pith. sign in

arxiv: 2410.01473 · v3 · submitted 2024-10-02 · 💻 cs.CV

SinkSAM-Net: Knowledge-Driven Self-Supervised Sinkhole Segmentation Using Topographic Priors and Segment Anything Model

Osher Rafaeli , Tal Svoray , Ariel Nahlieli This is my paper

Pith reviewed 2026-05-23 20:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords sinkhole segmentationself-supervised learningSegment Anything Modeltopographic priorspseudo-label generationaerial imagerydrone imagerymonocular depth
0
0 comments X

The pith

SinkSAM-Net generates pseudo-labels from topographic closed depressions and SAM to train a sinkhole segmenter that reaches 95 percent of fully supervised performance.

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

The paper introduces a self-supervised pipeline that starts with traditional topographic detection of closed depressions, then refines those regions into pixel-level sinkhole masks using monocular depth estimates and coordinate-wise bounding box jittering. These masks serve as training targets for an iterative loop that improves both the prompts sent to the Segment Anything Model and a lightweight EfficientNetV2-UNet. The final prompt-free model is evaluated on a large collection of aerial and drone images covering many sinkhole types and dates. A sympathetic reader would care because manual pixel annotation of irregular, vegetation-obscured sinkholes is slow and expensive, yet accurate maps matter for infrastructure risk and soil conservation. The central result is that domain knowledge plus foundation models can substitute for most human labels.

Core claim

The authors claim that combining topographic computations of closed depressions with an iterative, geometry-aware prompt-based Segment Anything Model, refined at the pixel level by monocular depth information and coordinate-wise bounding box jittering, produces pseudo-labels sufficient to train a lightweight target model that achieves approximately 95 percent of the performance obtained from human-annotated supervision on diverse sinkhole datasets imaged from both aircraft and drones.

What carries the argument

The iterative pseudo-label generation loop that prompts the Segment Anything Model with topographic closed-depression detections and refines the outputs using monocular depth plus coordinate-wise bounding box jittering before feeding them to the EfficientNetV2-UNet target.

If this is right

  • The same unlabeled RGB sinkhole collections can be used to train models without any manual pixel labels.
  • Knowledge is transferred to a low-parameter, prompt-free model that runs quickly at inference time.
  • The framework maintains performance across multiple sinkhole formation dates and both aerial and high-resolution drone imagery sources.
  • Foundational models such as SAM and Depth Anything V2 become effective for this task once topographic and geometric priors are added to the prompt and refinement steps.

Where Pith is reading between the lines

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

  • The same prior-plus-foundation-model recipe could be tested on other irregular geomorphic features such as landslides or gully networks where topographic closed-depression logic does not directly apply.
  • Once the lightweight model exists, regular low-cost drone surveys could support near-real-time sinkhole inventory updates without repeated expert labeling.
  • The method's reliance on monocular depth suggests it may degrade in very flat terrain or under heavy canopy where depth estimates become unreliable; that boundary could be quantified on new test regions.

Load-bearing premise

The pseudo-labels created by depth-based boundary refinement and bounding-box jittering remain accurate enough across iterations that the target model keeps improving rather than reinforcing its own errors.

What would settle it

A controlled experiment that measures the intersection-over-union of the generated pseudo-labels against a held-out set of human annotations and shows that the downstream EfficientNetV2-UNet reaches well below 95 percent of supervised performance when those pseudo-labels are used for training.

Figures

Figures reproduced from arXiv: 2410.01473 by Ariel Nahlieli, Osher Rafaeli, Tal Svoray.

Figure 1
Figure 1. Figure 1: Venn diagram illustrating the SinkSAM approach of merging Compu [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Input data: Patches of RGB, ground truth annotated sinkholes, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: The study area is located in the northwestern part of the Negev region, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: SinkSAM framework: Stage 1: Depth estimation from an RGB image using DAV2. Stage 2: ”Fill sinks” techniques and a substraction of estimated depth from a sink-free raster resulting in delineation of closed depression. Stage 3: Prompts generation: a threshold value is used to remove small sinks and create bounding boxes for SAM. Finally, at Stage 4, the SAM tuned model utilizes an image encoder and mask deco… view at source ↗
Figure 5
Figure 5. Figure 5: Experimental setup: Each comparison was designed to test the performance of SinkSAM Framework: (A) closed depressions, identified through the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evaluated maps of sinkhole segmentation on the unseen Yaen site: It is visually evident that the photogrammetric closed depressions miss parts of the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance at different IoU thresholds: SinkSAM outperforms all [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Performance of sinkhole detection models: A bar plot of the confusion [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: SAM prompted by closed depressions: SAM improved sinkhole [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Closed depressions prompting: YOLO overestimates by incorrectly [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ground level sinkholes images from study area: Sinkholes can vary [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Photogrammetric DEM vs DAV2: Sinkholes delineation from evaluated maps, compared to the ground truth sinkhole masks, DAV2 closed depressions delineate the boundaries more precisely and ”fill” the entire area of the sinkholes, while DEM closed depressions only ”fill” parts of the sinkholes and do not fill small sinkholes at all [PITH_FULL_IMAGE:figures/full_fig_p010_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Monocular depth estimation: Elevation profile plots on small [PITH_FULL_IMAGE:figures/full_fig_p011_13.png] view at source ↗
read the original abstract

