Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Objectomaly: Objectness-Aware Refinement for OoD Segmentation with Structural Consistency and Boundary Precision

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 Objectomaly claims a training-free, three-stage refinement — averaging anomaly scores inside SAM masks, then sharpening contours with Laplacian and Gaussian filtering — reaches state-of-the-art OoD segmentation on SMIYC and RoadAnomaly.

desk verdict Plausible OASC idea but the MBP equation as written turns the output into a zero-filled edge map, contradicting the reported results; not ready for review. read the letter →

arxiv 2507.07460 v2 pith:526MJK4F submitted 2025-07-10 cs.CV cs.AI

Jeonghoon Song , Sunghun Kim , Jaegyun Im , Byeongjoon Noh This is my paper
classification cs.CVcs.AI
keywords out-of-distributionsegmentationanomalyobjectness-awarecalibrationSegmentAnythingModelboundaryrefinementautonomousdrivingpost-hocroad-scenedetection
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

Objectomaly is a post-hoc, training-free refinement for out-of-distribution (OoD) road-scene segmentation. The paper's claim is that the three failures of mask-based detectors — blurred boundaries between adjacent objects, fragmented anomaly scores inside one object, and false positives from textured background — all come from scoring pixels instead of objects, and that fixing them requires object-level priors. Its recipe is to take the raw anomaly map of a detector such as Mask2Anomaly, replace every pixel score inside each SAM-generated instance mask with the mask's mean score, and then apply a Laplacian filter and Gaussian smoothing to the boundary. On SMIYC AnomalyTrack, ObstacleTrack, and RoadAnomaly, the paper reports pixel-level AuPRC up to 96.99, FPR95 down to 0.07, and component-level F1 up to 83.44, all without retraining the detector.

What carries the argument

The load-bearing mechanism is objectness-aware score calibration. Precisely, SAM (Segment Anything Model, a class-agnostic promptable segmenter) produces overlapping instance masks $S_k$; Objectomaly sorts them by area, computes $\alpha_k = \frac{1}{|S'_k|} \sum_{x \in S'_k} A_{\mathrm{init}}(x)$, and sets $A_{\mathrm{score}}(x) = \alpha_k$ for $x$ inside the mask while leaving outside pixels unchanged. This replacement removes within-object score variance and suppresses background noise, under the premise that SAM's masks delineate true anomaly objects. The second mechanism, MBP, computes the absolute Laplacian of the calibrated score map and smooths it with a Gaussian $G_\sigma$, sharpening object borders without committing to a hard binary threshold.

What would settle it

An experiment that would settle the claim: take the SMIYC ObstacleTrack images, measure Objectomaly's component-level F1, then rerun the OASC stage with artificially corrupted SAM masks — for example, deleting 20% of each ground-truth anomaly mask before computing $\alpha_k$, or unioning each anomaly mask with an equally sized road patch. If F1 collapses, the gain depends on SAM's mask completeness as assumed; if F1 survives, the boundary stage and not the mean-replacement assumption is carrying the improvement.

Watch

Extended reading notes

Core claim

The central discovery, as the authors state it, is that object structure is the missing axis in OoD segmentation: spatial consistency, not stronger pixel confidence, is what separates clean anomaly masks from noisy ones. In equations (9)-(10), each SAM mask $S'_k$ is assigned a single representative anomaly score $\alpha_k$, the average of the coarse map over the mask, and all pixels inside the mask take that value; this is what turns scattered high-confidence blobs and internally inconsistent objects into whole, uniformly scored regions. The Meticulous Boundary Precision stage then applies a Laplacian operator to the calibrated score map and convolves the result with a Gaussian kernel, keeping the map soft while aligning contours with image gradients. The paper argues that the two stages are synergistic: OASC supplies structural coherence and MBP supplies boundary precision, so their combined effect is larger than the sum of the individual ablation gains.

Load-bearing premise

The approach assumes that every anomaly is a coherent object that SAM segments completely and without background contamination, because Eq. (10) replaces every pixel in a mask with the mask mean; if SAM splits an anomaly or merges it with the road, that replacement either spreads nonzero scores into background or fragments the object, and the method inherits SAM's mistakes.

