Pith. sign in

REVIEW 4 major objections 5 minor 40 references

FastSmoothSAM: A Fast Smooth Method For Segment Anything Model

T0 review · 4 major / 5 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 Two rounds of B-Spline fitting smooth FastSAM's jagged segmentation edges while keeping real-time speed.

desk verdict A plausible smoothing post-process for FastSAM, but the accuracy claim is contradicted by the paper's own numbers and the evaluation is too thin to support it. read the letter →

arxiv 2507.15008 v1 pith:XUEGO3XE submitted 2025-07-20 cs.CV

Jiasheng Xu , Yewang Chen This is my paper
classification cs.CV
keywords imagesegmentationFastSAMB-SplinecurvefittingedgesmoothingadaptivesamplingCannydetectionreal-timepost-processing
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 proposes a post-processing pipeline called FastSmoothSAM that cleans up the jagged, pixelated boundaries produced by FastSAM, a real-time variant of the Segment Anything Model. The core claim is that two rounds of B-Spline curve fitting, a coarse fit followed by an adaptive, edge-aware fine fit, produce smoother and geometrically more faithful object contours without losing the real-time performance that makes FastSAM attractive. If true, this gives applications such as industrial automation, medical imaging, and autonomous driving a way to get cleaner instance masks at an added cost of roughly two milliseconds per mask on a mid-range GPU. The paper supports the claim with visual comparisons, curvature and Frechet-distance metrics against FastSAM and U-Net, runtime and memory measurements on COCO, and an ablation showing that all four stages contribute.

What carries the argument

The central mechanism is the Four-Stage-Edge-Curve-Fitting Method built on B-Spline curves, curves defined by control points and basis functions through the Cox-deBoor recursion. Stage 1 fits a degree-3 B-Spline to ordered contour points extracted from the FastSAM mask. Stage 2 dilates that curve into a band and runs Canny edge detection inside the band, producing candidate edge points that are presumed to lie near the true boundary. Stage 3 adaptively resamples: high-curvature regions, identified by the curvature formula, are sampled densely and anchored to Canny points, while low-curvature regions are sampled sparsely by averaging coarse-fit and Canny neighbors via KD-tree queries. Stage 4 fits a degree-2 B-Spline to these key ordered samples. The machinery does the work of turning an unordered, noisy edge-point cloud into an ordered, curvature-aware sample set that a least-squares B-Spline fit can follow.

What would settle it

Run the full pipeline on a dataset with ground-truth object boundaries and compute mean Frechet distance per mask, stratified by object size and corner sharpness; if the smoothed masks do not have lower Frechet distance than FastSAM's raw masks on sharp-cornered or small objects, the claim that the method improves edge accuracy without compromising geometric information is refuted for those cases.

Watch

Extended reading notes

Core claim

The paper's central claim is that the four-stage, two-round B-Spline fitting pipeline significantly improves the visual and analytical quality of object edges produced by FastSAM without compromising critical geometric information. FastSAM's YOLOv8-seg backbone gives real-time speed but yields jagged contours, and the paper argues these can be repaired by post-processing alone. A first B-Spline fit gives a coarse contour, a dilated band around it is used to harvest Canny edge points as candidate true-boundary evidence, adaptive sampling densifies points in high-curvature regions and merges Canny and coarse-fit evidence, and a second, finer B-Spline fit produces the final mask. The reported results show lower mean curvature, lower curvature variance, and smaller Frechet distance to ground truth than FastSAM on COCO and on a U-Net dataset, with per-mask overhead of about 2 ms on a GTX 3060Ti and 0.7 ms on a GTX 3090.

Load-bearing premise

The pipeline assumes that Canny edge points found inside a dilated band around the coarse B-Spline fit reliably mark the true object boundary, and the paper offers no quantitative check of that proxy before building the final fit on it.

Editorial extensions

If this is right

  • FastSAM outputs can be smoothed to visually cleaner, more accurate contours with negligible runtime penalty, about 2 ms per mask on a GTX 3060Ti, keeping real-time operation.
  • Lower mean curvature and curvature variance indicate smoother contours, while smaller Frechet distance indicates closer agreement with ground-truth boundaries than FastSAM.
  • Memory consumption stays near FastSAM's level, roughly 8 percent average GPU memory versus SAM's 13 percent, so the smoothing does not reintroduce SAM's resource burden.
  • Because the pipeline is a post-processor on masks, it applies to any YOLOv8-seg or FastSAM everything-mode output without retraining the segmenter.
  • Sharp-cornered objects and very small masks remain failure cases, since B-Spline smoothing rounds corners and sparse edge samples distort small shapes.

