Pith. sign in

REVIEW 3 major objections 6 minor 145 references

SemSegBench & DetecBench: Benchmarking Reliability and Generalization Beyond Classification

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Reliability and generalization do not reliably track clean-data accuracy in segmentation and detection models, a finding the paper supports with 6,139 evaluations of 137 models.

desk verdict A genuinely useful benchmark resource whose detection-side conclusions rest on a GAM3 metric that the paper itself admits is too low to distinguish the models. read the letter →

arxiv 2505.18015 v1 pith:JRAI5ACS submitted 2025-05-23 cs.CV cs.LG

Shashank Agnihotri , David Schader , Jonas Jakubassa , Nico Sharei , Simon Kral , Mehmet Ege Kaçar , Ruben Weber , Margret Keuper This is my paper
classification cs.CVcs.LG
keywords semanticsegmentationobjectdetectionrobustnessbenchmarkingadversarialattacksout-of-distributiongeneralizationcommoncorruptionsvisiontransformerbackbonesreliabilityandmeasures
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tackles a gap: robustness research has mostly been done on image classifiers, while the models actually deployed in safety-critical settings — semantic segmentation and object detection — have not been tested at scale. To close that gap, the authors build two open benchmarking tools and run 6,139 evaluations covering 76 segmentation models and 61 object detectors under adversarial attacks and synthetic image corruptions. The central claim is that reliability (performance under attack) and generalization (performance under distribution shift) do not reliably improve as clean-data accuracy improves: for object detection the two qualities are essentially uncorrelated with each other, and visible progress on i.i.d. benchmarks does not imply either. The paper additionally finds that transformer-based backbones generalize better to out-of-distribution data than CNN-based backbones in both tasks, and that synthetic corruptions track real-world adverse conditions closely enough to serve as a cheap proxy. If the claim stands, standard practice — judging progress by i.i.d. accuracy alone — understates how much dedicated design work is still needed before these models can be trusted in the field.

What carries the argument

Two summary metrics carry the analysis, each collapsing a large evaluation table into one number per model. The Reliability Measure (ReM) is the worst-case mIoU (segmentation) or mAP (detection) under a white-box adversarial attack, written $\ell_p$-ReM$^{\epsilon}_{\mathrm{iterations}}$ — for example $\ell_\infty$-ReM$^8_{20}$ denotes a 20-iteration attack with budget $\epsilon = 8/255$. The Generalization Ability Measure (GAM) is the worst mIoU or mAP across all synthetic corruptions at a fixed severity, denoted GAM3 for severity 3, computed over the 15 2D Common Corruptions for segmentation and over those plus eight 3D Common Corruptions for detection. With each model represented by (i.i.d. score, ReM, GAM), the paper computes Pearson correlations against i.i.d. performance, backbone type (CNN vs. vision transformer), and parameter count; those correlations are the machinery that converts thousands of individual evaluations into the paper's qualitative conclusions about when reliability and generalization do — or do not — follow accuracy.

What would settle it

Recompute the detection-side GAM3 after dropping the handful of corruptions (xy motion blur, z motion blur, 3D fog) that collapse every detector to near-zero mAP. If the strong i.i.d.-versus-GAM correlation survives but the claimed absence of a reliability-generalization correlation reverses, the detection conclusions are an artifact of the degenerate worst-case measure. Separately, run the same proxy validation the paper performs for segmentation — correlating synthetic corruption scores with a real adverse-conditions dataset — on object detectors, since the paper's own ACDC-style validation is done only for Cityscapes-trained segmentation models.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes three findings. First, for semantic segmentation, i.i.d. mIoU correlates strongly with reliability under $\ell_2$-norm attacks and with generalization as measured by GAM3, but standard $\ell_\infty$-norm attacks at $\epsilon = 8/255$ are so strong that nearly every model collapses, so reliability under that threat model carries almost no signal. Second, for object detection, i.i.d. mAP correlates strongly with generalization to corruptions, but reliability under adversarial attack correlates with neither accuracy nor generalization, and the worst-case generalization measure GAM3 is driven near zero for every detector by a small set of 3D corruptions (xy and z motion blur, and fog). Third, in both tasks transformer-based backbones generalize better out of distribution than CNN-based backbones, while parameter count matters only moderately and architecture choice matters more. The authors read these results as evidence that reliability and out-of-distribution generalization are not byproducts of chasing higher benchmark accuracy; each quality has to be engineered directly. They also report that performance on synthetic 2D corruptions correlates strongly with performance on the real adverse-conditions dataset ACDC, which they offer as the first large-scale support for treating synthetic corruptions as a proxy for real distribution shifts.