Editorial extensions

If this is right

  • Post-hoc deployment: any existing OoD detector that outputs a pixel score map can receive the refinement without retraining; the paper demonstrates this by swapping the CAS backbone across RPL, Maskomaly, RbA, UNO, and Mask2Anomaly.
  • Runtime flexibility: substituting SAM2, FastSAM, or MobileSAM inside OASC preserves most of the gain while drastically changing inference speed, which makes the pipeline tunable for embedded driving systems.
  • Safety metrics improve at the object level: component-level PPV and F1 gains imply autonomous planners receive whole anomaly objects rather than fragmented score blobs, reducing both missed obstacles and false alarms.
  • Near-zero FPR95 on ObstacleTrack implies the pipeline suppresses false positives from road texture and shadows, which is the failure mode that most limits the practical use of pixel-level anomaly scores.

Reading between the lines

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

  • As an editorial extension, the method's ceiling is bounded by the coarse anomaly map: OASC only renormalizes scores that CAS already emitted, so an anomaly entirely missed by the backbone cannot be recovered by mask averaging; a natural test is to run CAS with several backbones and feed the union of their score maps into OASC.
  • Because the calibration step is agnostic to semantics, the same objectness-aware score normalization could transfer to open-vocabulary segmentation or panoptic-anomaly setups that already produce instance masks, where the coarse score could come from a text-prompt or prototype distance rather than a softmax.
  • The mean is a crude representative statistic; replacing it with a max, quantile, or boundary-weighted statistic would change the precision-recall trade-off, and the ablation table suggests the choice interacts strongly with the dataset (SMIYC OT versus RA).
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 Objectomaly, a training-free post-hoc refinement framework for out-of-distribution (OoD) segmentation. It uses an existing OoD backbone (Mask2Anomaly) as the Coarse Anomaly Scoring (CAS) stage, calibrates pixel-level scores by replacing them with the mean score inside SAM-generated masks (Objectness-Aware Score Calibration, OASC), and then applies a Laplacian magnitude followed by Gaussian smoothing (Meticulous Boundary Precision, MBP). The authors report state-of-the-art pixel-level and component-level results on SMIYC AnomalyTrack, ObstacleTrack, and RoadAnomaly, along with ablations on stage contributions, SAM variants, and real-world dashcam footage.

Significance. The modular, training-free design is attractive, and the paper evaluates on standard benchmarks with a clear pipeline. I found no circular derivation or parameter fitting to the target metrics: the method relies on external components (Mask2Anomaly and SAM) rather than tuning a parameter to the reported outputs. However, the reported numbers cannot be reproduced from the algorithm as written in Eqs. (11)-(12), and several main-text claims are directly contradicted by the paper's own tables. Because these issues concern the core method definition and the validity of the headline results, the manuscript does not currently establish its central claims. The framework may be salvageable after a correct algorithm specification and a full audit of the experimental tables.

