Pith. sign in

REVIEW 2 major objections 6 minor 2 cited by

ER-LoRA: Effective-Rank Guided Adaptation for Weather-Generalized Depth Estimation

T0 review · 2 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read ER-LoRA claims that a low-rank adapter trained on daytime images alone lets a frozen vision foundation model generalize zero-shot to night, rain, fog, and snow depth estimation, outperforming full fine-tuning, prior PEFT methods, synthetic-

desk verdict ER-LoRA's zero-shot weather results look real, but the method quietly needs a full fine-tuning run to pick its directions, so the FFT comparisons are partly self-referential and the 'no full retraining' framing is misleading. read the letter →

arxiv 2509.00665 v2 pith:3L4V7X7Q submitted 2025-08-31 cs.CV cs.RO

Weilong Yan , Xin Zhang , Robby T. Tan This is my paper
classification cs.CVcs.RO
keywords monoculardepthestimationadverseweatherparameter-efficientfine-tuningvisionfoundationmodelslow-rankadaptationeffectiveranksingularvaluedecompositionzero-shotdomaingeneralization
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

This paper tries to establish that weather-generalized monocular depth estimation can be achieved without any synthetic adverse-weather data, by parameter-efficient fine-tuning (PEFT) of a frozen vision foundation model on a small set of clean daytime images. The proposed method, ER-LoRA, uses a Selecting-Tuning-Maintaining strategy: per-layer ranks are set from the entropy rank of the frozen weight matrices, the low-rank branch is initialized along singular directions that change most under full fine-tuning, and a regularization term protects the leading stable-rank directions. If the claim holds, robust depth under night, rain, fog, and snow reduces from a data-collection problem into a lightweight adaptation problem, and prior methods that rely on synthetic adverse data or full-backbone retraining are unnecessary. The paper reports zero-shot gains over full fine-tuning, prior PEFT methods, synthetic-data RMDE methods, and a depth foundation model across four real-world benchmarks.

What carries the argument

Two effective ranks of a weight matrix drive the design: the entropy rank (the exponential of the entropy of normalized singular values, capturing how evenly information is spread over singular directions) and the stable rank (the sum of squared normalized singular values, capturing how concentrated energy is in leading directions). The Selecting stage sets each layer's low-rank budget r as a scaled entropy rank and picks initialization directions by projecting the full-tuned residual weight onto the frozen weight's singular vectors; the Maintaining stage adds a regularization that penalizes changes along the leading stable-rank directions unless they coincide with selected task directions.

What would settle it

Compute the residual W' - W from a full fine-tuning run on the same 15K daytime images, then replace the selected top-r directions with r random singular directions of the frozen weights (same rank budget). If zero-shot AbsRel on nuScenes-night and nuScenes-rain does not materially worsen, the claim that task-aware direction selection drives the gains is falsified. Alternatively, measure the compute of the oracle fine-tuning run; if oracle + ER-LoRA training exceeds full fine-tuning cost, the parameter-efficiency claim needs qualification.

Watch

Extended reading notes

Core claim

The central claim is that a frozen DINOv2 visual encoder, adapted through a small trainable low-rank branch (about 8.7M parameters) using only about 15K daytime nuScenes images, generalizes to unseen night, rain, fog, and snow at zero shot, outperforming full fine-tuning, existing PEFT methods such as LoRA and SoMA, synthetic-adverse-data methods such as md4all, and the depth foundation model Depth Anything V2. The mechanism is the Selecting-Tuning-Maintaining (STM) strategy: per-layer rank budgets come from the entropy rank of each frozen weight matrix; the low-rank branch is initialized on the top singular directions of the residual between the frozen weights and a full-tuned weight; and a

Load-bearing premise

The method assumes that a full fine-tuned model on the same daytime training set is available to compute the residual whose top singular directions seed the low-rank branch; the paper never states how that full-tuned model is obtained or what it costs, and if those directions do not align with what is needed under night, fog, rain, and snow, the Selecting stage loses its value.

Editorial extensions

If this is right

  • If correct, adverse-weather robustness does not require collecting or synthesizing degraded imagery; a handful of clean daytime images plus a frozen foundation model suffices.
  • PEFT methods that target semantic tasks can be re-designed for geometry-centric tasks by replacing uniform rank allocation and random initialization with spectrum-guided rank scheduling and direction-aware initialization.
  • The stable-rank regularization offers a general recipe for retaining pretrained generalization while allowing task-specific low-rank updates, applicable beyond depth.
  • The reported gains (average AbsRel +7.3% vs full fine-tuning, +6.4% vs prior PEFT) position low-rank adaptation as a viable alternative to full retraining for domain-generalized dense prediction.
  • Supervised and self-supervised depth pipelines can share the same adaptation strategy, since the STM branch is agnostic to the training loss.

