens to reliably guide image features without being overwritten by them. A lightweight decoder reconstructs dense metric depth efficiently, without diffusion-based or refinement-heavy post-processing. To address the scarcity of paired training data, we introduce a comprehensive dToF simulation pipeline that reproduces the characteristics of flash, sub-VGA flash, and rotating sensors, including hardware-induced degradation, irregular sparsity, and realistic noise distributions. Trained entirely on synthetic data, our model achieves strong zero-shot generalization across 6 datasets and 3 real dToF devices, outperforming state-of-the-art approaches in both accuracy and computational efficiency. This establishes a robust and practical solution for dense metric depth completion from sparse direct ToF sensors. Our code and models are open-sourced. See https://vclab.kaist.ac.kr/cvpr2026p3.

Figures

Figures reproduced from arXiv: 2608.04737 by the authors.

Figure 1
Figure 1. Zero-shot generalization of our model across different dToF sensing conditions. Top: sparse depth inputs from three representative [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our method. Given an input RGB image and sparse dToF depth, we first normalize and upsample the sparse [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Left: in our masked joint attention, queries, keys, and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: We simulate three representative patterns of direct ToF [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Quantitative analysis on varying number of depth inputs. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison of metric depth maps, and reconstructed point clouds. The two examples are taken from the Library scene [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

69 extracted references · 41 canonical work pages

  1. [1]

    International Conference on 3D Vision (3DV) , year =

    Jonas Uhrig and Nick Schneider and Lukas Schneider and Uwe Franke and Thomas Brox and Andreas Geiger , title =. International Conference on 3D Vision (3DV) , year =

    work page

  2. [2]

    IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , primaryClass =

    Vitor Guizilini and Rares Ambrus and Sudeep Pillai and Allan Raventos and Adrien Gaidon , title =. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , primaryClass =

    work page

  3. [3]

    Dai and Andrea F

    Igor Vasiljevic and Nick Kolkin and Shanyi Zhang and Ruotian Luo and Haochen Wang and Falcon Z. Dai and Andrea F. Daniele and Mohammadreza Mostajabi and Steven Basart and Matthew R. Walter and Gregory Shakhnarovich , year =. CoRR , volume=. 2019 , url=

    2019

  4. [4]

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

    Evaluation of cnn-based single-image depth estimation methods , author=. Proceedings of the European Conference on Computer Vision (ECCV) Workshops , pages=

    work page

  5. [5]

    ops and Johannes L. Sch\

    Thomas Sch\"ops and Johannes L. Sch\"onberger and Silvano Galliani and Torsten Sattler and Konrad Schindler and Marc Pollefeys and Andreas Geiger , title =. Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    work page

  6. [6]

    European Conference on Computer Vision (ECCV) , year=

    DELTAR: Depth Estimation from a Light-weight ToF Sensor and RGB Image , author=. European Conference on Computer Vision (ECCV) , year=

    work page

  7. [7]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Objaverse: A universe of annotated 3d objects , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  8. [8]

    European Conference on Computer Vision , pages=

    Gen-lanenet: A generalized and scalable approach for 3d lane detection , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    work page 2020

Show all 69 references
  1. [9]

    Proceedings of the IEEE/CVF Winter Conference of Applications on Computer Vision (WACV) , year =

    Le, Hoang. Proceedings of the IEEE/CVF Winter Conference of Applications on Computer Vision (WACV) , year =

  2. [10]

    IEEE Robotics and Automation Letters , volume=

    Flow-motion and depth network for monocular stereo and beyond , author=. IEEE Robotics and Automation Letters , volume=. 2020 , publisher=

    2020

  3. [11]

    Susskind , title =

    Mike Roberts AND Jason Ramapuram AND Anurag Ranjan AND Atulit Kumar AND Miguel Angel Bautista AND Nathan Paczan AND Russ Webb AND Joshua M. Susskind , title =. International Conference on Computer Vision (ICCV) 2021 , year =

    2021

  4. [12]

    arXiv preprint arXiv:1912.09678 , year=

    Irs: A large naturalistic indoor robotics stereo dataset to train deep models for disparity and surface normal estimation , author=. arXiv preprint arXiv:1912.09678 , year=

    1912 arXiv

  5. [13]

    ACM Transactions on Graphics (ToG) , volume=

    3d ken burns effect from a single image , author=. ACM Transactions on Graphics (ToG) , volume=. 2019 , publisher=

    2019

  6. [14]

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

    Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  7. [15]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , pages=

    Mid-air: A multi-modal dataset for extremely low altitude drone flights , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , pages=

  8. [16]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deepmvs: Learning multi-view stereopsis , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  9. [17]

    European Conference on Computer Vision , pages=

    Structured3d: A large photo-realistic dataset for structured 3d modeling , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  10. [18]

    arXiv preprint arXiv:1810.08705 , year=

    Synscapes: A photorealistic synthetic dataset for street scene parsing , author=. arXiv preprint arXiv:1810.08705 , year=

    arXiv

  11. [19]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  12. [20]

    Neurocomputing , volume=

    All for one, and one for all: Urbansyn dataset, the third musketeer of synthetic driving scenes , author=. Neurocomputing , volume=. 2025 , publisher=

    2025

  13. [21]

    Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Fabio Tosi and Yiyi Liao and Carolin Schmitt and Andreas Geiger , title =. Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  14. [22]