major comments (4)
  1. [3.2.3, Eqs. (11)-(12)] The final anomaly map is defined as G_sigma * |∂²Ascore/∂x² + ∂²Ascore/∂y²|. After Eq. (10), Ascore is piecewise constant inside each SAM mask and in unmodified regions, so the Laplacian is zero throughout object interiors and nonzero only on narrow boundary rings; Gaussian smoothing cannot reconstruct interior scores. Pixel-level AuPRC/FPR95 and component-level F1 values therefore cannot follow from this map. No residual term, rescaling, or normalization is specified. This is an internal inconsistency between the stated algorithm and the reported results.
  2. [4.2.2 / Table 2] The text states that Objectomaly "outperforms all competing methods across all three metrics" on SMIYC AT, but Table 2 reports sIoU 43.70 for Objectomaly versus 68.01 for UNO and 49.77 for RPL. The sIoU gap is large, and the claim as written is false. The component-level superiority claim must be restated or the experiments re-run.
  3. [4.2.1 / Table 1] The text claims the lowest FPR95 across all evaluated datasets, but on RoadAnomaly Table 1 gives Objectomaly FPR95=9.92 while UNO achieves 9.20. The RA result is therefore not state-of-the-art on FPR95, and the stated generalization claim is contradicted by the paper's own table.
  4. [4.3.2 / Table 4] The FastSAM row reports AuPRC=11.75 and FPR95=92.65 for SMIYC OT, which is inconsistent with the claim that FastSAM performs competitively and with every other row in the table. If these entries are swapped or erroneous, corrected numbers are needed before the model-flexibility conclusion can be evaluated.
minor comments (5)
  1. [3.1.1] The sentence "The mask-level anomaly score is computed as [26]:" is duplicated immediately before Eq. (3); remove the second occurrence.
  2. [3.2.2 / Fig. 3] The text and Eq. (8) state that masks are processed in descending size order, but the Figure 3 caption says refinement proceeds from smaller objects to larger ones; please make the ordering consistent.
  3. [Table 1] In the Maskomaly row for SMIYC OT, the values "0.96 96.14" are the only occurrence of AuPRC below 1 in the table; please verify the column alignment or provide a reference for these "actual results".
  4. [Abstract] The abstract reports AuPRC 96.99 and FPR95 0.07 without specifying that these are SMIYC ObstacleTrack numbers; please qualify them by dataset.
  5. [OASC, Eq. (10)] The robustness of the OASC stage to imperfect SAM masks (over-segmentation or merging with background) is not analyzed; a quantitative failure-mode analysis would strengthen the generalization claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the refinement pipeline is empirical post-processing over external components, with no fitted parameter renamed as a prediction.

full rationale

The paper's central claim is that a three-stage post-processing pipeline (CAS, OASC, MBP) improves OoD segmentation. The pipeline takes a coarse anomaly map from Mask2Anomaly, calibrates it with SAM masks via Eq. (9)-(10), and refines boundaries via Eq. (11)-(12). None of these steps fits a parameter to the benchmark labels, nor does any equation define the output in terms of the final metric. The only self-citations (refs [16], [24], [25]) are introductory citations to the authors' prior work on semantic segmentation applications and are not load-bearing for the method or the results. The reported gains are externally evaluated against SMIYC and RoadAnomaly datasets, and the ablation study compares stages against a CAS-only baseline. The skeptic's concern about MBP—that Eqs. (11)-(12) output a blurred Laplacian edge magnitude rather than a refined anomaly score field—is a genuine correctness/reproducibility question about whether the stated algorithm can produce the claimed numbers, but it is not circularity: the output is not equivalent to the input by definition, nor is any fitted value being presented as a prediction. Therefore the circularity score is 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central evaluation depends on SAM behaving as a reliable objectness prior, on the benchmark protocol tolerating test-time use of an external segmenter, and on an ad hoc filter chain preserving anomaly-score semantics. The only reported free parameter is the unspecified Gaussian smoothing scale; several other implementation details are absent.

free parameters (1)
  • Gaussian smoothing sigma and kernel size = not reported
    Eq. (12) uses G_sigma with sigma controlling boundary sharpness versus noise suppression, but the value and kernel size are never specified; they are hand-chosen and directly affect FPR95 and boundary precision.
assumptions (3)
  • domain assumption OoD anomalies are coherent objects that SAM can segment into complete, contiguous masks.
    OASC replaces per-pixel anomaly scores with the mean per SAM mask (Eqs. 9-10). If SAM undersegments, oversegments, or merges the anomaly with background, the calibration corrupts the scores. This is the central premise of the refinement.
  • domain assumption Test-time use of a high-capacity external segmenter does not violate the benchmark protocol.
    OASC runs SAM on test images to derive masks. The benchmarks may permit test-time processing, but injecting a powerful external objectness prior is a protocol assumption that should be stated and defended.
  • ad hoc to paper Laplacian magnitude followed by Gaussian smoothing preserves or improves an anomaly score field.
    Eqs. (11)-(12) apply |Laplacian| and Gaussian smoothing and call the result boundary-refined soft anomaly scores. This filter chain is not derived, and it maps flat object interiors to near zero, so the assumption is both load-bearing and unproven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Objectomaly: Objectness-Aware Refinement for OoD Segmentation with Structural Consistency and Boundary Precision." pith.science (2026). https://pith.science/paper/526MJK4F