Reading between the lines

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

  • Editorial inference: the Selecting stage quietly depends on a full fine-tuning run to obtain the residual weight W' - W; if that oracle run is counted in total compute, the efficiency claim may shrink, and the method is best understood as a way to compress the knowledge of a full-tuned model into a low-rank branch rather than a pure parameter-efficient alternative.
  • Editorial inference: the direction-selection criterion (largest absolute projection of the FFT residual) favors directions that changed most on daytime data; whether those are the directions needed under nighttime photometric distortion is an assumption that the ablation table (Task_dir rows) supports only indirectly.
  • Editorial inference: a natural testable extension is to evaluate ER-LoRA on other geometric tasks where VFM semantic bias is weak, such as surface normal estimation or optical flow, or on other distribution shifts (dust, glare, underwater); the method predicts that its rank/regularization schedule transfers.
  • Editorial inference: because the pseudo-labels in the supervised pipeline come from Depth Anything V2, the comparison against that foundation model is partly a comparison against a teacher; an ablation with pseudo-labels removed or with a weaker teacher would isolate the contribution of the STM branch.
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

2 major / 6 minor

Summary. The paper proposes ER-LoRA, a parameter-efficient fine-tuning (PEFT) method for weather-generalized monocular depth estimation. Starting from a frozen DINOv2-L backbone, it decomposes the pretrained weights via SVD, allocates per-layer LoRA ranks using an entropy-based effective rank (Eq. 9), selects task-aware singular directions by projecting the full-tuned residual ΔW = W' - W onto the frozen singular vectors (Eqs. 10-11), initializes a low-rank branch from the selected components (Eqs. 12-14), and applies a stable-rank-based regularization to preserve principal directions (Eq. 15). Training uses only daytime nuScenes images with self-supervised photometric losses and, in some settings, LiDAR and Depth Anything V2 pseudo-label supervision. The paper reports zero-shot improvements over full fine-tuning (FFT), prior PEFT methods, synthetic-data RMDE methods, and Depth Anything V2 on night, rain, fog, and snow benchmarks.

Significance. If the empirical claims hold without hidden subsidies, the paper would make a strong case that weather generalization in depth estimation can be achieved by adapting a vision foundation model on normal daytime data, avoiding synthetic adverse data. The evaluation is genuinely out-of-domain, and the average AbsRel improvements cited in the abstract are approximately consistent with Tables 1-3. The effective-rank-guided rank allocation is a plausible and well-motivated idea, and the maintaining regularization is a sensible complement. However, two hidden dependencies substantially reduce the significance as stated: (1) the direction-selection step requires a converged full-tuned model W' to compute the residual, so the method is not an independent PEFT method and its FFT comparison is asymmetric; (2) training uses pseudo-labels from Depth Anything V2, which is itself a depth foundation model, making the comparison to that baseline in Table 3 difficult to interpret. These issues are fixable with disclosure and additional ablations, but they are load-bearing for the paper's central claims.

major comments (2)
  1. [Sec. 4.1-4.2, Eqs. (10)-(11)] The direction-selection step requires the full-tuned weight W' via ΔW = W' - W. The paper never states how W' is obtained, whether the FFT baselines in Tables 1-3 double as this oracle, or what the additional full fine-tuning run costs. This is load-bearing for the central claim: ER-LoRA is not a standalone PEFT method; it consumes the solution of a converged FFT model on the same training data, and the comparison to FFT is asymmetric. The ablation in Table 4 shows that removing Task_dir costs only ~0.25-0.35 AbsRel on nuScenes night/rain, suggesting the headline may survive without the oracle, but the method's framing as Selecting-Tuning-Maintaining and its claimed advantage over prior PEFT rest on this step. Please disclose the FFT run, report its compute cost, and provide an ablation without the FFT-based directions (e.g., random or spectral initialization).
  2. [Sec. 4.3, Eqs. (16)-(17), Table 3] The training objectives incorporate pseudo-labels from Depth Anything V2 in both the self-supervised and supervised settings, yet Table 3 compares against Depth Anything V2 as a baseline. This is a hidden subsidy: the method is partially a distillation of a 62M-data depth foundation model. The exact objective used for the self-supervised results in Tables 1-2 is also ambiguous: Eq. (16) lists only L_pe + L_smooth + L_reg, but the text immediately adds DA v2 pseudo-labels and a LiDAR loss; it is unclear which terms are active. Please state the exact objective for each table and add an ablation without DA v2 pseudo-labels to quantify the contribution of this external supervision.
minor comments (6)
  1. [Lemma 1] The proof of Lemma 1 is deferred to an appendix, but no appendix is included in the manuscript. Please include the proof or state that it is standard.
  2. [Sec. 4.2, Eq. (14)] The notation W0 is confusing: W0 denotes the frozen base, and at initialization W0 + BA = W. Clarify that W0 is the reparameterized frozen weight, not the original pretrained weight.
  3. [Sec. 5.4, Table 4] The ablation differences are small (0.25-0.35 AbsRel, about 1.5-2.7% relative). Report multiple seeds or a statistical test to establish that the Task_dir and Rankst gains are not within noise.
  4. [Sec. 5.2, Table 2] The section title and text call the evaluations 'zero-shot', but the RobotCar rows are marked as trained on corresponding training data. Qualify the language accordingly.
  5. [Sec. 5.3] The phrase 'reduces the average AbsRel of 5.5%' should read 'by 5.5%' and should specify whether the improvement is relative or absolute.
  6. [Sec. 5.2 references] Reference [29] (STPrivacy) does not appear to be a robust depth estimation method; check the citation in the list of robust-depth baselines.