Reading between the lines

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

  • One implication the paper leaves implicit is that the same four-stage recipe could be grafted onto other real-time segmenters, such as MobileSAM-style decoders or YOLACT, whenever their masks show boundary artifacts.
  • The method's accuracy hinges on the dilation radius, curvature threshold, and KD-tree search radius, but the paper does not report sensitivity to these, so a practical deployment would need to tune them per domain.
  • Because the fine fit uses degree-2 B-Splines, the output boundary is $C^1$-continuous, a natural fit for downstream geometry tasks such as measurement, CAD, and robotic grasping that need analytic curves.
  • A hybrid that detects corners by curvature thresholding and switches to polygonal segments at sharp vertices would directly address the paper's stated failure mode for sharp-edged objects.
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 / 5 minor

Summary. The paper proposes FastSmoothSAM, a post-processing pipeline that refines FastSAM segmentation masks by applying two rounds of B-spline curve fitting. The method operates in four stages: (1) coarse B-spline fitting to the initial mask contour, (2) Canny edge detection within a dilated band around the coarse curve, (3) adaptive sampling that combines the coarse curve, high-curvature points, and Canny edge candidates, and (4) a final fine B-spline fit. The authors claim that this approach improves the visual quality and analytical accuracy of object edges while preserving critical geometric information, and that it maintains real-time performance. Experiments report runtime and memory comparisons on COCO, smoothness metrics (curvature and Fréchet distance) on small COCO and UNet subsets, and zero-shot edge detection on BSDS500, with the latter showing mixed or slightly negative results compared to FastSAM.

Significance. If the central claim were established, FastSmoothSAM would offer a cheap, model-agnostic post-processing step that improves the usability of FastSAM masks in real-time applications. The paper does provide a clear algorithmic description, a complexity analysis, and runtime measurements showing only a modest overhead (roughly 0.7–2 ms per mask on the tested GPUs). These engineering contributions are useful. However, the claimed 'improvement in segmentation accuracy' is not supported by the paper's own quantitative results: the only large-scale evaluation (BSDS500 edge detection) shows FastSmoothSAM matching or slightly underperforming FastSAM on AP, R50, and OIS. The paper also lacks any standard segmentation accuracy metric (e.g., mAP or IoU) on a segmentation benchmark. The smoothness metrics used are partly circular, because B-spline fitting mechanically lowers curvature measures. The manuscript is therefore best viewed as a proposal for a smoothness-oriented post-processing pipeline, but the accuracy claim needs either substantiation or revision.