@misc{pith2026250707460,
  author       = {Pith},
  title        = {Pith review of: Objectomaly: Objectness-Aware Refinement for OoD Segmentation with Structural Consistency and Boundary Precision},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/526MJK4F}},
  note         = {Machine review of arXiv:2507.07460}
}
abstract

Out-of-Distribution (OoD) segmentation is critical for safety-sensitive applications like autonomous driving. However, existing mask-based methods often suffer from boundary imprecision, inconsistent anomaly scores within objects, and false positives from background noise. We propose \textbf{\textit{Objectomaly}}, an objectness-aware refinement framework that incorporates object-level priors. Objectomaly consists of three stages: (1) Coarse Anomaly Scoring (CAS) using an existing OoD backbone, (2) Objectness-Aware Score Calibration (OASC) leveraging SAM-generated instance masks for object-level score normalization, and (3) Meticulous Boundary Precision (MBP) applying Laplacian filtering and Gaussian smoothing for contour refinement. Objectomaly achieves state-of-the-art performance on key OoD segmentation benchmarks, including SMIYC AnomalyTrack/ObstacleTrack and RoadAnomaly, improving both pixel-level (AuPRC up to 96.99, FPR$_{95}$ down to 0.07) and component-level (F1$-$score up to 83.44) metrics. Ablation studies and qualitative results on real-world driving videos further validate the robustness and generalizability of our method. Code will be released upon publication.

Figures

Figures reproduced from arXiv: 2507.07460 by the authors.

Figure 1
Figure 1. The limitations in current OoD segmentation method: [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of Mask2Anomaly and the proposed Objectomaly. (a) Mask2Anomaly: An initial anomaly map is generated via [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the OASC process. Given the initial [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison across datasets: (a) SMIYC AT, (b) SMIYC OT, (c) RA. Each row shows input, ground truth, baselines [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results of Objectomaly on real-world dash [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 45 canonical work pages

  1. [1]

    Maskomaly: Zero-shot mask anomaly segmentation

    Jan Ackermann, Christos Sakaridis, and Fisher Yu. Maskomaly: Zero-shot mask anomaly segmentation. In 34th British Machine Vision Conference 2023, BMVC 2023, Aberdeen, UK, November 20-24, 2023 . BMV A, 2023. 1, 2, 5, 6

    work page 2023

  2. [2]

    Anomaly detection in autonomous driving: A survey

    Daniel Bogdoll, Maximilian Nitsche, and J Marius Z¨ollner. Anomaly detection in autonomous driving: A survey. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 4488–4499, 2022. 1

    work page 2022

  3. [3]

    Seg- mentmeifyoucan: A benchmark for anomaly segmen- tation

    Robin Chan, Krzysztof Lis, Svenja Uhlemeyer, Her- mann Blum, Sina Honari, Roland Siegwart, Pascal Fua, Mathieu Salzmann, and Matthias Rottmann. Seg- mentmeifyoucan: A benchmark for anomaly segmen- tation. In Thirty-fifth Conference on Neural Infor- mation Processing Systems Datasets and Benchmarks Track (Round 2), 2021. 1, 5

    work page 2021

  4. [4]

    Encoder-decoder with atrous separable convolution for semantic image segmentation

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European con- ference on computer vision (ECCV) , pages 801–818,

    work page

  5. [5]

    Ting Chen, Lala Li, Saurabh Saxena, Geoffrey Hin- ton, and David J. Fleet. A generalist framework for panoptic segmentation of images and videos. In Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 909–919, 2023. 2

    work page 2023

  6. [6]