Circularity Check

1 steps flagged · score 5.0 of 10

ER-LoRA's 'task-aware direction' selection is a hidden full-fine-tuning oracle, making the FFT comparison asymmetric, though the zero-shot evaluation itself is not fitted.

  1. fitted input called prediction [Sec. 4.1, Eq. 10–11; Sec. 4.2, Eqs. 12–14]
    "we step further to find how weight matrices change from Wq to W′q after full fine-tuning (FFT) on a few daytime data. Here, the residual weight is defined as ∆Wq = W′q − Wq. ... d = |diag(U ⊤∆W V )| ∈ R K×1 ... we obtain directions with top r values in d as task-aware directions: I = Top-r(d)"

    The 'task-aware directions' that initialize ER-LoRA are constructed directly from the converged full-fine-tuned weight W′ on the same training set. Thus ER-LoRA is not an independent PEFT method: it requires a full fine-tuning run (the very FFT baseline it claims to outperform) as an oracle. The later comparisons to FFT in Tables 1–3 are asymmetric, because the low-rank branch is initialized in the directions of the FFT residual—the adaptation is a low-rank projection of the FFT solution. The cost and source of W′ are undisclosed, and the abstract's claim of avoiding 'full-backbone retraining' is misleading. The target-domain accuracy numbers remain genuinely zero-shot, so the circularity is partial rather than total.

full rationale

The zero-shot evaluation is out-of-domain: night, rain, fog, and snow are never used in training or hyperparameter selection, so the AbsRel improvements are not fitted to the test domains. The core circularity is methodological: Eq. 10 defines d using ΔW = W′ − W, where W′ is the full-tuned weight on the same daytime data, and Eq. 11 selects top-r directions from that residual. This is a fitted input used as the method's 'task-aware directions' without disclosing the required full fine-tuning run. However, Table 4 shows that removing Task_dir costs only a small fraction of the total gain (~0.25–0.35 AbsRel), so the central result does not entirely reduce to the oracle. There is no load-bearing self-citation or imported uniqueness theorem; the self-citations present (e.g., [65]) are contextual. Overall, the paper has a real hidden-dependency problem but not full logical circularity.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

The method's measurable contribution is a low-rank adapter on frozen DINOv2; everything else is borrowed. The central free parameters are α (rank scaling, unreported), γ, and the unstated loss weights, all hand-chosen rather than derived. The key axioms are standard linear algebra (Eckart-Young-Mirsky), trust that DINOv2's frozen directions remain robust under light perturbation, the photometric-consistency assumption on daytime video, and the new oracle assumption that the full-tuned residual's top projections identify directions that transfer to unseen weather. No invented entities.

free parameters (7)
  • alpha (rank scaling in Eq. 9) = unreported
    r = α × Rank_en(W); sets every per-layer LoRA rank budget. Value never stated; it is the main knob controlling trainable capacity (8.7M params reported but α not given).
  • gamma in effective-rank definitions = unreported
    Both Eq. 7 and Eq. 8 are written with a general exponent γ; Lemma 1 assumes γ=1, but the paper never states which γ the experiments use.
  • L_reg coefficient = implied 1.0
    Eq. 15 has no explicit weight; whether the strength was tuned (and on which data) is unstated.
  • velocity loss weight = unreported
    Sec. 5.2 says a 'weak velocity loss' is incorporated in self-supervised training, following [17, 52], but gives no formulation or weight.
  • self-supervised loss weights (L_pe, L_smooth) = unreported
    Eq. 16 sums photometric, smooth, and regularization losses with no coefficients reported.
  • supervised loss balance alpha1:alpha2 = 2:1
    Stated in Sec. 4.3 but hand-chosen.
  • Depth Anything V2 pseudo-label source = unreported variant/checkpoint
    Sec. 4.3 uses DA v2 pseudo-labels for dense supervision; the checkpoint family (small/base/large, metric or relative) is unspecified, affecting pseudo-label quality and comparison fairness.
assumptions (7)
  • standard math Eckart-Young-Mirsky: top-r singular components approximate the core of a matrix
    Invoked in Sec. 3.3 to justify SVD-based decomposition and low-rank reasoning for weight matrices.
  • domain assumption Entropy-rank and stable-rank measure useful information dispersion of weight matrices
    Secs. 3.3, 4.1 borrow [43, 41, 53] and assume these quantities identify how many directions to tune and which to preserve.
  • domain assumption Photometric consistency holds on daytime training sequences
    Sec. 3.1 follows Monodepth2 [18]; the self-supervised objective in Eq. 1 relies on this assumption being valid for the daytime data only.
  • domain assumption DINOv2 frozen representations transfer to depth if perturbed only along few directions
    Secs. 4.1-4.2 assume shallow layers should be barely touched and that low-rank tuning preserves robustness across weather domains.
  • ad hoc to paper Top-r absolute projections of the full-tuned residual onto frozen directions identify task-aware, transferable directions
    Eqs. 10-11 define the Selecting stage; this oracle assumption is the paper's own heuristic and is untested outside the reported settings.
  • domain assumption Depth Anything V2 pseudo-labels are reliable dense supervision on daytime data
    Sec. 4.3 adds normalized L1 loss against DA v2 predictions; the method's supervised results depend on the quality of this teacher.
  • domain assumption Scale-and-shift aligned evaluation is a meaningful comparison of predicted depth
    Sec. 5.2 follows MiDaS [5]; all tables, including DA v2 comparisons, use this alignment, which discards metric scale information.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ER-LoRA: Effective-Rank Guided Adaptation for Weather-Generalized Depth Estimation." pith.science (2026). https://pith.science/paper/3L4V7X7Q

