Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Depth Anything at Any Condition

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 The paper proposes DepthAnything-AC, a foundation monocular depth estimation model that keeps its accuracy in dark, fog, snow, and blur conditions through perturbation-based consistency regularization and a spatial distance constraint…

desk verdict A sensible perturbation-based consistency fine-tune of Depth Anything V2 with a new spatial distance loss, but the real-world gains are small and the strongest benchmark is in-distribution; still worth refereeing. read the letter →

arxiv 2507.01634 v1 pith:X4XAIJNC submitted 2025-07-02 cs.CV cs.AI

Boyuan Sun , Modi Jin , Bowen Yin , Qibin Hou This is my paper
classification cs.CVcs.AI
keywords monoculardepthestimationfoundationmodeladverseweatherrobustnessnighttimeconsistencyregularizationspatialdistanceconstraintzero-shotgeneralizationsyntheticperturbation
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's goal is to make monocular depth estimation as reliable in darkness, fog, snow, and blur as it is in clear conditions, without needing new labeled data in those conditions. The proposed model, DepthAnything-AC, starts from an existing foundation depth model and fine-tunes it on 540K unlabeled ordinary images to which synthetic degradations are applied, demanding consistent depth predictions before and after perturbation. A second supervision, the spatial distance constraint, preserves the geometric relationship between image patches so that object boundaries and details survive the corruptions. The paper claims this outperforms state-of-the-art approaches on real-world and synthetic degraded benchmarks while keeping general-scene performance essentially intact.

What carries the argument

The central objects are two added supervisions layered on the frozen-encoder fine-tune of Depth Anything V2. The perturbation-based consistency framework uses an affine-invariant loss — one that normalizes disparity by its mean and mean absolute deviation, so scale offsets do not count as errors — to force identical predictions from clean and synthetically corrupted views of the same image, while a knowledge-distillation loss from the frozen teacher guards general performance on clean views. The Spatial Distance Relation (SDR) is the geometric prior: for every patch pair, it forms $S_D=\sqrt{S_p^2+S_d^2}$, where $S_p$ is Euclidean distance in the image plane and $S_d$ is absolute disparity difference, and the spatial distance loss aligns the corrupted branch's SDR with the teacher's clean SDR. These two mechanisms carry the argument: consistency transfers robustness from simulated degradations, and SDR supplies patch-level relative-geometry supervision that sharpens boundaries and details.

What would settle it

Run the released DepthAnything-AC on a held-out set of authentic fog, rain, and nighttime images with dense ground truth and compare its AbsRel and delta1 against Depth Anything V2; if it is not better there, or if the gains are no larger than benchmark noise on the rain and fog subsets, the synthetic-to-real transfer claim fails. A second check is to remove the spatial distance loss and measure boundary precision and recall on the DA-2K dark subset; if object-boundary quality does not drop, the SDR is not carrying the claimed improvement.

Watch

Extended reading notes

Core claim

DepthAnything-AC is a foundation monocular depth estimation model built by fine-tuning Depth Anything V2 with the encoder frozen. Its first mechanism is perturbation-based consistency: an unlabeled image receives ordinary weak augmentation for one branch and a synthetic corruption — darkness, weather, blur, or contrast — for another branch, and the student must give the same affine-invariant disparity prediction for both; a frozen copy of the base model supervises the clean branch so that general ability is not lost. Its second mechanism, the Spatial Distance Relation (SDR), computes a per-pair geometric distance $S_D=\sqrt{S_p^2+S_d^2}$ from image-plane position distance and disparity difference, then trains the corrupted branch to match the SDR of the clean teacher branch, which the authors show restores semantic boundaries and fine details. In the paper's experiments this combination beats earlier foundation and robustness-oriented depth models on dark, fog, snow, and blur subsets of DA-2K, on NuScenes-night and RobotCar-night, and on KITTI-C darkness, snow, motion blur, and Gaussian noise, while general benchmarks stay at the base model's level.

Load-bearing premise

The approach assumes that synthetic darkness, fog, snow, and blur applied to clean images behave like the real-world versions, because the model never trains on actual corrupted image-depth pairs.

Editorial extensions

If this is right

  • A foundation depth model can be made resilient to dark, fog, snow, and blur using a small unlabeled set of ordinary images rather than large collections of real corrupted image-depth pairs.
  • The frozen-teacher design keeps the base model's general-scene behavior essentially unchanged, so the same recipe should transfer to stronger future foundation depth models.
  • The spatial distance constraint improves boundary and detail quality under corruption, which should help downstream uses of depth in 3D reconstruction, robot manipulation, and autonomous driving perception.
  • The largest measured improvements appear on high-resolution DA-2K-style benchmarks; the method's practical benefits will show up most where fine detail matters, since gains on sparse LiDAR-style benchmarks are smaller.

Reading between the lines

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

  • The paper does not test whether its synthetic perturbation pool is the ceiling: real rain and fog subsets show near-zero improvement, so a natural next experiment is to add photometrically plausible rain, haze, and real unlabeled adverse clips to the perturbation distribution.
  • Nothing in the SDR formulation ties it to disparity; the same patch-pair geometric loss could supervise surface-normal, semantic-segmentation, or other dense prediction heads, though the paper does not explore this.
  • The frozen-encoder ablation suggests most of the benefit comes from preserving the pretrained representation while adapting the decoder; a cheaper variant that trains only the decoder head is an untested consequence of that finding.
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. The paper proposes DepthAnything-AC, a monocular depth estimation model obtained by fine-tuning DepthAnythingV2 with an unsupervised consistency-regularization objective on unlabeled images. The training applies synthetic perturbations (dark, weather, blur, contrast) and encourages the student to produce consistent depth predictions on original and perturbed views, while a frozen teacher provides guidance on original views. The authors additionally propose a Spatial Distance Constraint that compares pairwise patch distances (combining image-position distance and disparity difference) between the perturbed student prediction and the original teacher prediction. They report zero-shot results on a self-constructed multi-condition DA-2K benchmark, on real-world night/weather benchmarks (NuScenes-night, Robotcar-night, DrivingStereo), on the synthetic KITTI-C benchmark, and on general benchmarks, claiming state-of-the-art robustness while preserving general performance.