Soil sinkholes significantly influence soil degradation, infrastructure vulnerability, and landscape evolution. However, their irregular shapes, combined with interference from shadows and vegetation, make it challenging to accurately quantify their properties using remotely sensed data. In addition, manual annotation can be laborious and costly. In this study, we introduce a novel self-supervised framework for sinkhole segmentation, termed SinkSAM-Net, which integrates traditional topographic computations of closed depressions with an iterative, geometry-aware, prompt-based Segment Anything Model (SAM). We generate high-quality pseudo-labels through pixel-level refinement of sinkhole boundaries by integrating monocular depth information with random prompts augmentation technique named coordinate-wise bounding box jittering (CWBJ). These pseudo-labels iteratively enhance a lightweight EfficientNetV2-UNet target model, ultimately transferring knowledge to a prompt-free, low-parameter, and fast inference model. Our proposed approach achieves approximately 95\% of the performance obtained through manual supervision by human annotators. The framework's performance was evaluated on a large sinkhole database, covering diverse sinkhole dateset-induced sinkholes using both aerial and high-resolution drone imagery. This paper presents the first self-supervised framework for sinkhole segmentation, demonstrating the robustness of foundational models (such as SAM and Depth Anything V2) when combined with prior topographic and geometry knowledge and an iterative self-learning pipeline. SinkSAM-Net has the potential to be trained effectively on extensive unlabeled RGB sinkholes datasets, achieving comparable performance to a supervised model. The code and interactive demo for SinkSAM-Net are available at https://osherr1996.github.io/SinkSAMNet

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces SinkSAM-Net, a self-supervised framework for sinkhole segmentation in aerial and high-resolution drone imagery. It generates pseudo-labels by combining topographic closed-depression detection with iterative, geometry-aware prompting of the Segment Anything Model (SAM), refined via monocular depth from Depth Anything V2 and coordinate-wise bounding-box jittering (CWBJ). These pseudo-labels train a lightweight EfficientNetV2-UNet target model, which is claimed to reach approximately 95% of the performance of a fully supervised baseline trained on human annotations. The work positions itself as the first self-supervised method for this task and releases code and a demo.

Significance. If the performance claims are substantiated with quantitative evidence, the result would show that topographic priors plus off-the-shelf foundation models can produce usable pseudo-labels for remote-sensing segmentation, enabling training on large unlabeled RGB datasets with limited annotation cost. This would be a concrete demonstration of knowledge-driven self-supervision in a domain where manual labeling is expensive.

major comments (2)
  1. [Abstract] Abstract and method description: the headline claim of reaching ~95% of supervised performance is asserted without any quantitative tables, error bars, dataset splits, ablation results, or direct metrics (e.g., IoU or boundary F-score) comparing the generated pseudo-labels against a held-out set of human annotations prior to the self-training loop. This leaves the central performance gap unverifiable.
  2. [Method (pseudo-label generation)] Pseudo-label generation step: the assumption that monocular-depth refinement plus CWBJ produces masks sufficiently close to human quality to support the reported performance is stated but not independently verified with any pre-training quantitative check against ground-truth annotations.
minor comments (1)
  1. [Abstract] Abstract: 'dateset-induced' appears to be a typo for 'dataset-induced'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The two major comments both concern the need for explicit, pre-training quantitative verification of pseudo-label quality against held-out human annotations. We address each point below and will revise the manuscript to include the requested metrics and checks.

read point-by-point responses

  1. Referee: [Abstract] Abstract and method description: the headline claim of reaching ~95% of supervised performance is asserted without any quantitative tables, error bars, dataset splits, ablation results, or direct metrics (e.g., IoU or boundary F-score) comparing the generated pseudo-labels against a held-out set of human annotations prior to the self-training loop. This leaves the central performance gap unverifiable.