    Per-pixel classification is not all you need for seman- tic segmentation

    Bowen Cheng, Alex Schwing, and Alexander Kirillov. Per-pixel classification is not all you need for seman- tic segmentation. Advances in neural information pro- cessing systems, 34:17864–17875, 2021. 2

    work page 2021

  7. [7]

    Masked- attention mask transformer for universal image seg- mentation

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexander Kirillov, and Rohit Girdhar. Masked- attention mask transformer for universal image seg- mentation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 1290–1299, 2022. 2

    work page 2022

  8. [8]

    Outlier detection by ensembling uncertainty with negative ob- jectness

    Anja Deli ´c, Matej Grcic, and Sini ˇsa ˇSegvi´c. Outlier detection by ensembling uncertainty with negative ob- jectness. In 35th British Machine Vision Conference 2024, BMVC 2024, Glasgow, UK, November 25-28,

    work page 2024

Show all 46 references
  1. [9]

    Imagenet: A large-scale hierarchi- cal image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In 2009 IEEE conference on com- puter vision and pattern recognition , pages 248–255. Ieee, 2009. 5

    2009

  2. [10]

    Exploring the limits of out-of-distribution detection

    Stanislav Fort, Jie Ren, and Balaji Lakshminarayanan. Exploring the limits of out-of-distribution detection. Advances in neural information processing systems , 34:7068–7081, 2021. 2

    2021

  3. [11]

    Generalize or detect? towards ro- bust semantic segmentation under multiple distribu- tion shifts

    Zhitong Gao, Bingnan Li, Mathieu Salzmann, and Xuming He. Generalize or detect? towards ro- bust semantic segmentation under multiple distribu- tion shifts. Advances in Neural Information Process- ing Systems, 37:52014–52039, 2024. 2

    2024

  4. [12]

    Densehybrid: Hybrid anomaly detection for dense open-set recognition

    Matej Grci ´c, Petra Bevandi ´c, and Sini ˇsa ˇSegvi´c. Densehybrid: Hybrid anomaly detection for dense open-set recognition. In European Conference on Computer Vision, pages 500–517. Springer, 2022. 1

    2022

  5. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 5

    2016

  6. [14]

    Unleashing multispectral video’s po- tential in semantic segmentation: A semi-supervised viewpoint and new UA V-view benchmark

    Wei Ji, Jingjing Li, Wenbo Li, Yilin Shen, Li cheng, and Hongxia Jin. Unleashing multispectral video’s po- tential in semantic segmentation: A semi-supervised viewpoint and new UA V-view benchmark. In The Thirty-eighth Annual Conference on Neural Informa- tion Processing Syste...

    2024

  7. [15]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. In Proceedings of the IEEE/CVF international conference on computer vi- sion, pages 4015–4026, 2023. 2, 4

    2023

  8. [16]

    Saliencymix+: Noise-minimized image mixing method with saliency map in data aug- mentation

    Hajeong Lee, Zhixiong Jin, Jiyoung Woo, and Byeongjoon Noh. Saliencymix+: Noise-minimized image mixing method with saliency map in data aug- mentation. IEEE Access, 2025. 1

    2025

  9. [17]

    Open-set semantic seg- mentation for point clouds via adversarial prototype framework

    Jianan Li and Qiulei Dong. Open-set semantic seg- mentation for point clouds via adversarial prototype framework. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 9425–9434, 2023. 2

    2023

  10. [18]

    Shiyu Liang, Yixuan Li, and R. Srikant. Enhancing the reliability of out-of-distribution image detection in neural networks. In International Conference on Learning Representations, 2018. 2

    2018

  11. [19]

    Detecting the unexpected via image resynthesis

    Krzysztof Lis, Krishna Nakka, Pascal Fua, and Math- ieu Salzmann. Detecting the unexpected via image resynthesis. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision , pages 2152– 2161, 2019. 1, 5

    2019

  12. [20]

    Residual pattern learning for pixel-wise out- of-distribution detection in semantic segmentation

    Yuyuan Liu, Choubo Ding, Yu Tian, Guansong Pang, Vasileios Belagiannis, Ian Reid, and Gustavo Carneiro. Residual pattern learning for pixel-wise out- of-distribution detection in semantic segmentation. In Proceedings of the IEEE/CVF International Confer- ence on Computer Visio...