Significance. If the claims are substantiated, the paper would offer a practical, data-efficient recipe for adapting foundation depth models to adverse conditions without labeled corrupted data, and the released code would be useful to the community. The training setup uses a small unlabeled corpus (540K images), and the authors provide ablation studies for several components. However, the strength of the evidence is limited by the evaluation design: the main benchmark with the largest gains is constructed from the same perturbation families used in training, and several hyperparameters appear to be selected using the test benchmarks. The Spatial Distance Constraint, as formulated, may reduce to a reweighted consistency loss on disparity differences rather than a genuinely new geometric prior. These issues weaken the central claim of broad real-world robustness and need to be addressed.

major comments (4)
  1. [Appendix A.1 and Table 2] The multi-condition DA-2K benchmark is constructed by applying the same perturbation categories used in training (dark, fog, snow, blur; see Sec. 3.1 and Table 8). The gains on this benchmark therefore primarily measure the model's ability to handle perturbations from the training distribution, not its transfer to genuinely unseen corruption types. To support the claim of robustness beyond the training augmentation distribution, the authors should evaluate on held-out perturbation families (e.g., noise, JPEG artifacts, elastic deformations, or other weather types) or explicitly present DA-2K results only as an in-distribution check.
  2. [Tables 9, 10, 12 and Sec. 4.2] The choices of perturbation probabilities (Table 10a), distance type (Table 10b), training paradigm (Table 10b), and loss weights (Table 12) are selected based on performance on NuScenes-night, Robotcar-night, DS-cloud, and Gaussian (KITTI-C), all of which are test benchmarks in the main comparison. This test-set usage makes the reported zero-shot numbers optimistic and prevents a clean claim of a general, hyperparameter-free recipe. The authors should employ a held-out validation set (e.g., a subset of their unlabeled training pool with synthetic held-out perturbations, or a separate real-world validation set) for all configuration choices, and then report results on the current test benchmarks.
  3. [Sec. 3.2, Eqs. (5)-(8)] The position relation Sp in Eq. (5) depends only on the fixed patch coordinates and is identical for the perturbed student prediction and the original teacher prediction. Consequently, in the loss of Eq. (8), the position term acts as a constant, data-independent weighting on the difference of disparity-relation matrices; it does not inject any novel scene-dependent spatial information. The claim that SDR 'encodes rich semantic information' and provides 'position-aware optimization' is therefore not supported by the formulation. An ablation that isolates Sp (e.g., setting Sp to zero versus using the full SDR) would clarify whether the position term contributes anything beyond a reweighted consistency of disparity differences.
  4. [Table 3 and Sec. 4.2] On real-world benchmarks, the reported gains are small and uneven: DS-rain and DS-fog show identical AbsRel, DS-cloud improves by only 0.003 in delta1, NuScenes-night improves by 0.002 in delta1, while only Robotcar-night shows a larger delta1 gain of 0.037. Given that training uses only synthetic perturbations, the central claim of outperforming prior methods 'in real-world ... scenarios' rests on a single benchmark. The 'Limited improvements on traditional benchmarks' paragraph provides a plausible explanation for why traditional benchmarks may be insensitive, but it does not supply positive evidence of transfer to real corruptions. The authors should either provide additional real-world evaluations (e.g., dense ground-truth depth from other sensors or domains) or substantially temper the real-world claim.
minor comments (6)
  1. [Eq. (4) and Fig. 3] The notation for the frozen teacher is inconsistent: Sec. 3.1 says the initial foundation model F is frozen, but Eq. (4) uses F_hat; Fig. 3 also labels 'Frozen Trained' in an unclear way. Please clarify that F is the trainable student and F_hat is the frozen teacher.
  2. [Table 2] In the robustdepth row, the DA-2K blur value is printed as '0686' instead of '0.686'.
  3. [Fig. 3 caption] The caption contains a typo: 'DepthAnyting-AC' should be 'DepthAnything-AC'.
  4. [Eqs. (5)-(7)] The sets in Eqs. (5)-(7) are used to denote matrices; please use explicit matrix notation (e.g., SD ∈ R^{HW×HW}) for clarity, and define the index ranges of m,n.
  5. [Sec. 3.2] The physical interpretation 'geometric distance' is somewhat misleading because the disparity term is not a metric distance; consider phrasing such as 'pseudo-geometric distance' or 'relation magnitude'.
  6. [Table 2 heading] The metric 'Accuracy' is used for the DA-2K benchmark; please define it in the main text or refer to Depth Anything V2 for the precise pairwise comparison protocol.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline gains are partly by construction: hyperparameters are tuned on the same real benchmarks reported as zero-shot, and the largest synthetic gains come from a benchmark built with the training perturbation families.

  1. fitted input called prediction [Section 4.3 and Appendix B.1/B.4, Tables 9, 10a, and 12 versus Table 3/4 'zero-shot' results]
    "As shown, the model achieves the best performance when the probability of blur augmentation is set to 0.1 and that of weather augmentation to 0.2. ... Overall, setting [λ1, λ2, λ3] to [1/3, 1/3, 1/3] yields the best performance across benchmarks."

    The perturbation probabilities, loss weights, and spatial-distance scheme are selected using the same NuScenes-night, Robotcar-night, and Gaussian test columns that later appear as zero-shot results in the main tables. The reported final gains, e.g., Robotcar-night δ1 0.518 to 0.555, are therefore the outcome of a search over the evaluation benchmarks, not predictions from a fixed model configuration. This is a fitted-input-called-prediction situation: the configuration is chosen by optimizing on the test data, so the headline zero-shot numbers are partly a restatement of the selection criterion rather than independent evidence.

  2. self definitional [Section 3.1, Figure 2, Section 4.2 and Appendix A.1 (multi-condition DA-2K benchmark)]
    "we further built the multi-condition DA-2K benchmark to evaluate the robust capability of foundation MDE models. Specifically, we augment the original DA-2K images with various types of perturbations, namely dark, fog, snow and blur."