major comments (4)
  1. [Abstract and §5.3.3, Table 4] The central claim that the method 'improves segmentation accuracy' and preserves 'critical geometric information' is contradicted by the paper's own BSDS500 edge detection results. Table 4 shows FastSmoothSAM achieving ODS 0.737 vs. FastSAM 0.734 (a negligible improvement), but OIS 0.749 vs. 0.753, AP 0.620 vs. 0.634, and R50 0.678 vs. 0.696 (all worse). The text acknowledges these are 'slightly lower' but does not reconcile this with the abstract's accuracy claim. Since edge detection is the only large-scale quantitative benchmark in the paper, this discrepancy is load-bearing for the paper's main assertion. The authors should either provide a standard segmentation accuracy metric (e.g., mask mAP or IoU) on PASCAL VOC or COCO that demonstrates a real accuracy gain, or revise the claim to be specifically about smoothness rather than accuracy.
  2. [§5.3.3, curvature metric and Table 3] The curvature smoothness metric is not an independent evaluation of the method's benefit. Because FastSmoothSAM explicitly fits a degree-2 or degree-3 B-spline curve via least squares, the resulting contour has lower curvature and lower curvature variance essentially by construction. Comparing the mean curvature of FastSAM's raw contour with the B-spline-fitted contour is therefore circular. The Fréchet distance to ground truth in Table 3 provides a more meaningful measure, but it is reported on only 80 COCO and 200 UNet images, with no error bars or significance tests. The paper should either report a ground-truth-anchored metric on a larger set (with confidence intervals) or present a user study for visual quality; otherwise the 'smoothing improves accuracy' conclusion is not supported.
  3. [§4.2.2 and §4.2.3] The correctness of Stage 3 (adaptive sampling) and Stage 4 (fine fitting) rests entirely on the assumption that Canny edge points inside the dilated band REGmask are reliable samples of the true object boundary. The paper never validates this proxy. If Canny produces missing edges, texture edges, or biased localization at corners, the final B-spline will inherit those errors. The paper's own failure cases in Fig. 8 and Fig. 9—sharp corners and small objects being visibly distorted—are consistent with an unvalidated Canny proxy and with the geometric limitations of low-degree B-splines. The authors should provide a quantitative validation of Canny precision/recall within the band, or at minimum characterize the conditions under which this proxy is reliable. Absent this, the claim of 'without compromising critical geometric information' is not established.
  4. [§5.3.3, Table 3] The quantitative comparison in Table 3 is based on very small samples: 80 COCO images and 200 UNet images, with no standard deviation or statistical significance tests. The reported Fréchet distance differences (e.g., 1.81 vs. 2.24 on COCO) may not be statistically meaningful. Since this is one of the only quantitative supports for the accuracy claim, the authors should either substantially increase the evaluation set, report per-image distributions and significance tests, or explicitly describe these results as illustrative.
minor comments (5)
  1. [§5.3.3, Eq. (10)] Equation (10) appears to have a dimensional inconsistency: the numerator is in units of length^3 while the denominator is in units of length^2 (and the expression as written mixes variables inconsistently). The standard discrete curvature formula for point sequences should be used.
  2. [Table 3] The table contains a typo 'FastSmnoothSAM' in the dataset column. Also, the UNet row for FastSmoothSAM and FastSAM lists the dataset as 'UNet' but the text says 'UNet dataset'; please clarify the dataset source and split.
  3. [§1, Fig. 1] The notation 'F our-Stage' and 'F our-Stage-Edge-Curve-Fitting Method' uses an unusual spacing; please correct to 'Four-Stage'.
  4. [Fig. 4 caption] There is a typo: 'Inport image' should be 'Input image'.
  5. [§5.3.2, Table 1] The runtime table reports averages per image and per mask, but does not state how many runs were used for averaging or the variance. Adding standard deviations would improve reproducibility.

Circularity Check

1 steps flagged · score 3.0 of 10

The curvature smoothness metric is self-confirmatory for a B-spline smoothing method; external Fréchet and BSDS500 results provide partial but mixed grounding.

  1. self definitional [Section 5.3.3, Eq. (10), Table 3 vs. Section 4.2.4, Stage 4]
    "1) Curvature is commonly used to evaluate the smoothness performance of curves. ... The results in Table 3 indicate that compared to FastSAM, U-Net in COCO, UNet dataset, FastSmoothSAM exhibits better smoothness in contour edges. ... B-Spline fitting is performed again on the key ordered samples Ps using Alg. 1 with a degree of 2."

    The smoothness measured by Eq. (10) is exactly the quantity the method reduces by construction: Stage 4 fits a degree-2 B-spline to the sampled edge points, and B-spline least-squares fitting is a smoothing operation. Reporting lower mean curvature and curvature variance for FastSmoothSAM than for FastSAM therefore restates the mechanism of the method rather than independently validating it. This metric is offered as evidence for the claimed improvement in edge quality, but the curvature portion of that evidence is guaranteed by the fitting procedure. The Fréchet-distance column is external, and the BSDS500 results are mixed, with the paper reporting slightly lower AP, R50, and OIS for FastSmoothSAM, so the curvature metric remains a self-confirmatory part of the evaluation.

full rationale

The paper's central derivation is a post-processing pipeline: FastSAM masks are fit with a coarse B-spline, Canny edges inside a dilated band are used as candidate true boundaries, adaptive sampling selects points, and a final degree-2 B-spline is fit to those points. This pipeline has independent content and is not a fitted parameter renamed as a prediction. The only author self-citation ([32], a KD-tree neighbor-search paper by co-author Yewang Chen) is standard implementation support and not load-bearing. One evaluation criterion, however, is circular in effect: the curvature smoothness metric (Section 5.3.3, Eq. 10, Table 3) measures exactly the property that B-spline fitting is designed to enforce, so the reported improvement in smoothness is a consequence of the algorithm's definition rather than evidence of accuracy. The Fréchet-distance comparisons to ground truth provide some external grounding, but they are reported on small samples, and the BSDS500 edge-detection evaluation (Table 4) shows that FastSmoothSAM does not consistently beat FastSAM. The paper's own limitations for sharp-cornered and small objects further temper the central claim, but these are correctness concerns rather than circularity. Overall, the circularity is limited to the self-confirmatory smoothness metric, giving a score of 3.