    Authors: We agree that the abstract states the ~95% figure without the supporting quantitative details the referee requests. While the experiments section of the manuscript reports final model performance, it does not isolate a direct, pre-loop comparison of the generated pseudo-labels to held-out ground truth. In the revision we will add a dedicated table (and accompanying text) that reports IoU, boundary F-score, and related metrics for the pseudo-labels versus held-out human annotations, together with error bars obtained from repeated runs and explicit dataset-split information. This addition will make the headline claim directly verifiable before the self-training loop is described. revision: yes

  2. Referee: [Method (pseudo-label generation)] Pseudo-label generation step: the assumption that monocular-depth refinement plus CWBJ produces masks sufficiently close to human quality to support the reported performance is stated but not independently verified with any pre-training quantitative check against ground-truth annotations.

    Authors: The referee is correct that the manuscript currently asserts the quality of the monocular-depth refinement and CWBJ step without an independent quantitative check against ground truth prior to training. We will add a new evaluation subsection that performs exactly this check: pseudo-masks produced by the full pseudo-label pipeline are compared to held-out human annotations using the same metrics (IoU, boundary F-score, etc.) and the results are reported with error bars. This verification will be placed in both the method and results sections of the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: pseudo-labels generated from external topographic and foundation-model priors

full rationale

The derivation chain begins with independent topographic closed-depression detection, followed by monocular depth refinement via Depth Anything V2 and SAM prompted by coordinate-wise bounding-box jittering. These steps produce pseudo-labels that are then used to train the EfficientNetV2-UNet; the pseudo-label generation does not invoke the target model or any fitted parameters derived from it. No equations reduce a claimed prediction to its own inputs by construction, no self-citations are load-bearing, and no uniqueness theorems or ansatzes are smuggled in. The 95% performance claim is an empirical comparison against a supervised baseline on the sinkhole dataset and does not constitute a self-referential derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the domain assumption that topographic closed depressions plus SAM refinement yield usable pseudo-labels; no free parameters or invented entities are introduced in the abstract description.

axioms (1)
  • domain assumption Topographic computations of closed depressions provide reliable initial locations for sinkholes that can be refined by SAM and depth cues
    Invoked in the description of pseudo-label generation from traditional topographic computations integrated with SAM.

pith-pipeline@v0.9.0 · 5828 in / 1336 out tokens · 24630 ms · 2026-05-23T20:16:07.594050+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

85 extracted references · 85 canonical work pages · 3 internal anchors

  1. [1]

    Geomorphology and natural hazards in karst areas: A review,

    J. De Waele, F. Guti ´errez, M. Parise, and L. Plan, “Geomorphology and natural hazards in karst areas: A review,” Geomorphology, vol. 134, no. 1, pp. 1–8, 2011. Geomorphology and Natural Hazards in Karst Areas

    work page 2011

  2. [2]

    Sinkholes and subsidence: Karst and cavernous rocks in engineering and construction,

    T. Waltham, F. Bell, and M. Culshaw, “Sinkholes and subsidence: Karst and cavernous rocks in engineering and construction,” Springer-Praxis, 01 2005

    work page 2005

  3. [3]

    Guti ´errez, J

    F. Guti ´errez, J. P. Galve, P. Lucha, C. Casta ˜neda, J. Bonachea, and J. Guerrero, “Integrating geomorphological mapping, trenching, insar and gpr for the identification and characterization of sinkholes: A review and application in the mantled evaporite karst of the ebro valley (ne spain),” Geomorphology, vol. 134, no. 1-2, pp. 144–156, 2011

    work page 2011

  4. [4]

    Identification, pre- diction, and mitigation of sinkhole hazards in evaporite karst areas,

    F. Gutierrez, A. H. Cooper, and K. S. Johnson, “Identification, pre- diction, and mitigation of sinkhole hazards in evaporite karst areas,” Environmental Geology, vol. 53, pp. 1007–1022, 2008

    work page 2008

  5. [5]

    Subsurface erosion by soil piping: significance and research needs,

    A. Bernatek-Jakiel and J. Poesen, “Subsurface erosion by soil piping: significance and research needs,” Earth-Science Reviews , vol. 185, pp. 1107–1128, 2018

    work page 2018

  6. [6]

    The mapping of closed depressions and its contribution to the geodiversity inventory,

    E. Pardo-Ig ´uzquiza and P. Dowd, “The mapping of closed depressions and its contribution to the geodiversity inventory,” International Journal of Geoheritage and Parks , vol. 9, no. 4, pp. 480–495, 2021

    work page 2021

  7. [7]

    A comparison of accuracy and cost of lidar versus stand exam data for landscape management on the malheur national forest,