    The training objective optimizes consistency between clean and perturbed views using exactly these perturbation families, and the largest reported gains (Table 2, e.g., DA-2K blur 0.862 to 0.880) are measured on a benchmark built by applying the same perturbation families. That benchmark therefore measures whether the consistency objective was optimized in-distribution, not whether the model transfers to genuinely unseen noise. Because the paper uses this self-built benchmark to support the claim of outperforming in synthetic noise scenarios, the evaluation target is partly defined by the same perturbation operators used as training inputs, making that part of the evidence circular.

full rationale

The perturbation-consistency mechanism itself is not intrinsically circular: the losses in Eqs. (1)-(8) are standard consistency and self-distillation terms anchored by a frozen teacher model, and no load-bearing uniqueness theorem or self-citation chain is invoked. The circularity is in the evaluation and selection protocol. First, hyperparameters (blur/weather probabilities, loss weights, distance type) are chosen using the exact real-world benchmarks (NuScenes-night, Robotcar-night) and the Gaussian KITTI-C subset that are later reported as zero-shot results, so those numbers are partly selected rather than predicted. Second, the multi-condition DA-2K benchmark, which shows the most pronounced gains, is constructed by applying the same dark/fog/snow/blur perturbations used during training; it validates in-distribution consistency rather than transfer to real degradation. The paper's own 'Limited improvements' paragraph concedes that gains on independent real benchmarks are limited: on DS-rain and DS-fog AbsRel is unchanged (0.125 and 0.103), and DS-cloud improves only 0.151 to 0.149. KITTI-C provides some independent signal, but its Gaussian subset is also used in hyperparameter selection, so even that benchmark is not fully clean. Overall, this is partial circularity: some headline predictions reduce by construction to test-set tuning and to the training perturbation distribution, though the underlying method is not defined in terms of its outputs. Score 6.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several hand-set or tuned quantities: the three loss weights (default 1/3 each, ablated in Table 12), the probabilities of applying blur and weather perturbations (0.1 and 0.2, selected using results in Table 10a), and the dataset sampling ratios in the 540K training set. The method also assumes that a frozen teacher model provides reliable supervision, that synthetic perturbations represent real-world degradations, and that the affine-invariant loss is appropriate for relative depth. No new physical entities are introduced; the spatial distance relation is a loss construct derived from teacher predictions.

free parameters (5)
  • Loss weights lambda1, lambda2, lambda3 = 1/3, 1/3, 1/3 (default)
    Weights for consistency, distillation, and spatial distance losses; ablated in Table 12 and selected based on performance on NuScenes-night and Robotcar-night.
  • Blur perturbation probability = 0.1
    Probability of applying blur perturbation per image; tuned in Table 10a using NuScenes-night and Robotcar-night results.
  • Weather perturbation probability = 0.2
    Probability of applying weather (fog, snow, contrast) perturbation per image; tuned in Table 10a.
  • Perturbation severity levels = Not fully specified
    Intensity settings for dark, fog, snow, blur, and contrast are described qualitatively in Table 8, with no exact numeric values per severity level.
  • Dataset sampling ratios = Varies (e.g., DIML 8.6%, SA-1B 1.3%, AODRaw 10%)
    Proportions of each dataset used to build the 540K training set are hand-chosen; no sensitivity analysis is provided.
assumptions (4)
  • domain assumption The frozen teacher model DepthAnythingV2 provides reliable pseudo-labels for unperturbed images.
    Invoked in Eqn 4 (knowledge distillation loss) and Eqn 8 (spatial distance loss); assumes teacher predictions are accurate enough to guide the student.
  • domain assumption Simulated perturbations (darkness, weather, blur, contrast) cover the relevant real-world degradation space.
    Stated in Section 3.1: the goal is to simulate common real-world scenarios. If this coverage is incomplete, robustness may not transfer to unseen real corruptions.
  • standard math The affine-invariant loss is an appropriate supervision metric for relative depth.
    Used in Eqn 2 for consistency and distillation, borrowed from DepthAnything and MiDaS; this is a standard loss for relative depth estimation.
  • ad hoc to paper Spatial distance computed as the Euclidean combination of position distance and disparity distance captures semantic boundaries.
    Section 3.2 introduces this geometric prior; the paper provides no external validation that this specific combination is optimal, and it is presented as a contribution of this work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Depth Anything at Any Condition." pith.science (2026). https://pith.science/paper/X4XAIJNC

@misc{pith2026250701634,
  author       = {Pith},
  title        = {Pith review of: Depth Anything at Any Condition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X4XAIJNC}},
  note         = {Machine review of arXiv:2507.01634}
}
read the original abstract

We present Depth Anything at Any Condition (DepthAnything-AC), a foundation monocular depth estimation (MDE) model capable of handling diverse environmental conditions. Previous foundation MDE models achieve impressive performance across general scenes but not perform well in complex open-world environments that involve challenging conditions, such as illumination variations, adverse weather, and sensor-induced distortions. To overcome the challenges of data scarcity and the inability of generating high-quality pseudo-labels from corrupted images, we propose an unsupervised consistency regularization finetuning paradigm that requires only a relatively small amount of unlabeled data. Furthermore, we propose the Spatial Distance Constraint to explicitly enforce the model to learn patch-level relative relationships, resulting in clearer semantic boundaries and more accurate details. Experimental results demonstrate the zero-shot capabilities of DepthAnything-AC across diverse benchmarks, including real-world adverse weather benchmarks, synthetic corruption benchmarks, and general benchmarks. Project Page: https://ghost233lism.github.io/depthanything-AC-page Code: https://github.com/HVision-NKU/DepthAnythingAC

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Geometric Reciprocity: Unlocking Self-Supervision for Stereoscopic Video Generation

    cs.CV 2026-07 accept novelty 7.0 of 10

    The Geometric Reciprocity Theorem proves that DIBR disocclusion masks for stereo synthesis equal the pixels lost during reverse warping, enabling self-supervised stereo inpainting training from monocular images alone.