    2023

  13. [21]

    Swin 9 transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin 9 transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision , pages 10012–10022, 2021. 2

    2021

  14. [22]

    Towards ac- curate unified anomaly segmentation

    Wenxin Ma, Qingsong Yao, Xiang Zhang, Zhelong Huang, Zihang Jiang, and S.Kevin Zhou. Towards ac- curate unified anomaly segmentation. In Proceedings of the Winter Conference on Applications of Computer Vision (WACV), pages 1342–1352, 2025. 1

    2025

  15. [23]

    Rba: Segmenting unknown regions re- jected by all

    Nazir Nayal, Misra Yavuz, Joao F Henriques, and Fatma G¨uney. Rba: Segmenting unknown regions re- jected by all. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 711– 722, 2023. 1, 2, 5, 6

    2023

  16. [24]

    A novel method of predictive collision risk area estimation for proac- tive pedestrian accident prevention system in urban surveillance infrastructure

    Byeongjoon Noh and Hwasoo Yeo. A novel method of predictive collision risk area estimation for proac- tive pedestrian accident prevention system in urban surveillance infrastructure. Transportation research part C: emerging technologies, 137:103570, 2022. 1

    2022

  17. [25]

    Vision-based potential pedestrian risk analysis on unsignalized crosswalk using data mining techniques

    Byeongjoon Noh, Wonjun No, Jaehong Lee, and David Lee. Vision-based potential pedestrian risk analysis on unsignalized crosswalk using data mining techniques. Applied Sciences, 10(3):1057, 2020. 1

    2020

  18. [26]

    Unmasking anomalies in road-scene segmentation

    Shyam Nandan Rai, Fabio Cermelli, Dario Fontanel, Carlo Masone, and Barbara Caputo. Unmasking anomalies in road-scene segmentation. InProceedings of the IEEE/CVF International Conference on Com- puter Vision, pages 4037–4046, 2023. 1, 2, 3, 5, 6

    2023

  19. [27]

    SAM 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R ¨adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junting Pan, Kalyan Va- sudev Alwala, Nicolas Carion, Chao-Yuan Wu, Ross Girshick, Piotr Dollar, and Christoph Feich...

    2025

  20. [28]

    U-net: Convolutional networks for biomedical im- age segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical im- age segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, par...

    2015

  21. [29]

    Quantification of uncertainty with adversarial models

    Kajetan Schweighofer, Lukas Aichberger, Mykyta Ielanskyi, G ¨unter Klambauer, and Sepp Hochreiter. Quantification of uncertainty with adversarial models. Advances in Neural Information Processing Systems , 36:19446–19484, 2023. 1

    2023

  22. [30]

    Open-world semantic segmentation including class similarity

    Matteo Sodano, Federico Magistri, Lucas Nunes, Jens Behley, and Cyrill Stachniss. Open-world semantic segmentation including class similarity. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3184–3194, 2024. 1

    2024

  23. [31]

    Drive&segment: Unsupervised semantic segmentation of urban scenes via cross-modal distil- lation

    Antonin V obecky, David Hurych, Oriane Sim ´eoni, Spyros Gidaris, Andrei Bursuc, Patrick P ´erez, and Josef Sivic. Drive&segment: Unsupervised semantic segmentation of urban scenes via cross-modal distil- lation. In European Conference on Computer Vision, pages 478–495. Spring...