Assumptions & free parameters 7 free parameters · 3 assumptions · 0 invented entities

The pipeline rests on standard B-Spline math and on unverified assumptions about Canny edge reliability and mask topology. The many hand-chosen hyperparameters are not reported, so the method as described is under-specified.

free parameters (7)
  • B-Spline degree for coarse fit = 3
    Stage 1 (Section 4.2.1) fixes degree k=3; no sensitivity analysis.
  • B-Spline degree for fine fit = 2
    Stage 4 (Section 4.2.4) fixes degree k=2; no sensitivity analysis.
  • Curvature threshold theta = not reported
    Algorithm 2 uses theta to select high-curvature points; the value is not given.
  • KD-tree search radius r = not reported
    Algorithm 3 uses r for neighborhood queries; the value is not given.
  • Number of sampling knots n = not reported
    Algorithm 2 samples n knots on the curve; the value is not given.
  • Dilation size for REGmask = not reported
    Stage 2 dilates the coarse curve to build the mask; the dilation amount is unspecified.
  • Canny Gaussian sigma = not reported
    Section 4.2.2 adjusts sigma but reports no value.
assumptions (3)
  • standard math B-Spline basis functions and least-squares fitting provide an adequate smooth approximation of object contours.
    Invoked in Sections 3.1-3.2 before the method is constructed.
  • domain assumption Canny edge points inside the dilated coarse-fit region reliably approximate the true object boundary.
    Stage 2 (Section 4.2.2) and Algorithm 3 depend on this to select fitting points.
  • domain assumption FastSAM/YOLOv8-seg masks are single closed contours suitable for curve fitting.
    The pipeline assumes the mask has one ordered boundary; the paper's own failure cases (small objects, fragmented shapes in Fig. 9) show this does not always hold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FastSmoothSAM: A Fast Smooth Method For Segment Anything Model." pith.science (2026). https://pith.science/paper/XUEGO3XE

@misc{pith2026250715008,
  author       = {Pith},
  title        = {Pith review of: FastSmoothSAM: A Fast Smooth Method For Segment Anything Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XUEGO3XE}},
  note         = {Machine review of arXiv:2507.15008}
}
read the original abstract

Accurately identifying and representing object edges is a challenging task in computer vision and image processing. The Segment Anything Model (SAM) has significantly influenced the field of image segmentation, but suffers from high memory consumption and long inference times, limiting its efficiency in real-time applications. To address these limitations, Fast Segment Anything (FastSAM) was proposed, achieving real-time segmentation. However, FastSAM often generates jagged edges that deviate from the true object shapes. Therefore, this paper introduces a novel refinement approach using B-Spline curve fitting techniques to enhance the edge quality in FastSAM. Leveraging the robust shape control and flexible geometric construction of B-Splines, a four-stage refining process involving two rounds of curve fitting is employed to effectively smooth jagged edges. This approach significantly improves the visual quality and analytical accuracy of object edges without compromising critical geometric information. The proposed method improves the practical utility of FastSAM by improving segmentation accuracy while maintaining real-time processing capabilities. This advancement unlocks greater potential for FastSAM technology in various real-world scenarios, such as industrial automation, medical imaging, and autonomous systems, where precise and efficient edge recognition is crucial.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 19 canonical work pages

  1. [1]

    Digital Signal Processing 132:103812

    Kaur R, Singh S (2023) A comprehensive review of object detection with deep learning. Digital Signal Processing 132:103812. https://doi.org/10.1016/j.dsp.2022.103812

    work page arXiv 2023

  2. [2]

    Interna- tional Journal of Computational Intelligence Systems 16(1):126

    Sirisha U, Praveen SP, Srinivasu PN, et al (2023) Statistical analysis of design aspects of various yolo-based deep learning models for object detection. Interna- tional Journal of Computational Intelligence Systems 16(1):126. https://doi.org/10.1007/ s44196-023-00302-w

    work page 2023

  3. [3]