    S. Hummel, A. Hudak, E. Uebler, M. Falkowski, and K. Megown, “A comparison of accuracy and cost of lidar versus stand exam data for landscape management on the malheur national forest,” Journal of forestry, vol. 109, no. 5, pp. 267–273, 2011

    work page 2011

  8. [8]

    Dem generation with uav pho- togrammetry and accuracy analysis in sahitler hill,

    M. Uysal, A. Toprak, and N. Polat, “Dem generation with uav pho- togrammetry and accuracy analysis in sahitler hill,” Measurement, vol. 73, pp. 539–543, 2015

    work page 2015

  9. [9]

    Monocular depth estimation based on deep learning: An overview,

    C. Zhao, Q. Sun, C. Zhang, Y . Tang, and F. Qian, “Monocular depth estimation based on deep learning: An overview,” Science China Tech- nological Sciences, vol. 63, no. 9, pp. 1612–1627, 2020

    work page 2020

  10. [10]

    Dynamic semantically guided monocu- lar depth estimation for uav environment perception,

    V .-C. Miclea and S. Nedevschi, “Dynamic semantically guided monocu- lar depth estimation for uav environment perception,”IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–11, 2024

    work page 2024

  11. [11]

    Deep learning for monocular depth estimation: A review,

    Y . Ming, X. Meng, C. Fan, and H. Yu, “Deep learning for monocular depth estimation: A review,”Neurocomputing, vol. 438, pp. 14–33, 2021

    work page 2021

  12. [12]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,” 2024

    work page 2024

  13. [13]

    Sink removal from digital elevation model–a necessary evil for hydrological analysis,

    A. Sharma and K. Tiwari, “Sink removal from digital elevation model–a necessary evil for hydrological analysis,” Current science , vol. 117, pp. 1512–1515, 11 2019

    work page 2019

  14. [14]

    Using machine learning to identify karst sinkholes from lidar-derived topographic depressions in the bluegrass region of kentucky,

    J. Zhu, A. M. Nolte, N. Jacobs, and M. Ye, “Using machine learning to identify karst sinkholes from lidar-derived topographic depressions in the bluegrass region of kentucky,” Journal of Hydrology , vol. 588, p. 125049, 2020

    work page 2020

  15. [15]

    Identification of karst sinkholes in a forested karst landscape using airborne laser scanning data and water flow analysis,

    J. Hofierka, M. Gallay, P. Bandura, and J. ˇSaˇsak, “Identification of karst sinkholes in a forested karst landscape using airborne laser scanning data and water flow analysis,” Geomorphology, vol. 308, pp. 265–277, 2018

    work page 2018

  16. [16]

    Uav- based doline mapping in brazilian karst: A cave heritage protection reconnaissance,

    C. Ferreira, Y . Hussain, R. Uagoda, T. Silva, and R. Cicerelli, “Uav- based doline mapping in brazilian karst: A cave heritage protection reconnaissance,” Open Geosciences, 10 2023

    work page 2023

  17. [17]

    Deep learning in environ- mental remote sensing: Achievements and challenges,

    Q. Yuan, H. Shen, T. Li, Z. Li, S. Li, Y . Jiang, H. Xu, T. Weiwei, Q. Yang, J. Wang, J. Gao, and L. Zhang, “Deep learning in environ- mental remote sensing: Achievements and challenges,” Remote Sensing of Environment, vol. 241, p. 111716, 05 2020

    work page 2020

  18. [18]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” CoRR, vol. abs/1505.04597, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  19. [19]

    Deep learning in remote sensing: A comprehensive review and list of resources,

    X. X. Zhu, D. Tuia, L. Mou, G.-S. Xia, L. Zhang, F. Xu, and F. Fraundorfer, “Deep learning in remote sensing: A comprehensive review and list of resources,” IEEE Geoscience and Remote Sensing Magazine, vol. 5, no. 4, pp. 8–36, 2017

    work page 2017

  20. [20]

    Sinkhole detection by deep learning and data association,

    N. V . Hoai, N. M. Dung, and S. Ro, “Sinkhole detection by deep learning and data association,” in 2019 Eleventh International Conference on Ubiquitous and Future Networks (ICUFN) , pp. 211–213, 2019

    work page 2019

  21. [21]

    Performance analysis of deep learning cnn models for disease detection in plants using image segmentation,

    P. Sharma, Y . P. S. Berwal, and W. Ghai, “Performance analysis of deep learning cnn models for disease detection in plants using image segmentation,” Information Processing in Agriculture , vol. 7, no. 4, pp. 566–574, 2020

    work page 2020

  22. [22]

    Deep-learning-based automatic sinkhole recognition: Application to the eastern dead sea,