Load-bearing premise

The conclusions rest on the premise that one fixed list of synthetic corruptions, summarized by the worst score across that list, is a faithful measure of real-world generalization — even though for object detection a few 3D motion-blur corruptions push every detector's worst-case score to near zero.

Editorial extensions

If this is right

  • Model comparisons and leaderboards for segmentation and detection should report reliability and generalization alongside i.i.d. accuracy, since the two are not implied by accuracy.
  • Transformer-based backbones are the better default choice when deployment will face out-of-distribution conditions, with the advantage showing up consistently in both tasks.
  • Synthetic common corruptions can replace at least some real-world data collection for robustness screening of segmentation models: performance on them correlates with ACDC performance at Pearson 0.787 for mean corruption mIoU.
  • Because detection reliability and generalization are nearly uncorrelated, improving one (for example by adversarial training) should not be assumed to improve the other.
  • The released 6,139 pre-computed evaluations let the community test further robustness hypotheses without repeating the full, computationally expensive attack and corruption runs.

Reading between the lines

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

  • Editorial extension: the near-zero GAM3 for every detector implies the worst-case formulation is more useful for segmentation than for detection; a per-corruption or severity-curve report would separate universally hard 3D corruptions from the ones that actually discriminate between detectors.
  • Editorial extension: the decoupling the authors find for detection suggests that robustness certification for safety-critical use should be based on dedicated reliability and generalization tests rather than on clean-data accuracy thresholds.
  • Editorial extension: if the synthetic-proxy correlation holds beyond the single Cityscapes-to-ACDC pairing tested here, it would justify using synthetic corruption suites as a cheap screening stage in model selection, with real-world collection reserved for final validation.
  • Editorial extension: the interaction between backbone type and head design — attention-query detectors only show strong reliability with large Swin backbones — invites a controlled study that fixes the detector head and varies only the backbone to isolate where the robustness advantage originates.
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

3 major / 6 minor

Summary. This paper introduces SEMSEGBENCH and DETECBENCH, two benchmarking tools built on mmsegmentation and mmdetection, and uses them to evaluate 76 semantic segmentation models across four datasets and 61 object detectors across two datasets under adversarial attacks and common corruptions, for a total of 6139 logged evaluations. The paper defines a Reliability Measure (ReM) as worst-case performance under adversarial attacks and a Generalization Ability Measure (GAM) as worst-case performance across corruptions at a given severity, then reports correlations among i.i.d. performance, reliability, and generalization, with further analyses by backbone type, parameter count, and proposal method. The central claims are that reliability and generalization do not always improve with i.i.d. performance, that transformer-based backbones generalize better to OOD data for both tasks, and that synthetic corruptions can serve as a proxy for real-world distribution shifts.

Significance. The open-sourced tools and the 6139 pre-logged evaluations constitute a substantial community resource, and the paper goes well beyond classification-focused robustness benchmarks by covering dense prediction tasks with many architectures, backbones, and datasets. The segmentation-side correlation analyses and the ACDC-versus-synthetic-corruption proxy finding in Appendix A.1 are useful and generally well supported. The ReM and GAM definitions are transparent and do not involve fitted parameters. If the object-detection generalization analysis is reworked to avoid the degenerate worst-case GAM3, the paper would be a valuable reference for robustness research in semantic segmentation and object detection.