    EAI Endorsed Transactions on Pervasive Health and Technology 7(27):e6–e6

    Ramesh K, Kumar GK, Swapna K, et al (2021) A review of medical image segmentation algorithms. EAI Endorsed Transactions on Pervasive Health and Technology 7(27):e6–e6. https://doi.org/10.4108/eai.12-4-2021.169184

    arXiv 2021

  4. [4]

    IEEE Transactions on Intelligent Transportation Systems 22(3):1341–1360

    Feng D, Haase-Sch¨ utz C, Rosenbaum L, et al (2020) Deep multi-modal object detection and semantic segmentation for autonomous driving: Datasets, methods, and challenges. IEEE Transactions on Intelligent Transportation Systems 22(3):1341–1360. https://doi. org/10.1109/TITS.2020.2972974

    work page arXiv 2020

  5. [5]

    IEEE Transactions on Intel- ligent Transportation Systems 23(11):21405–21417

    Wang H, Chen Y, Cai Y, et al (2022) Sfnet-n: An improved sfnet algorithm for semantic segmentation of low-light autonomous driving road scenes. IEEE Transactions on Intel- ligent Transportation Systems 23(11):21405–21417. https://doi.org/10.1109/TITS.2022. 3177615

    doi:10.1109/tits.2022 2022

  6. [6]

    In: 2023 IEEE/CVF Inter- national Conference on Computer Vision (ICCV), pp 3992–4003, https://doi.org/10.1109/ ICCV51070.2023.00371 17

    Kirillov A, Mintun E, Ravi N, et al (2023) Segment anything. In: 2023 IEEE/CVF Inter- national Conference on Computer Vision (ICCV), pp 3992–4003, https://doi.org/10.1109/ ICCV51070.2023.00371 17

    arXiv 2023

  7. [7]

    Nature Communica- tions 15(1):654

    Ma J, He Y, Li F, et al (2024) Segment anything in medical images. Nature Communica- tions 15(1):654. https://doi.org/10.1038/s41467-024-44824-z

    doi:10.1038/s41467-024-44824-z 2024

  8. [8]

    In: Medical Image Segmentation Challenge

    Ali Q, Chen Y, Wong A (2024) Repvit-medsam: Efficient segment anything in the medical images. In: Medical Image Segmentation Challenge. Springer, p 195–205, https://doi.org/ 10.1007/978-3-031-81854-7 13

    work page doi:10.1007/978-3-031-81854-7 2024

Show all 40 references
  1. [9]

    In: Annual Conference on Medical Image Understand- ing and Analysis, Springer, pp 187–201, https://doi.org/10.1007/978-3-031-66958-3 14

    Paranjape JN, Nair NG, Sikder S, et al (2024) Adaptivesam: Towards efficient tuning of sam for surgical scene segmentation. In: Annual Conference on Medical Image Understand- ing and Analysis, Springer, pp 187–201, https://doi.org/10.1007/978-3-031-66958-3 14

    2024 doi

  2. [10]

    Electronics 12(12):2730

    Emek Soylu B, Guzel MS, Bostanci GE, et al (2023) Deep-learning-based approaches for semantic segmentation of natural scene images: A review. Electronics 12(12):2730. https://doi.org/10.3390/electronics12122730

    2023 doi

  3. [11]

    In: International Conference on Learning Representations, https://doi.org/10.48550/arXiv.2010.11929

    Dosovitskiy A, Beyer L, Kolesnikov A, et al (2021) An image is worth 16x16 words: Transformers for image recognition at scale. In: International Conference on Learning Representations, https://doi.org/10.48550/arXiv.2010.11929

    2021 doi arXiv

  4. [12]

    Engineering Applications of Artificial Intelligence 126:106669

    Thisanke H, Deshan C, Chamith K, et al (2023) Semantic segmentation using vision transformers: A survey. Engineering Applications of Artificial Intelligence 126:106669. https://doi.org/10.1016/j.engappai.2023.106669

    2023

  5. [13]

    Advances in Neural Information Processing Systems 35:4971–4982

    Zhang B, Tian Z, Tang Q, et al (2022) Segvit: Semantic segmentation with plain vision transformers. Advances in Neural Information Processing Systems 35:4971–4982. https: //doi.org/10.48550/arXiv.2210.05844