    O. Alrabayah, D. Caus, R. A. Watson, H. Z. Schulten, T. Weigel, L. R¨upke, and D. Al-Halbouni, “Deep-learning-based automatic sinkhole recognition: Application to the eastern dead sea,” Remote Sensing , vol. 16, no. 13, p. 2264, 2024

    work page 2024

  23. [23]

    Image segmentation using text and image prompts,

    T. L ¨uddecke and A. Ecker, “Image segmentation using text and image prompts,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 7086–7096, June 2022

    work page 2022

  24. [24]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection,

    S. Liu, Z. Zeng, T. Ren, F. Li, H. Zhang, J. Yang, Q. Jiang, C. Li, J. Yang, H. Su, J. Zhu, and L. Zhang, “Grounding dino: Marrying dino with grounded pre-training for open-set object detection,” 2024

    work page 2024

  25. [25]

    The segment anything model (sam) for remote sensing applications: From zero to one shot,

    L. P. Osco, Q. Wu, E. L. de Lemos, W. N. Gonc ¸alves, A. P. M. Ramos, J. Li, and J. M. Junior, “The segment anything model (sam) for remote sensing applications: From zero to one shot,” 2023

    work page 2023

  26. [26]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Doll´ar, and R. Girshick, “Segment anything,” 2023

    work page 2023

  27. [27]

    You only look once: Unified, real-time object detection,

    J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 779–788, 2016

    work page 2016

  28. [28]

    Active learning enabled low-cost cell image segmentation using bounding box annotation,

    Y . Zhu, Q. Yang, and L. Xu, “Active learning enabled low-cost cell image segmentation using bounding box annotation,” 2024

    work page 2024

  29. [29]

    An integrated framework to identify and analyze karst sinkholes,

    V . Zumpano, L. Pisano, and M. Parise, “An integrated framework to identify and analyze karst sinkholes,” Geomorphology, vol. 332, pp. 213–225, 2019

    work page 2019

  30. [30]

    A fast, simple and versatile algorithm to fill the depressions of digital elevation models,

    O. Planchon and F. Darboux, “A fast, simple and versatile algorithm to fill the depressions of digital elevation models,” CATENA, vol. 46, no. 2, pp. 159–176, 2002

    work page 2002

  31. [31]

    An efficient method for identifying and filling surface depressions in digital elevation models for hydrologic analysis and modelling,

    L. Wang and H. Liu, “An efficient method for identifying and filling surface depressions in digital elevation models for hydrologic analysis and modelling,” International Journal of Geographical Information Science, vol. 20, no. 2, pp. 193–213, 2006

    work page 2006

  32. [32]

    Another fast and simple dem depression-filling algorithm based on priority queue structure,

    L. Yong-He, Z. Wan-Chang, and X. Jing-Wen, “Another fast and simple dem depression-filling algorithm based on priority queue structure,” Atmospheric and Oceanic Science Letters , vol. 2, no. 4, pp. 214–219, 2009

    work page 2009

  33. [33]

    Developing efficient procedures for automated sinkhole extraction from lidar dems,

    X. Miao, X. Qiu, S.-S. Wu, J. Luo, D. Gouzie, and H. Xie, “Developing efficient procedures for automated sinkhole extraction from lidar dems,” Photogrammetric Engineering and Remote Sensing , vol. 79, pp. 545– 554, 06 2013

    work page 2013

  34. [34]

    Improved karst sinkhole mapping in kentucky using lidar techniques: A pilot study in floyds fork watershed.,

    J. Zhu, T. P. Taylor, J. C. Currens, and M. M. Crawford, “Improved karst sinkhole mapping in kentucky using lidar techniques: A pilot study in floyds fork watershed.,” Journal of Cave & Karst Studies, vol. 76, no. 3, 2014. 13

    work page 2014

  35. [35]

    Sinkhole detection and char- acterization using lidar-derived dem with logistic regression,

    Y . J. Kim, B. H. Nam, and H. Youn, “Sinkhole detection and char- acterization using lidar-derived dem with logistic regression,” Remote Sensing, vol. 11, no. 13, p. 1592, 2019

    work page 2019

  36. [36]

    Identification of soil piping-related depressions using an airborne lidar dem: Role of land use changes,

    A. Bernatek-Jakiel and M. Jakiel, “Identification of soil piping-related depressions using an airborne lidar dem: Role of land use changes,” Geomorphology, vol. 378, p. 107591, 2021

    work page 2021

  37. [37]

    A semi-automatic model for sinkhole identification in a karst area of zhijin county, china,