Reference graph

Works this paper leans on

107 extracted references · 57 canonical work pages · cited by 1 Pith paper

  1. [1]

    Adabins: Depth estimation using adaptive bins

    Shariq Farooq Bhat, Ibraheem Alhashim, and Peter Wonka. Adabins: Depth estimation using adaptive bins. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 4009–4018, 2021. 3

    2021

  2. [2]

    Zoedepth: Zero-shot transfer by combining relative and metric depth

    Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias Müller. Zoedepth: Zero-shot transfer by combining relative and metric depth. arXiv preprint arXiv:2302.12288, 2023. 2, 3

    arXiv 2023

  3. [3]

    Monocular depth estimation: A survey

    Amlaan Bhoi. Monocular depth estimation: A survey. arXiv preprint arXiv:1901.09402, 2019. 3

    arXiv 1901

  4. [4]

    Unsupervised scale-consistent depth and ego-motion learning from monocular video

    Jiawang Bian, Zhichao Li, Naiyan Wang, Huangying Zhan, Chunhua Shen, Ming-Ming Cheng, and Ian Reid. Unsupervised scale-consistent depth and ego-motion learning from monocular video. In Advances in Neural Information Processing Systems (NeurIPS), 2019. 3

    2019

  5. [5]

    Auto-rectify network for unsupervised indoor depth estimation

    Jia-Wang Bian, Huangying Zhan, Naiyan Wang, Tat-Jin Chin, Chunhua Shen, and Ian Reid. Auto-rectify network for unsupervised indoor depth estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2021. 1

    2021

  6. [6]

    Midas v3.1 – a model zoo for robust monocular relative depth estimation

    Reiner Birkl, Diana Wofk, and Matthias Müller. Midas v3.1 – a model zoo for robust monocular relative depth estimation. arXiv preprint arXiv:2307.14460, 2023. 3

    arXiv 2023

  7. [7]

    Depth pro: Sharp monocular metric depth in less than a second

    Aleksei Bochkovskii, AmaÃG, l Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073, 2024. 1, 2, 6, 7, 8, 20, 23

    arXiv 2024

  8. [8]

    A naturalistic open source movie for optical flow evaluation

    Daniel J Butler, Jonas Wulff, Garrett B Stanley, and Michael J Black. A naturalistic open source movie for optical flow evaluation. In Computer Vision–ECCV 2012: 12th European Conference on Computer Vision, Florence, Italy, October 7-13, 2012, Proceedings, Part VI 12, pages 611–625. Springer, 2012. 7, 17, 18

    2012

Show all 107 references
  1. [9]

    Virtual kitti 2, 2020

    Yohann Cabon, Naila Murray, and Martin Humenberger. Virtual kitti 2, 2020. 1, 6

    2020

  2. [10]

    nuscenes: A multimodal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern r...

  3. [11]

    Spatialbot: Precise spatial understanding with vision language models

    Wenxiao Cai, Yaroslav Ponomarenko, Jianhao Yuan, Xiaoqi Li, Wankou Yang, Hao Dong, and Bo Zhao. Spatialbot: Precise spatial understanding with vision language models. arXiv preprint arXiv:2406.13642,

    arXiv

  4. [12]

    Deep monocular depth estimation leveraging a large-scale outdoor stereo dataset

    Jaehoon Cho, Dongbo Min, Youngjung Kim, and Kwanghoon Sohn. Deep monocular depth estimation leveraging a large-scale outdoor stereo dataset. Expert Systems with Applications, 178:114877, 2021. 6

    2021

  5. [13]

    Diml/cvl rgb-d dataset: 2m rgb-d images of natural indoor and outdoor scenes

    Jaehoon Cho, Dongbo Min, Youngjung Kim, and Kwanghoon Sohn. Diml/cvl rgb-d dataset: 2m rgb-d images of natural indoor and outdoor scenes. arXiv preprint arXiv:2110.11590, 2021. 1, 6

    2021 arXiv

  6. [14]

    Boost 3d reconstruction using diffusion-based monocular camera calibration

    Junyuan Deng, Wei Yin, Xiaoyang Guo, Qian Zhang, Xiaotao Hu, Weiqiang Ren, Xiaoxiao Long, and Ping Tan. Boost 3d reconstruction using diffusion-based monocular camera calibration. arXiv preprint arXiv:2411.17240, 2024. 1

    2024

  7. [15]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at...

    2021

  8. [16]

    Depth map prediction from a single image using a multi-scale deep network

    David Eigen, Christian Puhrsch, and Rob Fergus. Depth map prediction from a single image using a multi-scale deep network. Advances in neural information processing systems, 27, 2014. 1, 3

    2014

  9. [17]

    The pascal visual object classes (voc) challenge

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. International journal of computer vision, 88:303–338, 2010. 6

    2010

  10. [18]

    Depth-centric dehazing and depth-estimation from real-world hazy driving video

    Junkai Fan, Kun Wang, Zhiqiang Yan, Xiang Chen, Shangbing Gao, Jun Li, and Jian Yang. Depth-centric dehazing and depth-estimation from real-world hazy driving video. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 2852–2860, 2025. 3

    2025

  11. [19]

    Computer rendering of stochastic models

    Alain Fournier, Don Fussell, and Loren Carpenter. Computer rendering of stochastic models. Communica- tions of the ACM, 25(6):371–384, 1982. 18

    1982

  12. [20]

    Deep ordinal regression network for monocular depth estimation

    Huan Fu, Mingming Gong, Chaohui Wang, Kayhan Batmanghelich, and Dacheng Tao. Deep ordinal regression network for monocular depth estimation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2002–2011, 2018. 1, 3

    2002

  13. [21]

    Geowizard: Unleashing the diffusion priors for 3d geometry estimation from a single image

    Xiao Fu, Wei Yin, Mu Hu, Kaixuan Wang, Yuexin Ma, Ping Tan, Shaojie Shen, Dahua Lin, and Xiaoxiao Long. Geowizard: Unleashing the diffusion priors for 3d geometry estimation from a single image. In ECCV, 2024. 3

    2024

  14. [22]

    Virtual worlds as proxy for multi-object tracking analysis

    Adrien Gaidon, Qiao Wang, Yohann Cabon, and Eleonora Vig. Virtual worlds as proxy for multi-object tracking analysis. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pages 4340–4349, 2016. 6

    2016

  15. [23]

    R4dyn: Exploring radar for self-supervised monocular depth estimation of dynamic scenes

    Stefano Gasperini, Patrick Koch, Vinzenz Dallabetta, Nassir Navab, Benjamin Busam, and Federico Tombari. R4dyn: Exploring radar for self-supervised monocular depth estimation of dynamic scenes. In 2021 International Conference on 3D Vision (3DV), pages 751–760. IEEE, 2021. 3

    2021

  16. [24]

    Vision meets robotics: The kitti dataset

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset. International Journal of Robotics Research (IJRR), 2013. 1, 3, 7, 8, 17