    2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Tartanair: A dataset to push the limits of visual slam , author=. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2020 , organization=

    2020

  15. [23]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Vision transformers for dense prediction , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  16. [24]

    ArXiv , year =

    Scaling Rectified Flow Transformers for High-Resolution Image Synthesis , author =. ArXiv , year =

  17. [25]

    arXiv:2304.07193 , year=

    DINOv2: Learning Robust Visual Features without Supervision , author=. arXiv:2304.07193 , year=

    arXiv

  18. [26]

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

    Depth prompting for sensor-agnostic depth estimation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  19. [27]

    Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

    Consistent direct time-of-flight video depth super-resolution , author=. Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

  20. [28]

    European Conference on Computer Vision , pages=

    Disambiguating monocular depth estimation with a single transient , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  21. [29]

    arXiv preprint arXiv:2503.04171 , year=

    DuCos: Duality Constrained Depth Super-Resolution via Foundation Model , author=. arXiv preprint arXiv:2503.04171 , year=

    arXiv

  22. [30]

    Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

    Towards fast and accurate real-world depth super-resolution: Benchmark dataset and baseline , author=. Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

  23. [31]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Recurrent structure attention guidance for depth super-resolution , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  24. [32]

    2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Penet: Towards precise and efficient image guided depth completion , author=. 2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2021 , organization=

    2021

  25. [33]

    Proceedings of the ieee/cvf international conference on computer vision , pages=

    Unsupervised depth completion with calibrated backprojection layers , author=. Proceedings of the ieee/cvf international conference on computer vision , pages=

  26. [34]

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

    Wang, Yufei and Li, Bo and Zhang, Ge and Liu, Qi and Gao, Tao and Dai, Yuchao , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

    2023

  27. [35]

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

    Bilateral propagation network for depth completion , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  28. [36]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Completionformer: Depth completion with convolutions and vision transformers , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  29. [37]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Distilling monocular foundation model for fine-grained depth completion , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  30. [38]

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

    Zhu, Xuan and Xiang, Jijun and Wang, Xianqi and Liu, Longliang and Wang, Yu and Zhang, Hong and Guo, Fei and Yang, Xin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2025 , pages =

    2025

  31. [39]

    arXiv preprint arXiv:2504.01596 , year=

    DEPTHOR: Depth Enhancement from a Practical Light-Weight dToF Sensor and RGB Image , author=. arXiv preprint arXiv:2504.01596 , year=

    arXiv

  32. [40]

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

    HFD-Teacher: High-Frequency Depth Distillation from Depth Foundation Models for Enhanced Depth Completion , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  33. [41]

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

    Omni-dc: Highly robust depth completion with multiresolution depth integration , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  34. [42]

    arXiv preprint arXiv:2505.10565 , year=

    Depth Anything with Any Prior , author=. arXiv preprint arXiv:2505.10565 , year=

    arXiv

  35. [43]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Prompting depth anything for 4k resolution accurate metric depth estimation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  36. [44]

    arXiv preprint arXiv:2412.18153 , year=

    Depthlab: From partial to complete , author=. arXiv preprint arXiv:2412.18153 , year=

  37. [45]

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

    Marigold-dc: Zero-shot monocular depth completion with guided diffusion , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  38. [46]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Repurposing diffusion-based image generators for monocular depth estimation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  39. [47]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  40. [48]

    FirstName LastName , title =

  41. [49]

    FirstName Alpher , title =

  42. [50]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  43. [51]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  44. [52]

    FirstName Alpher and FirstName Gamow , title =

  45. [53]

    2020 , url =

    Intel Corporation , title =. 2020 , url =

    2020

  46. [54]

    arXiv preprint arXiv:2111.08897 , year=

    Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data , author=. arXiv preprint arXiv:2111.08897 , year=

    arXiv

  47. [55]

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

    Scannet++: A high-fidelity dataset of 3d indoor scenes , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  48. [56]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Metric3d: Towards zero-shot metric 3d prediction from a single image , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  49. [57]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  50. [58]

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

    Towards zero-shot scale-aware monocular depth estimation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  51. [59]

    Depth Pro: Sharp Monocular Metric Depth in Less Than a Second , author=

  52. [60]

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

    UniDepth: Universal monocular metric depth estimation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  53. [61]

    arXiv preprint arXiv:2507.02546 , year=

    MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details , author=. arXiv preprint arXiv:2507.02546 , year=

    arXiv

  54. [62]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Depth anything: Unleashing the power of large-scale unlabeled data , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  55. [63]

    Advances in Neural Information Processing Systems , volume=

    Depth anything v2 , author=. Advances in Neural Information Processing Systems , volume=

  56. [64]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  57. [65]

    European Conference on Computer Vision , pages=

    Geowizard: Unleashing the diffusion priors for 3d geometry estimation from a single image , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  58. [66]

    Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction , author=

  59. [67]

    Advances in Neural Information Processing Systems , volume=

    Learning affinity via spatial propagation networks , author=. Advances in Neural Information Processing Systems , volume=

  60. [68]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Learning depth with convolutional spatial propagation network , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2019 , publisher=

    2019

  61. [69]

    European Conference on Computer Vision , pages=

    OGNI-DC: Robust depth completion with optimization-guided neural iterations , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.