    H. Chen, T. Oguchi, and P. Wu, “A semi-automatic model for sinkhole identification in a karst area of zhijin county, china,” in International Conference on Intelligent Earth Observing and Applications 2015 , vol. 9808, pp. 209–216, SPIE, 2015

    work page 2015

  38. [38]

    Distinguishing actual and artefact depressions in digital elevation data,

    J. B. Lindsay and I. F. Creed, “Distinguishing actual and artefact depressions in digital elevation data,”Computers & Geosciences, vol. 32, no. 8, pp. 1192–1204, 2006. Spatial Modeling for Environmental and Hazard Management

    work page 2006

  39. [39]

    Sinkholenet: A novel rgb-slope sinkhole dataset and deep weakly-supervised learning framework for sinkhole classification and localization,

    A. Yavariabdi, H. Kusetogullari, O. Orhan, E. Uray, V . Demir, T. Celik, and E. Mendi, “Sinkholenet: A novel rgb-slope sinkhole dataset and deep weakly-supervised learning framework for sinkhole classification and localization,” The Egyptian Journal of Remote Sensing and Space Sciences, vol. 26, no. 4, pp. 966–973, 2023

    work page 2023

  40. [40]

    Detection of sinkholes and landslides in a semi-arid en- vironment using deep-learning methods, uav images, and topographical derivatives,

    N. Kariminejad, A. Mondini, M. Hosseinalizadeh, F. Golkar, and H. R. Pourghasemi, “Detection of sinkholes and landslides in a semi-arid en- vironment using deep-learning methods, uav images, and topographical derivatives,” 2023

    work page 2023

  41. [41]

    Automatic segmentation of sinkholes using a convolutional neural network,

    M. U. Rafique, J. Zhu, and N. Jacobs, “Automatic segmentation of sinkholes using a convolutional neural network,” Earth and Space Science, vol. 9, no. 2, p. e2021EA002195, 2022

    work page 2022

  42. [42]

    Morphological characteristics and distri- bution of dolines in slovenia, a study of a lidar-based doline map of slovenia,

    A. Mihevc and R. Mihevc, “Morphological characteristics and distri- bution of dolines in slovenia, a study of a lidar-based doline map of slovenia,” Acta Carsologica, vol. 50, May 2021

    work page 2021

  43. [43]

    Amfenet: An adaptive multiscale feature fusion enhancement network for sinkhole detection,

    G. Zhu, Y . Niu, L. Ruan, and X. Zhang, “Amfenet: An adaptive multiscale feature fusion enhancement network for sinkhole detection,” IEEE Geoscience and Remote Sensing Letters , vol. 21, pp. 1–5, 2024

    work page 2024

  44. [44]

    Deep monocular depth estima- tion via integration of global and local predictions,

    Y . Kim, H. Jung, D. Min, and K. Sohn, “Deep monocular depth estima- tion via integration of global and local predictions,” IEEE Transactions on Image Processing , vol. 27, no. 8, pp. 4131–4144, 2018

    work page 2018

  45. [45]

    Unsupervised monocular depth estimation with left-right consistency,

    C. Godard, O. Mac Aodha, and G. J. Brostow, “Unsupervised monocular depth estimation with left-right consistency,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 270–279, 2017

    work page 2017

  46. [46]

    Know your limits: Accuracy of long range stereoscopic object measurements in practice,

    P. Pinggera, D. Pfeiffer, U. Franke, and R. Mester, “Know your limits: Accuracy of long range stereoscopic object measurements in practice,” in Computer Vision – ECCV 2014 (D. Fleet, T. Pajdla, B. Schiele, and T. Tuytelaars, eds.), (Cham), pp. 96–111, Springer International Publishing, 2014

    work page 2014

  47. [47]

    Deep Convolutional Neural Fields for Depth Estimation from a Single Image

    F. Liu, C. Shen, and G. Lin, “Deep convolutional neural fields for depth estimation from a single image,” CoRR, vol. abs/1411.6387, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  48. [48]

    Monocular depth estimation with improved long-range accuracy for uav environment perception,

    V .-C. Miclea and S. Nedevschi, “Monocular depth estimation with improved long-range accuracy for uav environment perception,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–15, 2022

    work page 2022

  49. [49]

    Deep learning for monocular depth estimation from uav images,

    L. Madhuanand, F. Nex, and M. Y . Yang, “Deep learning for monocular depth estimation from uav images,” ISPRS Annals of the Photogramme- try, Remote Sensing and Spatial Information Sciences , vol. V-2-2020, pp. 451–458, 2020

    work page 2020

  50. [50]