    2013

  17. [25]

    Unsupervised monocular depth estimation with left-right consistency

    Clément Godard, Oisin Mac Aodha, and Gabriel J Brostow. Unsupervised monocular depth estimation with left-right consistency. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 270–279, 2017. 1

    2017

  18. [26]

    Digging into self-supervised monocular depth estimation

    Clément Godard, Oisin Mac Aodha, Michael Firman, and Gabriel J Brostow. Digging into self-supervised monocular depth estimation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3828–3838, 2019. 1

    2019

  19. [27]

    Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020. 3 11

    2020

  20. [28]

    Depthfm: Fast monocular depth estimation with flow matching, 2024

    Ming Gui, Johannes Schusterbauer, Ulrich Prestel, Pingchuan Ma, Dmytro Kotovenko, Olga Grebenkova, Stefan Andreas Baumann, Vincent Tao Hu, and Björn Ommer. Depthfm: Fast monocular depth estimation with flow matching, 2024. 3

    2024

  21. [29]

    Multi-view reconstruction via sfm-guided monocular depth estimation

    Haoyu Guo, He Zhu, Sida Peng, Haotong Lin, Yunzhi Yan, Tao Xie, Wenguan Wang, Xiaowei Zhou, and Hujun Bao. Multi-view reconstruction via sfm-guided monocular depth estimation. In CVPR, 2025. 3

    2025

  22. [30]

    Lotus: Diffusion-based visual foundation model for high-quality dense prediction

    Jing He, Haodong Li, Wei Yin, Yixun Liang, Leheng Li, Kaiqiang Zhou, Hongbo Liu, Bingbing Liu, and Ying-Cong Chen. Lotus: Diffusion-based visual foundation model for high-quality dense prediction. arXiv preprint arXiv:2409.18124, 2024. 3

    2024 arXiv

  23. [31]

    Distill any depth: Distillation creates a stronger monocular depth estimator

    Xiankang He, Dongyan Guo, Hongji Li, Ruibo Li, Ying Cui, and Chi Zhang. Distill any depth: Distillation creates a stronger monocular depth estimator. arXiv preprint arXiv: 2502.19204, 2025. 1

    2025 arXiv

  24. [32]

    Metric3d v2: A versatile monocular geometric foundation model for zero- shot metric depth and surface normal estimation

    Mu Hu, Wei Yin, Chi Zhang, Zhipeng Cai, Xiaoxiao Long, Hao Chen, Kaixuan Wang, Gang Yu, Chunhua Shen, and Shaojie Shen. Metric3d v2: A versatile monocular geometric foundation model for zero- shot metric depth and surface normal estimation. IEEE Transactions on Pattern Analysi...

    2024

  25. [33]

    Drivingworld: Constructingworld model for autonomous driving via video gpt

    Xiaotao Hu, Wei Yin, Mingkai Jia, Junyuan Deng, Xiaoyang Guo, Qian Zhang, Xiaoxiao Long, and Ping Tan. Drivingworld: Constructingworld model for autonomous driving via video gpt. arXiv preprint arXiv:2412.19505, 2024. 1

    2024 arXiv

  26. [34]

    Repurposing diffusion-based image generators for monocular depth estimation

    Bingxin Ke, Anton Obukhov, Shengyu Huang, Nando Metzger, Rodrigo Caye Daudt, and Konrad Schindler. Repurposing diffusion-based image generators for monocular depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1, 3

    2024

  27. [35]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1, 2023. 1

    2023

  28. [36]

    Deep stereo confidence prediction for depth estimation

    Sunok Kim, Dongbo Min, Bumsub Ham, Seungryong Kim, and Kwanghoon Sohn. Deep stereo confidence prediction for depth estimation. In 2017 ieee international conference on image processing (icip), pages 992–996. IEEE, 2017. 6

    2017

  29. [37]

    Structure selective depth superreso- lution for rgb-d cameras

    Youngjung Kim, Bumsub Ham, Changjae Oh, and Kwanghoon Sohn. Structure selective depth superreso- lution for rgb-d cameras. IEEE Transactions on Image Processing, 25(11):5227–5238, 2016. 6

    2016

  30. [38]

    Deep monocular depth estimation via integration of global and local predictions

    Youngjung Kim, Hyungjoo Jung, Dongbo Min, and Kwanghoon Sohn. Deep monocular depth estimation via integration of global and local predictions. IEEE transactions on Image Processing, 27(8):4131–4144,

  31. [39]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4015–4026, 2023. 6

    2023

  32. [40]

    Cottereau, and Wei Tsang Ooi

    Lingdong Kong, Shaoyuan Xie, Hanjiang Hu, Lai Xing Ng, Benoit R. Cottereau, and Wei Tsang Ooi. Robodepth: Robust out-of-distribution depth estimation under corruptions. In Advances in Neural Information Processing Systems, 2023. 1, 3, 7, 8, 17

    2023

  33. [41]

    Visual robotic manipulation with depth-aware pretraining

    Jinming Li, Wanying Wang, Yaxin Peng, Chaomin Shen, Yichen Zhu, and Zhiyuan Xu. Visual robotic manipulation with depth-aware pretraining. In 2024 IEEE International Conference on Robotics and Biomimetics (ROBIO), pages 843–850. IEEE, 2024. 1

    2024

  34. [42]

    Manipllm: Embodied multimodal large language model for object-centric robotic manipulation

    Xiaoqi Li, Mingxu Zhang, Yiran Geng, Haoran Geng, Yuxing Long, Yan Shen, Renrui Zhang, Jiaming Liu, and Hao Dong. Manipllm: Embodied multimodal large language model for object-centric robotic manipulation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Patter...