major comments (3)
  1. [§3.2, §4.2.1, Fig. 3a] For object detection, GAM3 is degenerate because a few 3D corruptions drive mAP to near zero for essentially all detectors. Appendix Fig. 31 shows that xy motion blur, z motion blur, and fog 3D reduce mAP to roughly 0.005–0.02 across the board, so the GAM3 values in Fig. 3a cluster in the range 0.002–0.018 with very little dynamic range. The paper itself states in Sec. 4.2.1 that “the low GAM3 scores hardly allow to draw any conclusions,” yet Fig. 3a uses GAM3 for the Pearson 0.810 i.i.d.-versus-generalization correlation, and Sec. 4.2.2 uses it for the transformer-versus-CNN generalization comparison. Because the ranking is determined mostly by which model fails slightly less on a floor-effect corruption, the detection-specific conclusions in Sec. 4.2 and Table 1 are not supported as stated. Please redo the detection generalization analysis with the mean mAP over corruptions (already reported in Appendix B.2) or with a variant that excludes or separately aggregates the floor-effect corruptions, and report how the conclusions change.
  2. [§3.2, Appendix K] The GAM definition is sensitive to the composition of the corruption set, and for detection the set is not stable. Appendix K reports that the h265 crf and h265 abr corruptions were computed for most models but excluded from GAM because they were computationally infeasible for Conditional-DETR and Co-DETR. With only eight 3D corruptions and several near-zero entries, adding or removing a single corruption can change GAM3 rankings materially. This is not a mere implementation detail: it means that the “fixed” corruption set underlying GAM3 is a choice that affects the object-detection findings. Please provide a stability analysis, such as leave-one-corruption-out rank correlations, and report the detection generalization results for the mean-mAP and 2D-only variants so that the sensitivity of the headline claims is visible.
  3. [§4.1.1, §4.2.1, Figs. 2a and 3a] The Pearson correlations are computed across models that heavily share backbones and architectures (for example, many ResNet50 and ResNet101 variants), so the observations are not independent, and no confidence intervals or significance tests are reported. This is especially problematic for the small-range GAM3 scatter in Fig. 3a, where a handful of points can determine the reported r = 0.810. Please report rank correlations and bootstrap confidence intervals, or explicitly discuss the non-independence of the samples as a limitation when interpreting the strength of these correlations.
minor comments (6)
  1. [§1, Contributions] The sentence “This paper clearly deserves acceptance.” at the end of the contribution list is not appropriate for a scientific manuscript and should be removed.
  2. [§4.2.1] The text first reports a weak positive correlation between i.i.d. performance and reliability under ℓ∞ attacks (Pearson 0.392 in Fig. 3a, left) and later concludes “there exists no correlation between performance and reliability”; this should be made consistent.
  3. [Fig. 3a] The middle panel of Fig. 3a labels the x-axis “i.i.d. mIoU,” but the object-detection analysis uses mAP; this appears to be a copy-paste error from Fig. 2a and should be corrected.
  4. [Fig. 3b] The axis labels in the supplied Fig. 3b (i.i.d. mIoU, ℓ2-ReM, GAM3) appear to be reused from the segmentation figure and should be replaced with the object-detection metrics (mAP, ℓ∞-ReM, GAM3).
  5. [References, [90]] Reference [90] is cited for the DINO object detector, but the reference given is Caron et al.'s self-supervised vision transformer paper; the correct citation is the DINO object detection paper by Zhang et al.
  6. [Throughout] There are several typos and incomplete sentences, e.g., “Both tools are now publicly.” in Sec. 1, “lipshitz” for “Lipschitz,” and “descent i.i.d. performance” for “decent i.i.d. performance”; these should be corrected in a revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark analysis is data-driven, with metrics defined straightforwardly and no prediction that reduces to a fitted input or self-citation chain.

full rationale

The paper's central claims are empirical findings from a large benchmark, not mathematical derivations from assumed conclusions. The Reliability Measure (ReM) and Generalization Ability Measure (GAM) are explicit definitions: ReM is the mIoU/mAP under a fixed white-box attack, and GAM is the worst mIoU/mAP across a fixed corruption set at a given severity. These definitions do not encode the paper's conclusions, such as 'reliability and generalization do not always improve with i.i.d. performance' or 'transformer backbones generalize better'; those conclusions are read off the resulting scatter plots and correlations. The use of CosPGD from the authors' prior work is a standard, publicly available attack method, and its inclusion is an experimental protocol choice rather than a load-bearing self-citation: the paper does not invoke CosPGD to prove a uniqueness theorem or to forbid alternative interpretations. The appendix claim that synthetic corruptions correlate with real-world ACDC performance is supported by independently measured correlations between held-out evaluations, not by construction. The detection GAM3 floor effect, where a few 3D corruptions drive all detectors near zero, is a legitimate validity concern about the chosen metric, but it is not circularity: the metric is defined before the analysis and the poor GAM3 values are unexpected empirical results, not inputs that force the stated conclusions. No fitted parameter is renamed as a prediction, and no equation reduces to another equation by definition. Therefore the derivation chain is self-contained with respect to the paper's stated contributions.