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

    R. Birkl, D. Wofk, and M. M ¨uller, “Midas v3.1 – a model zoo for robust monocular relative depth estimation,” 2023

    work page 2023

  51. [51]

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

    L. Yang, B. Kang, Z. Huang, X. Xu, J. Feng, and H. Zhao, “Depth anything: Unleashing the power of large-scale unlabeled data,” 2024

    work page 2024

  52. [52]

    To- wards robust monocular depth estimation in non-lambertian surfaces,

    J. Zhang, J. Li, Y . Huang, Y . Wang, J. Zheng, L. Shen, and Z. Cao, “To- wards robust monocular depth estimation in non-lambertian surfaces,” arXiv preprint arXiv:2408.06083 , 2024

    work page arXiv 2024

  53. [53]

    Depth any canopy: Leveraging depth foundation models for canopy height estimation,

    D. R. Cambrin, I. Corley, and P. Garza, “Depth any canopy: Leveraging depth foundation models for canopy height estimation,” arXiv preprint arXiv:2408.04523, 2024

    work page arXiv 2024

  54. [54]

    Segment anything model (sam) for medical image segmentation: A preliminary review,

    L. Zhang, X. Deng, and Y . Lu, “Segment anything model (sam) for medical image segmentation: A preliminary review,” in 2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) , pp. 4187–4194, 2023

    work page 2023

  55. [55]

    Sam-helps-shadow:when segment any- thing model meet shadow removal,

    X. Zhang, C. Gu, and S. Zhu, “Sam-helps-shadow:when segment any- thing model meet shadow removal,” 2023

    work page 2023

  56. [56]

    Segment anything for microscopy,

    A. Archit, S. Nair, N. Khalid, P. Hilt, V . Rajashekar, M. Freitag, S. Gupta, A. Dengel, S. Ahmed, and C. Pape, “Segment anything for microscopy,” bioRxiv, 2023

    work page 2023

  57. [57]

    Application of segment anything model for civil infrastructure defect assessment,

    M. Ahmadi, A. G. Lonbar, A. Sharifi, A. T. Beris, M. Nouri, and A. S. Javidi, “Application of segment anything model for civil infrastructure defect assessment,” arXiv preprint arXiv:2304.12600 , 2023

    work page arXiv 2023

  58. [58]

    A flexible deep learning crater detection scheme using segment anything model (sam),

    I. Giannakis, A. Bhardwaj, L. Sam, and G. Leontidis, “A flexible deep learning crater detection scheme using segment anything model (sam),” Icarus, vol. 408, p. 115797, 2024

    work page 2024

  59. [59]

    Semantic segmentation of glaciological features across multiple remote sensing platforms with the segment anything model (sam),

    S. Shankar, L. A. Stearns, and C. van der Veen, “Semantic segmentation of glaciological features across multiple remote sensing platforms with the segment anything model (sam),” Journal of Glaciology , pp. 1–10, 2023

    work page 2023

  60. [60]

    Road-sam: Adapting the seg- ment anything model to road extraction from large very-high-resolution optical remote sensing images,

    W. Feng, F. Guan, C. Sun, and W. Xu, “Road-sam: Adapting the seg- ment anything model to road extraction from large very-high-resolution optical remote sensing images,” IEEE Geoscience and Remote Sensing Letters, vol. 21, pp. 1–5, 2024

    work page 2024

  61. [61]

    Sam- assisted remote sensing imagery semantic segmentation with object and boundary constraints,

    X. Ma, Q. Wu, X. Zhao, X. Zhang, M.-O. Pun, and B. Huang, “Sam- assisted remote sensing imagery semantic segmentation with object and boundary constraints,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–16, 2024

    work page 2024

  62. [62]

    Prompt- based segmentation at multiple resolutions and lighting conditions using segment anything model 2,

    O. Rafaeli, T. Svoray, R. Blushtein-Livnon, and A. Nahlieli, “Prompt- based segmentation at multiple resolutions and lighting conditions using segment anything model 2,” 2024

    work page 2024

  63. [63]

    SAM 2: Segment Anything in Images and Videos

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R ¨adle, C. Rolland, L. Gustafson, et al. , “Sam 2: Segment anything in images and videos,” arXiv preprint arXiv:2408.00714 , 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  64. [64]

    Geosam: Fine-tuning sam with sparse and dense visual prompting for automated segmentation of mobility infrastructure,

    R. I. Sultan, C. Li, H. Zhu, P. Khanduri, M. Brocanelli, and D. Zhu, “Geosam: Fine-tuning sam with sparse and dense visual prompting for automated segmentation of mobility infrastructure,” 2024

    work page 2024

  65. [65]