    2024

  35. [43]

    Patchfusion: An end-to-end tile-based framework for high-resolution monocular metric depth estimation

    Zhenyu Li, Shariq Farooq Bhat, and Peter Wonka. Patchfusion: An end-to-end tile-based framework for high-resolution monocular metric depth estimation. 2024. 1

    2024

  36. [44]

    Megadepth: Learning single-view depth prediction from internet photos

    Zhengqi Li and Noah Snavely. Megadepth: Learning single-view depth prediction from internet photos. In Computer Vision and Pattern Recognition (CVPR), 2018. 6 12

    2018

  37. [45]

    Towards raw object detection in diverse conditions

    Zhong-Yu Li, Xin Jin, Boyuan Sun, Chun-Le Guo, and Ming-Ming Cheng. Towards raw object detection in diverse conditions. arXiv preprint arXiv:2411.15678, 2024. 6, 7

    2024 arXiv

  38. [46]

    Prompting depth anything for 4k resolution accurate metric depth estimation

    Haotong Lin, Sida Peng, Jingxiao Chen, Songyou Peng, Jiaming Sun, Minghuan Liu, Hujun Bao, Jiashi Feng, Xiaowei Zhou, and Bingyi Kang. Prompting depth anything for 4k resolution accurate metric depth estimation. 2024. 1

    2024

  39. [47]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer vision–ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceedings...

    2014

  40. [48]

    Single image depth estimation from predicted semantic labels

    Beyang Liu, Stephen Gould, and Daphne Koller. Single image depth estimation from predicted semantic labels. In 2010 IEEE computer society conference on computer vision and pattern recognition , pages 1253–1260. IEEE, 2010. 3

    2010

  41. [49]

    Self-supervised monocular depth estimation for all day images using domain separation

    Lina Liu, Xibin Song, Mengmeng Wang, Yong Liu, and Liangjun Zhang. Self-supervised monocular depth estimation for all day images using domain separation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12737–12746, 2021. 3

    2021

  42. [50]

    Depthlab: From partial to complete

    Zhiheng Liu, Ka Leong Cheng, Qiuyu Wang, Shuzhe Wang, Hao Ouyang, Bin Tan, Kai Zhu, Yujun Shen, Qifeng Chen, and Ping Luo. Depthlab: From partial to complete. arXiv preprint arXiv:2412.18153, 2024. 3

    2024

  43. [51]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 7

    2017 arXiv

  44. [52]

    1 year, 1000 km: The oxford robotcar dataset

    Will Maddern, Geoffrey Pascoe, Chris Linegar, and Paul Newman. 1 year, 1000 km: The oxford robotcar dataset. The International Journal of Robotics Research, 36(1):3–15, 2017. 1, 7, 8, 17

    2017

  45. [53]

    Stealing stable diffusion prior for robust monocular depth estimation

    Yifan Mao, Jian Liu, and Xianming Liu. Stealing stable diffusion prior for robust monocular depth estimation. arXiv preprint arXiv:2403.05056, 2024. 3

    2024 arXiv

  46. [54]

    Indoor segmentation and support inference from rgbd images

    Pushmeet Kohli Nathan Silberman, Derek Hoiem and Rob Fergus. Indoor segmentation and support inference from rgbd images. In ECCV, 2012. 7, 8, 17, 18

    2012

  47. [55]

    Depth helps: Improving pre-trained rgb-based policy with depth information injection, 2024

    Xincheng Pang, Wenke Xia, Zhigang Wang, Bin Zhao, Di Hu, Dong Wang, and Xuelong Li. Depth helps: Improving pre-trained rgb-based policy with depth information injection, 2024. 1

    2024

  48. [56]

    Sharpdepth: Sharpening metric depth predictions using diffusion distillation

    Duc-Hai Pham, Tung Do, Phong Nguyen, Binh-Son Hua, Khoi Nguyen, and Rang Nguyen. Sharpdepth: Sharpening metric depth predictions using diffusion distillation. arXiv preprint arXiv:2411.18229, 2024. 3

    2024 arXiv

  49. [57]

    Deep learning-based depth estimation methods from monocular image and videos: A comprehensive survey

    Uchitha Rajapaksha, Ferdous Sohel, Hamid Laga, Dean Diepeveen, and Mohammed Bennamoun. Deep learning-based depth estimation methods from monocular image and videos: A comprehensive survey. ACM Computing Surveys, 56(12):1–51, 2024. 3

    2024

  50. [58]

    Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer

    René Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, and Vladlen Koltun. Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(3), 2022. 2, 7, 9

    2022

  51. [59]

    Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer

    René Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, and Vladlen Koltun. Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(3), 2022. 3

    2022

  52. [60]

    High-resolution image synthesis with latent diffusion models, 2021

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models, 2021. 3

    2021

  53. [61]

    Self-supervised monocular depth estimation: Let’s talk about the weather

    Kieran Saunders, George V ogiatzis, and Luis J Manso. Self-supervised monocular depth estimation: Let’s talk about the weather. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8907–8917, 2023. 2, 3, 6, 7, 8

    2023

  54. [62]

    Learning depth from single monocular images

    Ashutosh Saxena, Sung Chung, and Andrew Ng. Learning depth from single monocular images. Advances in neural information processing systems, 18, 2005. 3 13

    2005

  55. [63]

    Monocular depth estimation using diffusion models

    Saurabh Saxena, Abhishek Kar, Mohammad Norouzi, and David J Fleet. Monocular depth estimation using diffusion models. arXiv preprint arXiv:2302.14816, 2023. 3

    2023 arXiv

  56. [64]

    Schönberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger

    Thomas Schöps, Johannes L. Schönberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger. A multi-view stereo benchmark with high-resolution images and multi- camera videos. In Conference on Computer Vision and Pattern Recognition (CVPR),...

    2017

  57. [65]

    Realmdreamer: Text-driven 3d scene generation with inpainting and depth diffusion

    Jaidev Shriram, Alex Trevithick, Lingjie Liu, and Ravi Ramamoorthi. Realmdreamer: Text-driven 3d scene generation with inpainting and depth diffusion. 2025. 1

    2025

  58. [66]

    Depthmaster: Taming diffusion models for monocular depth estimation

    Ziyang Song, Zerong Wang, Bo Li, Hao Zhang, Ruijie Zhu, Li Liu, Peng-Tao Jiang, and Tianzhu Zhang. Depthmaster: Taming diffusion models for monocular depth estimation. arXiv preprint arXiv:2501.02576,

    arXiv

  59. [67]

    Ec-depth: Exploring the consistency of self-supervised monocular depth estimation under challenging scenes

    Ziyang Song, Ruijie Zhu, Chuxin Wang, Jiacheng Deng, Jianfeng He, and Tianzhu Zhang. Ec-depth: Exploring the consistency of self-supervised monocular depth estimation under challenging scenes. arXiv preprint arXiv:2310.08044, 2023. 6, 7, 8, 17

    arXiv

  60. [68]