    2022 doi arXiv

  6. [14]

    International Journal of Computer Vision 132(4):1126–1147

    Zhang B, Liu L, Phan MH, et al (2024) Segvit v2: Exploring efficient and continual semantic segmentation with plain vision transformers. International Journal of Computer Vision 132(4):1126–1147. https://doi.org/10.1007/s11263-023-01894-8

    2024 doi

  7. [15]

    arXiv preprint arXiv:230612156 https://doi.org/10.48550/arXiv.2306.12156

    Zhao X, Ding W, An Y, et al (2023) Fast segment anything. arXiv preprint arXiv:230612156 https://doi.org/10.48550/arXiv.2306.12156

    2023 doi arXiv

  8. [16]

    Unser M, Aldroubi A, Eden M (1993) B-spline signal processing. i. theory. IEEE transactions on signal processing 41(2):821–833. https://doi.org/10.1109/78.193220

    1993 doi

  9. [17]

    In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 10062–10071, https://doi.org/10.1109/CVPR52729.2023.00970

    Pak B, Lee J, Jin KH (2023) B-spline texture coefficients estimator for screen content image super-resolution. In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 10062–10071, https://doi.org/10.1109/CVPR52729.2023.00970

    2023

  10. [18]

    mesh-based multi-objective deformable image registration

    Andreadis G, Mulder JI, Bouter A, et al (2024) A tournament of transformation models: B-spline-based vs. mesh-based multi-objective deformable image registration. In: Medical Imaging 2024: Image Processing, SPIE, pp 374–384, https://doi.org/10.1117/12.2692876

    2024 doi

  11. [19]

    IEEE Transactions on pat- tern analysis and machine intelligence (6):679–698

    Canny J (1986) A computational approach to edge detection. IEEE Transactions on pat- tern analysis and machine intelligence (6):679–698. https://doi.org/10.1109/TPAMI.1986. 4767851

    1986 doi

  12. [20]

    arXiv preprint arXiv:230614289 https://doi.org/10.48550/arXiv

    Zhang C, Han D, Qiao Y, et al (2023) Faster segment anything: Towards lightweight sam for mobile applications. arXiv preprint arXiv:230614289 https://doi.org/10.48550/arXiv. 2306.14289

    2023 doi arXiv

  13. [21]

    arXiv preprint arXiv:231209579 https://doi.org/10.48550/arXiv.2312.09579

    Zhang C, Han D, Zheng S, et al (2023) Mobilesamv2: Faster segment anything to everything. arXiv preprint arXiv:231209579 https://doi.org/10.48550/arXiv.2312.09579

    2023 doi arXiv

  14. [22]

    In: 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp 16111–16121, https://doi.org/10.1109/ CVPR52733.2024.01525 18

    Xiong Y, Varadarajan B, Wu L, et al (2024) Efficientsam: Leveraged masked image pretraining for efficient segment anything. In: 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp 16111–16121, https://doi.org/10.1109/ CVPR52733.2024.01525 18

    2024

  15. [23]

    In: Pro- ceedings of the 37th International Conference on Neural Information Processing Systems

    Zou X, Yang J, Zhang H, et al (2023) Segment everything everywhere all at once. In: Pro- ceedings of the 37th International Conference on Neural Information Processing Systems. Curran Associates Inc., Red Hook, NY, USA, NIPS ’23, https://doi.org/10.5555/3666122. 3666990

    2023 doi

  16. [25]

    In: Advances in Neural Information Processing Systems, vol 37

    Wang X, Yang J, Darrell T (2024) Segment anything without supervision. In: Advances in Neural Information Processing Systems, vol 37. Curran Associates, Inc., pp 138731– 138755, https://doi.org/10.48550/arXiv.2406.20081

    2024 doi arXiv

  17. [26]

    In: 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp 9156–9165, https: //doi.org/10.1109/ICCV.2019.00925

    Bolya D, Zhou C, Xiao F, et al (2019) Yolact: Real-time instance segmentation. In: 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp 9156–9165, https: //doi.org/10.1109/ICCV.2019.00925

    2019

  18. [27]