    2022

  24. [32]

    Pixood: Pixel-level out-of-distribution detection

    Tom ´aˇs V oj´ıˇr, Jan ˇSochman, and Ji ˇr´ı Matas. Pixood: Pixel-level out-of-distribution detection. In Euro- pean Conference on Computer Vision, pages 93–109. Springer, 2024. 1

    2024

  25. [33]

    Semflow: Bind- ing semantic segmentation and image synthesis via rectified flow

    Chaoyang Wang, Xiangtai Li, Lu Qi, Henghui Ding, Yunhai Tong, and Ming-Hsuan Yang. Semflow: Bind- ing semantic segmentation and image synthesis via rectified flow. Advances in Neural Information Pro- cessing Systems, 37:138981–139001, 2025. 1

    2025

  26. [34]

    Segrefiner: Towards model-agnostic segmentation refinement with discrete diffusion process

    Mengyu Wang, Henghui Ding, Jun Hao Liew, Jiajun Liu, Yao Zhao, and Yunchao Wei. Segrefiner: Towards model-agnostic segmentation refinement with discrete diffusion process. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. 2

    2023

  27. [35]

    Toward real ultra image segmentation: Leveraging surround- ing context to cultivate general segmentation model

    Sai Wang, Yutian Lin, Yu Wu, and Bo Du. Toward real ultra image segmentation: Leveraging surround- ing context to cultivate general segmentation model. Advances in Neural Information Processing Systems , 37:129227–129249, 2025. 1

    2025

  28. [36]

    Conditional boundary loss for semantic segmentation

    Dongyue Wu, Zilin Guo, Aoyan Li, Changqian Yu, Changxin Gao, and Nong Sang. Conditional boundary loss for semantic segmentation. IEEE Transactions on Image Processing, 32:3717–3731, 2023. 1

    2023

  29. [37]

    3d open-vocabulary panoptic segmen- tation with 2d-3d vision-language distillation

    Zihao Xiao, Longlong Jing, Shangxuan Wu, Alex Zi- hao Zhu, Jingwei Ji, Chiyu Max Jiang, Wei-Chih Hung, Thomas Funkhouser, Weicheng Kuo, Anelia Angelova, et al. 3d open-vocabulary panoptic segmen- tation with 2d-3d vision-language distillation. In Eu- ropean Conference on Compu...

    2024

  30. [38]

    Barlow twins: Self-supervised learn- ing via redundancy reduction

    Jure Zbontar, Li Jing, Ishan Misra, Yann LeCun, and St´ephane Deny. Barlow twins: Self-supervised learn- ing via redundancy reduction. In International con- ference on machine learning , pages 12310–12320. PMLR, 2021. 5

    2021

  31. [39]

    Faster segment anything: Towards lightweight sam for mobile applications, 2023

    Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, Seungkyu Lee, and Choong Seon Hong. Faster segment anything: Towards lightweight sam for mobile applications, 2023. 8

    2023

  32. [40]

    Csl: Class-agnostic structure- constrained learning for segmentation including the unseen

    Hao Zhang, Fang Li, Lu Qi, Ming-Hsuan Yang, and Narendra Ahuja. Csl: Class-agnostic structure- constrained learning for segmentation including the unseen. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 7078–7086, 2024. 2 10

    2024

  33. [41]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiao- gang Wang, and Jiaya Jia. Pyramid scene parsing network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2881– 2890, 2017. 2

    2017

  34. [42]

    Segment every out-of-distribution object

    Wenjie Zhao, Jia Li, Xin Dong, Yu Xiang, and Yun- hui Guo. Segment every out-of-distribution object. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 3910– 3920, 2024. 1

    2024

  35. [43]

    Fast segment anything

    Xu Zhao, Wenchao Ding, Yongqi An, Yinglong Du, Tao Yu, Min Li, Ming Tang, and Jinqiao Wang. Fast segment anything. arXiv preprint arXiv:2306.12156,

    arXiv

  36. [44]

    Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers

    Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. In Proceedings of the IEEE/CVF conference on ...

    2021

  37. [45]

    Deformable {detr}: De- formable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiao- gang Wang, and Jifeng Dai. Deformable {detr}: De- formable transformers for end-to-end object detection. In International Conference on Learning Representa- tions, 2021. 5 11

    2021

  38. [2024]

    BMV A, 2024. 2, 5, 6

    2024

Pith tools

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