@misc{pith2026250900665,
  author       = {Pith},
  title        = {Pith review of: ER-LoRA: Effective-Rank Guided Adaptation for Weather-Generalized Depth Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3L4V7X7Q}},
  note         = {Machine review of arXiv:2509.00665}
}
read the original abstract

Monocular depth estimation under adverse weather conditions (e.g.\ rain, fog, snow, and nighttime) remains highly challenging due to the lack of reliable ground truth and the difficulty of learning from unlabeled real-world data. Existing methods often rely on synthetic adverse data with pseudo-labels, which suffer from domain gaps, or employ self-supervised learning, which violates photometric assumptions in adverse scenarios. In this work, we propose to achieve weather-generalized depth estimation by Parameter-Efficient Fine-Tuning (PEFT) of Vision Foundation Models (VFMs), using only a small amount of high-visibility (normal) data. While PEFT has shown strong performance in semantic tasks such as segmentation, it remains underexplored for geometry -- centric tasks like depth estimation -- especially in terms of balancing effective adaptation with the preservation of pretrained knowledge. To this end, we introduce the Selecting-Tuning-Maintaining (STM) strategy, which structurally decomposes the pretrained weights of VFMs based on two kinds of effective ranks (entropy-rank and stable-rank). In the tuning phase, we adaptively select the proper rank number as well as the task-aware singular directions for initialization, based on the entropy-rank and full-tuned weight; while in the maintaining stage, we enforce a principal direction regularization based on the stable-rank. This design guarantees flexible task adaptation while preserving the strong generalization capability of the pretrained VFM. Extensive experiments on four real-world benchmarks across diverse weather conditions demonstrate that STM not only outperforms existing PEFT methods and full fine-tuning but also surpasses methods trained with adverse synthetic data, and even the depth foundation model

Figures

Figures reproduced from arXiv: 2509.00665 by the authors.

Figure 1
Figure 1. (a) Row 1: Distribution of singular values within VFM’s frozen weights. Row 2: Distri [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The ER-LoRA pipeline. In the Selecting phase, we adaptively allocate proper rank numbers [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Qualitative Results under extreme conditions. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Actor as Its Own Critic: Unifying Region Understanding and Localization via CycleGRPO

    cs.CV 2026-07 accept novelty 7.0 of 10

    A single MLLM jointly improves region captioning and localization by rewarding captions that let it reconstruct the original mask, needing only region inputs.

  2. GlowGS: Generative Semantic Feature Learning for 3D Gaussian Splatting in Nighttime Glow Scenes

    cs.CV 2026-05 unverdicted novelty 5.0 of 10

    GlowGS improves 3D Gaussian Splatting in nighttime glow scenes via semantic feature generation from diffusion models and novel-view semantic learning with vision foundation models.

Reference graph

Works this paper leans on

88 extracted references · 61 canonical work pages · cited by 2 Pith papers

  1. [1]

    Domain-adaptive 2d human pose estimation via dual teachers in extremely low-light conditions

    Yihao Ai, Yifei Qi, Bo Wang, Yu Cheng, Xinchao Wang, and Robby T Tan. Domain-adaptive 2d human pose estimation via dual teachers in extremely low-light conditions. InEuropean Conference on Computer Vision, pages 221–239. Springer, 2024

    2024

  2. [2]

    Zoedepth: Zero-shot transfer by combining relative and metric depth, 2023

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

    2023

  3. [3]

    Learning frequency-adapted vision foundation model for domain generalized semantic segmentation

    Qi Bi, Jingjun Yi, Hao Zheng, Haolan Zhan, Yawen Huang, Wei Ji, Yuexiang Li, and Yefeng Zheng. Learning frequency-adapted vision foundation model for domain generalized semantic segmentation. In Advances in Neural Information Processing Systems (NeurIPS), volume 37, pages 94047–94072, 2024

    2024

  4. [4]

    Unsupervised scale-consistent depth learning from video.International Journal of Computer Vision (IJCV), 2021