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

The benchmark introduces the ReM and GAM metrics, which are definitional and not fitted to data. The main free parameters are severity levels, attack budgets, step sizes, iteration counts, and the post-hoc exclusion of h265 corruptions. The analysis relies on domain assumptions about synthetic corruptions, white-box attacks, and model representativeness. No new physical or conceptual entities are postulated.

free parameters (7)
  • GAM severity level = 3
    All corruption evaluations use severity 3; this choice is not derived or justified in depth.
  • epsilon for l-infinity attacks = 8/255
    Standard adversarial budget from prior work; adopted without derivation.
  • epsilon for l2 attacks = 64
    Taken from the authors' CosPGD paper [46]; a hand-chosen budget.
  • alpha for l-infinity attacks = 0.01
    Step size used for all l-infinity attacks; from prior literature.
  • alpha for l2 attacks = 0.1
    Step size used for all l2 attacks; from prior literature.
  • attack iterations = 20
    Justified by saturation observations in prior work, but still a manual selection.
  • h265 corruption exclusion = excluded from GAM
    The h265 crf and h265 abr corruptions were not computed for two models and were then excluded from GAM for all models as a post-hoc fairness adjustment.
assumptions (3)
  • domain assumption Synthetic common corruptions are a reliable proxy for real-world distribution shifts
    The paper uses synthetic corruptions as the OOD measure throughout and supports this with correlations to ACDC in Appendix A.1, but correlation is not proof of proxy validity.
  • domain assumption White-box adversarial attacks indicate worst-case reliability
    The ReM metric is based solely on untargeted white-box attacks, following prior work; this is an adopted assumption rather than a proven property.
  • domain assumption The selected mmsegmentation and mmdetection checkpoints are representative of state-of-the-art methods
    The benchmark samples a subset of available architectures, backbones, and checkpoints, and conclusions about 'systematic weaknesses' assume this subset represents the broader model population.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SemSegBench & DetecBench: Benchmarking Reliability and Generalization Beyond Classification." pith.science (2026). https://pith.science/paper/JRAI5ACS

@misc{pith2026250518015,
  author       = {Pith},
  title        = {Pith review of: SemSegBench & DetecBench: Benchmarking Reliability and Generalization Beyond Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JRAI5ACS}},
  note         = {Machine review of arXiv:2505.18015}
}
read the original abstract

Reliability and generalization in deep learning are predominantly studied in the context of image classification. Yet, real-world applications in safety-critical domains involve a broader set of semantic tasks, such as semantic segmentation and object detection, which come with a diverse set of dedicated model architectures. To facilitate research towards robust model design in segmentation and detection, our primary objective is to provide benchmarking tools regarding robustness to distribution shifts and adversarial manipulations. We propose the benchmarking tools SEMSEGBENCH and DETECBENCH, along with the most extensive evaluation to date on the reliability and generalization of semantic segmentation and object detection models. In particular, we benchmark 76 segmentation models across four datasets and 61 object detectors across two datasets, evaluating their performance under diverse adversarial attacks and common corruptions. Our findings reveal systematic weaknesses in state-of-the-art models and uncover key trends based on architecture, backbone, and model capacity. SEMSEGBENCH and DETECBENCH are open-sourced in our GitHub repository (https://github.com/shashankskagnihotri/benchmarking_reliability_generalization) along with our complete set of total 6139 evaluations. We anticipate the collected data to foster and encourage future research towards improved model reliability beyond classification.

Figures

Figures reproduced from arXiv: 2505.18015 by the authors.