    Defeat-net: General monocular depth via simulta- neous unsupervised representation learning

    Jaime Spencer, Richard Bowden, and Simon Hadfield. Defeat-net: General monocular depth via simulta- neous unsupervised representation learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020. 3

    2020

  61. [69]

    Corrmatch: Label propagation via correlation matching for semi-supervised semantic segmentation

    Boyuan Sun, Yuqi Yang, Le Zhang, Ming-Ming Cheng, and Qibin Hou. Corrmatch: Label propagation via correlation matching for semi-supervised semantic segmentation. IEEE Computer Vision and Pattern Recognition (CVPR), 2024. 3, 9

    2024

  62. [70]

    Llava-scissor: Token compression with semantic connected components for video llms

    Boyuan Sun, Jiaxing Zhao, Xihan Wei, and Qibin Hou. Llava-scissor: Token compression with semantic connected components for video llms. arXiv preprint arXiv:2506.21862, 2025. 1

    2025 arXiv

  63. [71]

    Sc-depthv3: Robust self-supervised monocular depth estimation for dynamic scenes

    Libo Sun, Jia-Wang Bian, Huangying Zhan, Wei Yin, Ian Reid, and Chunhua Shen. Sc-depthv3: Robust self-supervised monocular depth estimation for dynamic scenes. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2023. 1

    2023

  64. [72]

    Depth estimation from image structure

    Antonio Torralba and Aude Oliva. Depth estimation from image structure. IEEE Transactions on pattern analysis and machine intelligence, 24(9):1226–1238, 2002. 3

    2002

  65. [73]

    Diffusion models for monocular depth estimation: Overcoming challenging conditions

    Fabio Tosi, Pierluigi Zama Ramirez, and Matteo Poggi. Diffusion models for monocular depth estimation: Overcoming challenging conditions. In European Conference on Computer Vision (ECCV), 2024. 3

    2024

  66. [74]

    Unsupervised monocular depth estimation for night-time images using adversarial domain feature adaptation

    Madhu Vankadari, Sourav Garg, Anima Majumder, Swagat Kumar, and Ardhendu Behera. Unsupervised monocular depth estimation for night-time images using adversarial domain feature adaptation. In Com- puter Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020...

    2020

  67. [75]

    When the sun goes down: Repairing photometric losses for all-day depth estimation

    Madhu Vankadari, Stuart Golodetz, Sourav Garg, Sangyun Shin, Andrew Markham, and Niki Trigoni. When the sun goes down: Repairing photometric losses for all-day depth estimation. arXiv preprint arXiv:2206.13850, 2022. 3

    2022 arXiv

  68. [76]

    Dai, Andrea F

    Igor Vasiljevic, Nick Kolkin, Shanyi Zhang, Ruotian Luo, Haochen Wang, Falcon Z. Dai, Andrea F. Daniele, Mohammadreza Mostajabi, Steven Basart, Matthew R. Walter, and Gregory Shakhnarovich. DIODE: A Dense Indoor and Outdoor DEpth Dataset. CoRR, abs/1908.00463, 2019. 7, 17, 18

    1908 arXiv

  69. [77]

    Vggt: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. 1

    2025

  70. [78]

    Weatherdepth: Curriculum contrastive learning for self-supervised depth estimation under adverse weather conditions

    Jiyuan Wang, Chunyu Lin, Lang Nie, Shujun Huang, Yao Zhao, Xing Pan, and Rui Ai. Weatherdepth: Curriculum contrastive learning for self-supervised depth estimation under adverse weather conditions. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages ...

    2024

  71. [79]

    Digging into contrastive learning for robust depth estimation with diffusion models

    Jiyuan Wang, Chunyu Lin, Lang Nie, Kang Liao, Shuwei Shao, and Yao Zhao. Digging into contrastive learning for robust depth estimation with diffusion models. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 4129–4137, 2024. 3 14

    2024

  72. [80]

    Regularizing nighttime weirdness: Efficient self-supervised monocular depth estimation in the dark

    Kun Wang, Zhenyu Zhang, Zhiqiang Yan, Xiang Li, Baobei Xu, Jun Li, and Jian Yang. Regularizing nighttime weirdness: Efficient self-supervised monocular depth estimation in the dark. In Proceedings of the IEEE/CVF international conference on computer vision, pages 16055–16064, 2021. 3

    2021

  73. [81]

    Depth-conditioned dynamic message propagation for monocular 3d object detection

    Li Wang, Liang Du, Xiaoqing Ye, Yanwei Fu, Guodong Guo, Xiangyang Xue, Jianfeng Feng, and Li Zhang. Depth-conditioned dynamic message propagation for monocular 3d object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 454–...

    2021

  74. [82]

    Tacodepth: Towards efficient radar-camera depth estimation with one-stage fusion

    Yiran Wang, Jiaqi Li, Chaoyi Hong, Ruibo Li, Liusheng Sun, Xiao Song, Zhe Wang, Zhiguo Cao, and Guosheng Lin. Tacodepth: Towards efficient radar-camera depth estimation with one-stage fusion. In CVPR, 2025. 3

    2025

  75. [83]

    Realtime depth estimation and obstacle detection from monocular video

    Andreas Wedel, Uwe Franke, Jens Klappstein, Thomas Brox, and Daniel Cremers. Realtime depth estimation and obstacle detection from monocular video. In Joint Pattern Recognition Symposium, pages 475–484. Springer, 2006. 3

    2006

  76. [84]

    Toward practical monocular indoor depth estimation

    Cho-Ying Wu, Jialiang Wang, Michael Hall, Ulrich Neumann, and Shuochen Su. Toward practical monocular indoor depth estimation. In CVPR, 2022. 3

    2022

  77. [85]

    Structure-guided ranking loss for single image depth prediction

    Ke Xian, Jianming Zhang, Oliver Wang, Long Mai, Zhe Lin, and Zhiguo Cao. Structure-guided ranking loss for single image depth prediction. In The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020. 6

    2020

  78. [86]

    Depthsplat: Connecting gaussian splatting and depth

    Haofei Xu, Songyou Peng, Fangjinhua Wang, Hermann Blum, Daniel Barath, Andreas Geiger, and Marc Pollefeys. Depthsplat: Connecting gaussian splatting and depth. In CVPR, 2025. 3

    2025

  79. [87]

    Synthetic-to-real self-supervised robust depth estimation via learning with motion and structure priors

    Weilong Yan, Ming Li, Haipeng Li, Shuwei Shao, and Robby T Tan. Synthetic-to-real self-supervised robust depth estimation via learning with motion and structure priors. arXiv preprint arXiv:2503.20211,

    arXiv

  80. [88]

    Drivingstereo: A large-scale dataset for stereo matching in autonomous driving scenarios

    Guorun Yang, Xiao Song, Chaoqin Huang, Zhidong Deng, Jianping Shi, and Bolei Zhou. Drivingstereo: A large-scale dataset for stereo matching in autonomous driving scenarios. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 7, 17

    2019

  81. [89]

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

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 1, 2, 3, 5, 6...