    Jia-Wang Bian, Huangying Zhan, Naiyan Wang, Zhichao Li, Le Zhang, Chunhua Shen, Ming-Ming Cheng, and Ian Reid. Unsupervised scale-consistent depth learning from video.International Journal of Computer Vision (IJCV), 2021

    2021

  5. [5]

    Midas v3.1 – a model zoo for robust monocular relative depth estimation.arXiv preprint arXiv:2307.14460, 2023

    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

    arXiv 2023

  6. [6]

    Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom

    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. arXiv preprint arXiv:1903.11027, 2019

    arXiv 1903

  7. [7]

    Xiao Cao, Beibei Lin, Bo Wang, Zhiyong Huang, and Robby T. Tan. Ssnerf: Sparse view semi-supervised neural radiance fields with augmentation.arXiv preprint arXiv:2408.09144, 2024

    work page Pith review arXiv 2024

  8. [8]

    Xiao Cao, Beibei Lin, Bo Wang, Zhiyong Huang, and Robby T. Tan. 3dswapping: Texture swapping for 3d object from single reference image.arXiv preprint arXiv:2503.18853, 2025

    work page arXiv 2025

Show all 88 references
  1. [9]

    On the robustness of language guidance for low-level vision tasks: Findings from depth estimation

    Agneet Chatterjee, Tejas Gokhale, Chitta Baral, and Yezhou Yang. On the robustness of language guidance for low-level vision tasks: Findings from depth estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2794–2803, June 2024

    2024

  2. [10]

    Adapt- former: Adapting vision transformers for scalable visual recognition.arXiv preprint arXiv:2205.13535, 2022

    Shoufa Chen, Chongjian Ge, Zhan Tong, Jiangliu Wang, Yibing Song, Jue Wang, and Ping Luo. Adapt- former: Adapting vision transformers for scalable visual recognition.arXiv preprint arXiv:2205.13535, 2022

    2022 arXiv

  3. [11]

    The approximation of one matrix by another of lower rank.Psychometrika, 1(3):211–218, 1936

    Carl Eckart and Gale Young. The approximation of one matrix by another of lower rank.Psychometrika, 1(3):211–218, 1936

    1936

  4. [12]

    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. InProceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2, NIPS’14, page 2366–2374, Cambridge, MA, USA, 2014....

    2014

  5. [13]

    Make lora great again: Boosting lora with adaptive singular values and mixture-of-experts optimization alignment, 2025

    Chenghao Fan, Zhenyi Lu, Sichen Liu, Xiaoye Qu, Wei Wei, Chengfeng Gu, and Yu Cheng. Make lora great again: Boosting lora with adaptive singular values and mixture-of-experts optimization alignment, 2025

    2025

  6. [14]

    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, pages xxxxx–xxxxx, 2025

    2025

  7. [15]

    Eva: Exploring the limits of masked visual representation learning at scale.arXiv preprint arXiv:2211.07636, 2022

    Yuxin Fang, Wen Wang, Binhui Xie, Quan Sun, Ledell Wu, Xinggang Wang, Tiejun Huang, Xinlong Wang, and Yue Cao. Eva: Exploring the limits of masked visual representation learning at scale.arXiv preprint arXiv:2211.07636, 2022. 10

    2022 arXiv

  8. [16]

    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 European Conference on Computer Vision, pages 241–258. Springer, 2024

    2024

  9. [17]

    Robust monocular depth estimation under challenging conditions

    Stefano Gasperini, Nils Morbitzer, HyunJun Jung, Nassir Navab, and Federico Tombari. Robust monocular depth estimation under challenging conditions. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 8177–8186, 2023

    2023

  10. [18]

    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. InProceedings of the IEEE/CVF international conference on computer vision, pages 3828–3838, 2019

    2019

  11. [19]

    Clément Godard, Oisin Mac Aodha, and Gabriel J. Brostow. Unsupervised monocular depth estimation with left-right consistency. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6602–6611, 2017

    2017

  12. [20]

    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

    2024

  13. [21]

    Masked autoencoders are scalable vision learners.arXiv:2111.06377, 2021

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners.arXiv:2111.06377, 2021

    2021 arXiv

  14. [22]

    Delving deep into rectifiers: Surpassing human-level performance on imagenet classification, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification, 2015

    2015

  15. [23]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. InInternational Conference on Learning Representations, 2022

    2022

  16. [24]

    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 Analysis...

    2024

  17. [25]

    Visual prompt tuning, 2022

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning, 2022

    2022

  18. [26]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  19. [27]

    Deeper depth prediction with fully convolutional residual networks

    Iro Laina, Christian Rupprecht, Vasileios Belagiannis, Federico Tombari, and Nassir Navab. Deeper depth prediction with fully convolutional residual networks. In2016 Fourth International Conference on 3D Vision (3DV), pages 239–248, 2016

    2016

  20. [28]

    The power of scale for parameter-efficient prompt tuning, 2021

    Brian Lester, Rami Al-Rfou, and Noah Constant. The power of scale for parameter-efficient prompt tuning, 2021

    2021

  21. [29]

    Stprivacy: Spatio-temporal privacy-preserving action recognition

    Ming Li, Xiangyu Xu, Hehe Fan, Pan Zhou, Jun Liu, Jia-Wei Liu, Jiahe Li, Jussi Keppo, Mike Zheng Shou, and Shuicheng Yan. Stprivacy: Spatio-temporal privacy-preserving action recognition. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 5...