    Remote Sensing 13(18):3551

    Harmening C, Hobmaier C, Neuner H (2021) Laser scanner–based deformation analysis using approximating b-spline surfaces. Remote Sensing 13(18):3551. https://doi.org/10. 3390/rs13183551

    2021

  19. [28]

    Remote Sensing 12(5):829

    Kermarrec G, Kargoll B, Alkhatib H (2020) Deformation analysis using b-spline sur- face with correlated terrestrial laser scanner observations—a bridge under load. Remote Sensing 12(5):829. https://doi.org/10.3390/rs12050829

    2020 doi

  20. [29]

    Computers & Geosciences 178:105408

    Wang Y (2023) Partially overlapped grain shape restoration of the granular rock images. Computers & Geosciences 178:105408. https://doi.org/10.1016/j.cageo.2023.105408

    2023

  21. [30]

    Com- puter Methods in Applied Mechanics and Engineering 299:337–365

    Giannelli C, J¨ uttler B, Kleiss SK, et al (2016) Thb-splines: An effective mathematical technology for adaptive refinement in geometric design and isogeometric analysis. Com- puter Methods in Applied Mechanics and Engineering 299:337–365. https://doi.org/10. 1016/j.cma.2015.11.002

    2016

  22. [31]

    In: Proceedings of the ACM SIGGRAPH Symposium on High Performance Graphics

    Wu Z, Zhao F, Liu X (2011) Sah kd-tree construction on gpu. In: Proceedings of the ACM SIGGRAPH Symposium on High Performance Graphics. Association for Computing Machinery, HPG ’11, pp 71–78, https://doi.org/10.1145/2018323.201833

    2011

  23. [32]

    Information Sciences 472:145–162

    Chen Y, Zhou L, Tang Y, et al (2019) Fast neighbor search by using revised kd tree. Information Sciences 472:145–162. https://doi.org/10.1016/j.ins.2018.09.012

    2019 doi

  24. [33]

    In: Computer Vision–ECCV 2014, Springer

    Lin TY, Maire M, Belongie S, et al (2014) Microsoft coco: Common objects in context. In: Computer Vision–ECCV 2014, Springer. Springer International Publishing, Cham, pp 740–755, https://doi.org/10.1007/978-3-319-10602-1 48

    2014 doi

  25. [34]

    In: Navab N, Hornegger J, Wells WM, et al (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

    Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. In: Navab N, Hornegger J, Wells WM, et al (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. Springer International Publishing, Cham, pp 234–24...

    2015 doi

  26. [35]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 33(5):898–

    Arbel´ aez P, Maire M, Fowlkes C, et al (2011) Contour detection and hierarchical image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence 33(5):898–

    2011

  27. [36]

    a talk at the Stanford Artificial Project in 1968:271–272

    Sobel I, Feldman G, et al (1968) A 3x3 isotropic gradient operator for image processing. a talk at the Stanford Artificial Project in 1968:271–272. https://doi.org/10.13140/RG.2. 1.1912.4965

    1968

  28. [37]

    In: 2015 IEEE International Con- ference on Computer Vision (ICCV), pp 1395–1403, https://doi.org/10.1109/ICCV.2015

    Xie S, Tu Z (2015) Holistically-nested edge detection. In: 2015 IEEE International Con- ference on Computer Vision (ICCV), pp 1395–1403, https://doi.org/10.1109/ICCV.2015. 164 19

    2015 doi

  29. [38]

    In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 1392– 1402, https://doi.org/10.1109/CVPR52688.2022.00146

    Pu M, Huang Y, Liu Y, et al (2022) Edter: Edge detection with transformer. In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 1392– 1402, https://doi.org/10.1109/CVPR52688.2022.00146

    2022

  30. [39]

    International journal of computer vision 59:167–181

    Felzenszwalb PF, Huttenlocher DP (2004) Efficient graph-based image segmentation. International journal of computer vision 59:167–181. https://doi.org/10.1023/B:VISI. 0000022288.19776.77

    2004

  31. [40]

    International journal of computer vision 111:98–136

    Everingham M, Eslami SA, Van Gool L, et al (2015) The pascal visual object classes challenge: A retrospective. International journal of computer vision 111:98–136. https: //doi.org/10.1007/s11263-014-0733-5 20

    2015 doi

  32. [916]

    https://doi.org/10.1109/TPAMI.2010.161

    2010 doi

Pith tools

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