    Segment anything model for road network graph extraction,

    C. Hetang, H. Xue, C. Le, T. Yue, W. Wang, and Y . He, “Segment anything model for road network graph extraction,” 2024

    work page 2024

  66. [66]

    On aligning SAM to remote sensing data,

    R. Sahay and A. Savakis, “On aligning SAM to remote sensing data,” in Geospatial Informatics XIV (K. Palaniappan and G. Seetharaman, eds.), vol. 13037, p. 1303703, International Society for Optics and Photonics, SPIE, 2024

    work page 2024

  67. [67]

    Comprehensive multimodal segmentation in medical imaging: Combining yolov8 with sam and hq- sam models,

    S. Pandey, K.-F. Chen, and E. B. Dam, “Comprehensive multimodal segmentation in medical imaging: Combining yolov8 with sam and hq- sam models,” in Proceedings of the IEEE/CVF international conference on computer vision , pp. 2592–2598, 2023

    work page 2023

  68. [68]

    Devel- oping approaches in building classification and extraction with synergy of yolov8 and sam models,

    A. Khatua, A. Bhattacharya, A. K. Goswami, and B. H. Aithal, “Devel- oping approaches in building classification and extraction with synergy of yolov8 and sam models,” Spatial Information Research , pp. 1–20, 2024

    work page 2024

  69. [69]

    Using image segmentation models to analyse high-resolution earth observation data: new tools to monitor disease risks in changing environments,

    F. Trujillano, G. Jimenez, E. Manrique, N. F. Kahamba, F. Okumu, N. Apollinaire, G. Carrasco-Escobar, B. Barrett, and K. Fornace, “Using image segmentation models to analyse high-resolution earth observation data: new tools to monitor disease risks in changing environments,” International Journal of Health Geographics, vol. 23, no. 1, p. 13, 2024

    work page 2024

  70. [70]

    Topological data analysis guided segment anything model prompt optimization for zero-shot segmentation in biological imaging,

    R. Glatt and S. Liu, “Topological data analysis guided segment anything model prompt optimization for zero-shot segmentation in biological imaging,” 2023

    work page 2023

  71. [71]

    Meteorological data and reports

    Israel Meteorological Service, “Meteorological data and reports.” https: //ims.gov.il, n.d. n.d

    work page

  72. [72]

    Soil quality and aggregation in runoff water harvesting forestry systems in the semi-arid israeli negev,

    I. Stavi and E. Argaman, “Soil quality and aggregation in runoff water harvesting forestry systems in the semi-arid israeli negev,” CATENA, vol. 146, pp. 88–93, 2016. Dan H. Yaalon Memorial Issue

    work page 2016

  73. [73]

    Piping formation and distri- bution in the semi-arid northern negev environment: A new conceptual model,

    A. Nahlieli, T. Svoray, and E. Argaman, “Piping formation and distri- bution in the semi-arid northern negev environment: A new conceptual model,” CATENA, vol. 213, p. 106201, 2022

    work page 2022

  74. [74]

    Arcgis desktop: Release 10,

    C. E. S. R. I. Redlands, “Arcgis desktop: Release 10,” 2011

    work page 2011

  75. [75]

    Yolov9: Learning what you want to learn using programmable gradient information,

    C.-Y . Wang, I.-H. Yeh, and H.-Y . M. Liao, “Yolov9: Learning what you want to learn using programmable gradient information,” 2024

    work page 2024

  76. [76]

    Segment anything in medical images,

    J. Ma, Y . He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nature Communications, vol. 15, Jan. 2024

    work page 2024

  77. [77]

    Learn how to fine-tune the segment anything model (sam),

    A. Bonnet, “Learn how to fine-tune the segment anything model (sam),” Encord Blog, April 2023. Accessed: 2024-09-04

    work page 2023

  78. [78]

    Ultralytics yolo,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolo,” jan 2023. Available: https://github.com/ultralytics/ultralytics

    work page 2023

  79. [79]

    Morphometric comparisons between automated and manual karst depression inventories in apalachicola national forest, florida, and mammoth cave national park, kentucky, usa,

    J. Wall, D. Bohnenstiehl, K. Wegmann, and N. Levine, “Morphometric comparisons between automated and manual karst depression inventories in apalachicola national forest, florida, and mammoth cave national park, kentucky, usa,” Natural Hazards, vol. 85, 01 2017

    work page 2017

  80. [80]

    Performance of human annotators in object detection and segmentation of remotely sensed data,

    R. Blushtein-Livnon, T. Svoray, and M. Dorman, “Performance of human annotators in object detection and segmentation of remotely sensed data,” 2024

    work page 2024

Showing first 80 references.