    2023

  22. [30]

    Depthformer: Exploiting long-range correlation and local information for accurate monocular depth estimation.arXiv preprint arXiv:2203.14211, 2022

    Zhenyu Li, Zehui Chen, Xianming Liu, and Junjun Jiang. Depthformer: Exploiting long-range correlation and local information for accurate monocular depth estimation.arXiv preprint arXiv:2203.14211, 2022

    2022 arXiv

  23. [31]

    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. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12737–12746, 2021

    2021

  24. [32]

    Dora: Weight-decomposed low-rank adaptation.arXiv preprint arXiv:2402.09353, 2024

    Shih-Yang Liu, Chien-Yi Wang, Hongxu Yin, Pavlo Molchanov, Yu-Chiang Frank Wang, Kwang- Ting Cheng, and Min-Hung Chen. Dora: Weight-decomposed low-rank adaptation.arXiv preprint arXiv:2402.09353, 2024

    2024 arXiv

  25. [33]

    Maddern, G

    W. Maddern, G. Pascoe, C. Linegar, and P. Newman. 1 year, 1000 km: The oxford robotcar dataset. International Journal of Robotics Research, page 0278364916679498, 2016

    2016

  26. [34]

    Pissa: Principal singular values and singular vectors adaptation of large language models.arXiv preprint arXiv:2404.02948, 2024

    Fanxu Meng, Zhaohui Wang, and Muhan Zhang. Pissa: Principal singular values and singular vectors adaptation of large language models.arXiv preprint arXiv:2404.02948, 2024. 11

    2024 arXiv

  27. [35]

    Maxime Oquab, Timothée Darcet, Theo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Russell Howes, Po-Yao Huang, Hu Xu, Vasu Sharma, Shang-Wen Li, Wojciech Galuba, Mike Rabbat, Mido Assran, Nicolas...

    2023

  28. [36]

    Ecodepth: Effective conditioning of diffusion models for monocular depth estimation

    Suraj Patni, Aradhye Agarwal, and Chetan Arora. Ecodepth: Effective conditioning of diffusion models for monocular depth estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 28285–28295, June 2024

    2024

  29. [37]

    Excavating the potential capacity of self-supervised monocular depth estimation

    Rui Peng, Ronggang Wang, Yawen Lai, Luyang Tang, and Yangang Cai. Excavating the potential capacity of self-supervised monocular depth estimation. InProceedings of the IEEE International Conference on Computer Vision (ICCV), 2021

    2021

  30. [38]

    Canadian adverse driving conditions dataset.The International Journal of Robotics Research, 2021

    Matthew Pitropov, Danson Evan Garcia, Jason Rebello, Michael Smart, Carlos Wang, Krzysztof Czarnecki, and Steven Lake Waslander. Canadian adverse driving conditions dataset.The International Journal of Robotics Research, 2021

    2021

  31. [39]

    Learning transferable visual models from natural language supervision.CoRR, abs/2103.00020, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision.CoRR, abs/2103.00020, 2021

    2021 arXiv

  32. [40]

    Vision transformers for dense prediction.ArXiv preprint, 2021

    René Ranftl, Alexey Bochkovskiy, and Vladlen Koltun. Vision transformers for dense prediction.ArXiv preprint, 2021

    2021

  33. [41]

    Benjamin Recht, Maryam Fazel, and Pablo A. Parrilo. Guaranteed minimum-rank solutions of linear matrix equations via nuclear norm minimization.SIAM Review, 52(3):471–501, January 2010

    2010

  34. [42]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10684–10695, June 2022

    2022

  35. [43]

    The effective rank: A measure of effective dimensionality

    Olivier Roy and Martin Vetterli. The effective rank: A measure of effective dimensionality. In2007 15th European Signal Processing Conference, pages 606–610, 2007

    2007

  36. [44]

    Sampling from large matrices: an approach through geometric functional analysis, 2006

    Mark Rudelson and Roman Vershynin. Sampling from large matrices: an approach through geometric functional analysis, 2006

    2006

  37. [45]

    Kieran Saunders, George V ogiatzis, and Luis J. Manso. Self-supervised monocular depth estimation: Let’s talk about the weather. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 8907–8917, October 2023

    2023

  38. [46]

    Mgnet: Monocular geometric scene understand- ing for autonomous driving

    Markus Schön, Michael Buchholz, and Klaus Dietmayer. Mgnet: Monocular geometric scene understand- ing for autonomous driving. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 15804–15815, 2021

    2021

  39. [47]

    Iebins: Iterative elastic bins for monocular depth estimation

    Shuwei Shao, Zhongcai Pei, Xingming Wu, Zhong Liu, Weihai Chen, and Zhengguo Li. Iebins: Iterative elastic bins for monocular depth estimation. InAdvances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  40. [48]