    2024

  82. [90]

    Depth anything v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2. Advances in Neural Information Processing Systems, 37:21875–21911, 2024. 1, 2, 3, 6, 7, 8, 9, 10, 17, 20, 21

    2024

  83. [91]

    Revisiting weak-to-strong consistency in semi-supervised semantic segmentation

    Lihe Yang, Lei Qi, Litong Feng, Wayne Zhang, and Yinghuan Shi. Revisiting weak-to-strong consistency in semi-supervised semantic segmentation. In CVPR, 2023. 3, 9

    2023

  84. [92]

    Dformer: Rethinking rgbd representation learning for semantic segmentation

    Bowen Yin, Xuying Zhang, Zhong-Yu Li, Li Liu, Ming-Ming Cheng, and Qibin Hou. Dformer: Rethinking rgbd representation learning for semantic segmentation. In ICLR, 2024. 1

    2024

  85. [93]

    Dformerv2: Geometry self-attention for rgbd semantic segmentation

    Bo-Wen Yin, Jiao-Long Cao, Ming-Ming Cheng, and Qibin Hou. Dformerv2: Geometry self-attention for rgbd semantic segmentation. In CVPR, 2025. 1

    2025

  86. [94]

    Metric3d: Towards zero-shot metric 3d prediction from a single image

    Wei Yin, Chi Zhang, Hao Chen, Zhipeng Cai, Gang Yu, Kaixuan Wang, Xiaozhi Chen, and Chunhua Shen. Metric3d: Towards zero-shot metric 3d prediction from a single image. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9043–9053, 2023. 3

    2023

  87. [95]

    Towards accurate reconstruction of 3d scene shape from a single monocular image

    Wei Yin, Jianming Zhang, Oliver Wang, Simon Niklaus, Simon Chen, Yifan Liu, and Chunhua Shen. Towards accurate reconstruction of 3d scene shape from a single monocular image. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2022. 1

    2022

  88. [96]

    Learning to recover 3d scene shape from a single image

    Wei Yin, Jianming Zhang, Oliver Wang, Simon Niklaus, Long Mai, Simon Chen, and Chunhua Shen. Learning to recover 3d scene shape from a single image. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn. (CVPR), 2021. 1

    2021

  89. [97]

    Ms-nerf: Multi-space neural radiance fields

    Ze-Xin Yin, Peng-Yi Jiao, Jiaxiong Qiu, Ming-Ming Cheng, and Bo Ren. Ms-nerf: Multi-space neural radiance fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 1–18, 2025. 1 15

    2025

  90. [98]

    Adding conditional control to text-to-image diffusion models, 2023

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models, 2023. 1

    2023

  91. [99]

    Towards scale-aware, robust, and generalizable unsupervised monocular depth estimation by integrating imu motion dynamics

    Sen Zhang, Jing Zhang, and Dacheng Tao. Towards scale-aware, robust, and generalizable unsupervised monocular depth estimation by integrating imu motion dynamics. In European Conference on Computer Vision, pages 143–160. Springer, 2022. 3, 6, 7, 8

    2022

  92. [100]

    Pattern-affinitive propagation across depth, surface normal and semantic segmentation

    Zhenyu Zhang, Zhen Cui, Chunyan Xu, Yan Yan, Nicu Sebe, and Jian Yang. Pattern-affinitive propagation across depth, surface normal and semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4106–4115, 2019. 1

    2019

  93. [101]

    Monocular depth estimation based on deep learning: An overview

    Chaoqiang Zhao, Qiyu Sun, Chongzhen Zhang, Yang Tang, and Feng Qian. Monocular depth estimation based on deep learning: An overview. Science China Technological Sciences, 63(9):1612–1627, 2020. 3

    2020

  94. [102]

    Unsupervised monocular depth estimation in highly complex environments

    Chaoqiang Zhao, Yang Tang, and Qiyu Sun. Unsupervised monocular depth estimation in highly complex environments. IEEE Transactions on Emerging Topics in Computational Intelligence, 6(5):1237–1246,

  95. [103]

    Llava-octopus: Unlocking instruction- driven adaptive projector fusion for video understanding

    Jiaxing Zhao, Boyuan Sun, Xiang Chen, Xihan Wei, and Qibin Hou. Llava-octopus: Unlocking instruction- driven adaptive projector fusion for video understanding. arXiv preprint arXiv:2501.05067, 2025. 1

    2025 arXiv

  96. [104]

    Steps: Joint self-supervised nighttime image enhancement and depth estimation

    Yupeng Zheng, Chengliang Zhong, Pengfei Li, Huan-ang Gao, Yuhang Zheng, Bu Jin, Ling Wang, Hao Zhao, Guyue Zhou, Qichao Zhang, et al. Steps: Joint self-supervised nighttime image enhancement and depth estimation. In 2023 IEEE International Conference on Robotics and Automation...

    2023

  97. [105]

    Scene parsing through ade20k dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing through ade20k dataset. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 633–641, 2017. 6

    2017

  98. [106]

    Semantic understanding of scenes through the ade20k dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Semantic understanding of scenes through the ade20k dataset. International Journal of Computer Vision, 127:302–321, 2019. 6

    2019

  99. [107]

    Sintel Final

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In Computer Vision (ICCV), 2017 IEEE International Conference on, 2017. 3 16 Appendix In order to provide a more complete illustration...

    2017

Pith tools

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