    Pvchat: Personalized video chat with one-shot learning.arXiv preprint arXiv:2503.17069, 2025

    Yufei Shi, Weilong Yan, Gang Xu, Yumeng Li, Yucheng Chen, Zhenxi Li, Fei Richard Yu, Ming Li, and Si Yong Yeo. Pvchat: Personalized video chat with one-shot learning.arXiv preprint arXiv:2503.17069, 2025

    2025 arXiv

  41. [49]

    Task-specific directions: Definition, exploration, and utilization in parameter efficient fine-tuning, 2025

    Chongjie Si, Zhiyi Shi, Shifan Zhang, Xiaokang Yang, Hanspeter Pfister, and Wei Shen. Task-specific directions: Definition, exploration, and utilization in parameter efficient fine-tuning, 2025

    2025

  42. [50]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14402–14413, 2020

    2020

  43. [51]

    Sc-depthv3: Robust self-supervised monocular depth estimation for dynamic scenes.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2023

    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

    2023

  44. [52]

    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. InEuropean Conference on Computer Vision (ECCV), 2024

    2024

  45. [53]

    Joel A. Tropp. An introduction to matrix concentration inequalities, 2015. 12

    2015

  46. [54]

    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. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, P...

    2020

  47. [55]

    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. InConference on Robot Learning, pages 1992–2003. PMLR, 2023

    1992

  48. [56]

    Milora: Harnessing minor singular components for parameter-efficient llm finetuning, 2024

    Hanqing Wang, Yixia Li, Shuo Wang, Guanhua Chen, and Yun Chen. Milora: Harnessing minor singular components for parameter-efficient llm finetuning, 2024

    2024

  49. [57]

    Scalable autoregressive monocular depth estimation.arXiv preprint arXiv:2411.11361, 2024

    Jinhong Wang, Jian Liu, Dongqi Tang, Weiqiang Wang, Wentong Li, Danny Chen, Jintai Chen, and Jian Wu. Scalable autoregressive monocular depth estimation.arXiv preprint arXiv:2411.11361, 2024

    2024 arXiv

  50. [58]

    Weatherdepth: Curriculum contrastive learning for self-supervised depth estimation under adverse weather conditions, 2023

    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, 2023

    2023

  51. [59]

    Digging into contrastive learning for robust depth estimation with diffusion models

    Jiyuan Wang, Lang Nie, Kang Liao, Shuwei Shao, and Yao Zhao. Digging into contrastive learning for robust depth estimation with diffusion models. InACM Int. Conf. Multimedia (ACMMM), pages 4129–4137, 10 2024

    2024

  52. [60]

    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. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 16055–16064, 2021

    2021

  53. [61]

    Planedepth: Self-supervised depth estimation via orthogonal planes

    Ruoyu Wang, Zehao Yu, and Shenghua Gao. Planedepth: Self-supervised depth estimation via orthogonal planes. In2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21425–21434, 2023

    2023

  54. [62]

    Monopcc: Photometric-invariant cycle constraint for monocular depth estimation of endoscopic images

    Zhiwei Wang, Ying Zhou, Shiquan He, Ting Li, Fan Huang, Qiang Ding, Xinxia Feng, Mei Liu, and Qiang Li. Monopcc: Photometric-invariant cycle constraint for monocular depth estimation of endoscopic images. Medical Image Analysis, 102:103534, 2025

    2025

  55. [63]

    The temporal opportunist: Self-supervised multi-frame monocular depth

    Jamie Watson, Oisin Mac Aodha, Victor Prisacariu, Gabriel Brostow, and Michael Firman. The temporal opportunist: Self-supervised multi-frame monocular depth. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1164–1174, June 2021

    2021

  56. [64]

    Stronger fewer & superior: Harnessing vision foundation models for domain generalized semantic segmentation

    Zhixiang Wei, Lin Chen, Yi Jin, Xiaoxiao Ma, Tianle Liu, Pengyang Ling, Ben Wang, Huaian Chen, and Jinjin Zheng. Stronger fewer & superior: Harnessing vision foundation models for domain generalized semantic segmentation. InProceedings of the IEEE/CVF Conference on Computer Vi...

    2024

  57. [65]

    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, 2025

    2025 arXiv

  58. [66]

    Tan, Bing Zeng, and Shuaicheng Liu

    Weilong Yan, Robby T. Tan, Bing Zeng, and Shuaicheng Liu. Deep homography mixture for single image rolling shutter correction. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 9834–9843, 2023

    2023

  59. [67]

    Transformer-based attention networks for continuous pixel-wise prediction

    Guanglei Yang, Hao Tang, Mingli Ding, Nicu Sebe, and Elisa Ricci. Transformer-based attention networks for continuous pixel-wise prediction. InICCV, 2021

    2021

  60. [68]

    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

    2019

  61. [69]

    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. InCVPR, 2024

    2024

  62. [70]

    Depth anything v2.arXiv:2406.09414, 2024

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2.arXiv:2406.09414, 2024

    2024 arXiv

  63. [71]

    Learning spectral-decomposed tokens for domain generalized semantic segmentation.arXiv preprint arXiv:2407.18568, 2024

    Jingjun Yi, Qi Bi, Hao Zheng, Haolan Zhan, Wei Ji, Yawen Huang, Yuexiang Li, and Yefeng Zheng. Learning spectral-decomposed tokens for domain generalized semantic segmentation.arXiv preprint arXiv:2407.18568, 2024. 13

    2024 arXiv

  64. [72]

    Udepth: Fast monocular depth estimation for visually-guided underwater robots

    Boxiao Yu, Jiayi Wu, and Md Jahidul Islam. Udepth: Fast monocular depth estimation for visually-guided underwater robots. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3116–3123. IEEE, 2023

    2023

  65. [73]

    Newcrfs: Neural window fully- connected crfs for monocular depth estimation

    Weihao Yuan, Xiaodong Gu, Zuozhuo Dai, Siyu Zhu, and Ping Tan. Newcrfs: Neural window fully- connected crfs for monocular depth estimation. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2022

    2022

  66. [74]

    Seokju Yun, Seunghye Chae, Dongheon Lee, and Youngmin Ro. Soma: Singular value decomposed minor components adaptation for domain generalizable representation learning.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  67. [75]

    Wordepth: Variational language prior for monocular depth estimation, 2024

    Ziyao Zeng, Daniel Wang, Fengyu Yang, Hyoungseob Park, Yangchao Wu, Stefano Soatto, Byung-Woo Hong, Dong Lao, and Alex Wong. Wordepth: Variational language prior for monocular depth estimation, 2024

    2024

  68. [76]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. InIEEE International Conference on Computer Vision (ICCV), 2023

    2023

  69. [77]

    Lite-mono: A lightweight cnn and transformer architecture for self-supervised monocular depth estimation

    Ning Zhang, Francesco Nex, George V osselman, and Norman Kerle. Lite-mono: A lightweight cnn and transformer architecture for self-supervised monocular depth estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 18537–185...

    2023

  70. [78]

    Adaptive domain generalization via online disagreement minimization

    Xin Zhang and Ying-Cong Chen. Adaptive domain generalization via online disagreement minimization. IEEE Transactions on Image Processing, 32:4247–4258, 2023

    2023

  71. [79]

    Xin Zhang and Robby T. Tan. Mamba as a bridge: Where vision foundation models meet vision language models for domain-generalized semantic segmentation.https://arxiv.org/abs/2504.03193, 2025

    2025 arXiv

  72. [80]

    Xin Zhang, Jinheng Xie, Yuan Yuan, Michael Bi Mi, and Robby T. Tan. Heap: Unsupervised object discovery and localization with contrastive grouping.arXiv preprint arXiv:2312.17492, 2024

    2024 arXiv

  73. [81]

    Unsupervised monocular depth estimation in highly complex environments.IEEE Transactions on Emerging Topics in Computational Intelligence, 6(5):1237–1246, 2022

    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, 2022

    2022

  74. [82]

    Monovit: Self-supervised monocular depth estimation with a vision transformer

    Chaoqiang Zhao, Youmin Zhang, Matteo Poggi, Fabio Tosi, Xianda Guo, Zheng Zhu, Guan Huang, Yang Tang, and Stefano Mattoccia. Monovit: Self-supervised monocular depth estimation with a vision transformer. In2022 International Conference on 3D Vision (3DV), pages 668–678. IEEE, 2022

    2022

  75. [83]

    Unleashing text-to-image diffusion models for visual perception.ICCV, 2023

    Wenliang Zhao, Yongming Rao, Zuyan Liu, Benlin Liu, Jie Zhou, and Jiwen Lu. Unleashing text-to-image diffusion models for visual perception.ICCV, 2023

    2023

  76. [84]

    Physical 3d adversarial attacks against monocular depth estimation in autonomous driving

    Junhao Zheng, Chenhao Lin, Jiahao Sun, Zhengyu Zhao, Qian Li, and Chao Shen. Physical 3d adversarial attacks against monocular depth estimation in autonomous driving. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24452–24461, 2024

    2024

  77. [85]

    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, and Dongbin Zhao. Steps: Joint self-supervised nighttime image enhancement and depth estimation. In2023 IEEE International Conference on Robotics and ...

    2023

  78. [86]

    Forkgan: Seeing into the rainy night

    Ziqiang Zheng, Yang Wu, Xinran Han, and Jianbo Shi. Forkgan: Seeing into the rainy night. InThe IEEE European Conference on Computer Vision (ECCV), August 2020

    2020

  79. [87]

    Tinghui Zhou, Matthew Brown, Noah Snavely, and David G. Lowe. Unsupervised learning of depth and ego-motion from video. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6612–6619, 2017

    2017

  80. [88]

    Unpaired image-to-image translation using cycle-consistent adversarial networkss

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networkss. InComputer Vision (ICCV), 2017 IEEE International Conference on, 2017. 14

    2017

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.