Figure 1
Figure 1. An overview of semantic segmentation (top) and object detection (bottom) methods [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Semantic Segmentation using the ADE20K dataset. The colors represent the architecture of the method, while the shapes of the markers represent the backbone of the respective method. All methods were trained on the train set of the ADE20K dataset. Please refer to the Appendix for results with other datasets i.e. Cityscapes and PASCAL VOC2012, additionally in the Appendix we show high correlation between performance a… view at source ↗
Figure 3
Figure 3. Object Detection using the MS-COCO dataset. The colors represent the backbone of the respective method, while different marker shapes represent the architecture of the method. All methods were trained on the train set of the MS-COCO dataset. The numbers in subcaptions for the respective subfigures are left to right. Second, we consider the correlation between i.i.d. performance and generalization ability to image co… view at source ↗
Figures from the paper (30 more)
Figure 4
Figure 4. Figure 4: To empirically determine if synthetic common corruptions such that those proposed by [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: To better understand the correlations from Figure 4, here we look at the correaltions between [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: To find correlations in observations across datasets, here we use the ADE20K dataset and [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Dataset used: ADE20K. The performance of semantic segmentation methods is usually captured across different metrics, namely: mIoU, mAcc, and aAcc. In the analysis of this work, we only used mIoU. Here we show that there is a high positive correlation between these metr…
Figure 9
Figure 9. Figure 9: We observe a very strong positive correlation between the different metrics: mIoU, mAcc, [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 8
Figure 8. Figure 8: Dataset used: Cityscapes. The performance of semantic segmentation methods is usually captured across different metrics, namely: mIoU, mAcc, and aAcc. In the analysis of this work, we only used mIoU. Here we show that there is a high positive correlation between these …
Figure 9
Figure 9. Figure 9: Dataset used: PASCAL VOC2012. The performance of semantic segmentation methods is usually captured across different metrics, namely: mIoU, mAcc, and aAcc. In the analysis of this work, we only used mIoU. Here we show that there is a high positive correlation between th…
Figure 10
Figure 10. Figure 10: Semantic Segmentation: Using the ADE20K dataset, here we analyze correlations in i.i.d. performance, reliability and generalization abilities of different methods. The colors represent the architecture of the method, while the shapes of the markers represent the backb…
Figure 11
Figure 11. Figure 11: Using the MS-COCO dataset, here we analyze i.i.d. performance, reliability, and [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]
Figure 12
Figure 12. Figure 12: Applying 3D Common Corruptions (y-axis) and 2D Common Corruptions (x-axis) to the [PITH_FULL_IMAGE:figures/full_fig_p027_12.png]
Figure 13
Figure 13. Figure 13: MS-COCO dataset is usually captured across different metrics, namely: mAP, [PITH_FULL_IMAGE:figures/full_fig_p028_13.png]
Figure 14
Figure 14. Figure 14: Illustrating changes in prediction due to different [PITH_FULL_IMAGE:figures/full_fig_p032_14.png]
Figure 15
Figure 15. Figure 15: F Model Zoo The trained checkpoints for all models available in SEMSEGBENCH can be obtained using the following lines of code: from semsegbench.evals import load_model model = load_model(model_name='upernet', backbone='InterImage-H', dataset='ADE20k') Users need to sp…
Figure 15
Figure 15. Figure 15: Illustrating changes in prediction due to different 2D Common Corruptions on a randomly [PITH_FULL_IMAGE:figures/full_fig_p035_15.png]
Figure 16
Figure 16. Figure 16: Dataset used: PASCAL VOC2012. The correlation in the performance of semantic segmentation methods against different attacks under the ℓ∞-norm and ℓ2-norm bounded attacks. The respective axis shows the name of the attack used. Colors are used to show different architec…
Figure 17
Figure 17. Figure 17: Dataset used: Cityscapes. The correlation in the performance of semantic segmentation methods against different attacks under the ℓ∞-norm and ℓ2-norm bounded attacks. The respective axis shows the name of the attack used. Colors are used to show different architecture…
Figure 18
Figure 18. Figure 18: Dataset used: ADE20K. The correlation in the performance of semantic segmentation methods against different attacks under the ℓ∞-norm and ℓ2-norm bounded attacks. The respective axis shows the name of the attack used. Colors are used to show different architectures an…
Figure 19
Figure 19. Figure 19: Dataset used: PASCAL VOC2012. Reliability of semantic segmentation methods against individual attacks. TOP: PGD, MIDDLE: SegPGD and BOTTOM: CosPGD, constrained under ℓ∞-norm (y-axis) and the ℓ2-norm (x-axis). 40 [PITH_FULL_IMAGE:figures/full_fig_p040_19.png]
Figure 20
Figure 20. Figure 20: Dataset used: Cityscapes. Reliability of semantic segmentation methods against individual attacks. TOP: PGD, MIDDLE: SegPGD and BOTTOM: CosPGD, constrained under ℓ∞-norm (y-axis) and the ℓ2-norm (x-axis). 41 [PITH_FULL_IMAGE:figures/full_fig_p041_20.png]
Figure 21
Figure 21. Figure 21: Dataset used: ADE20K. Reliability of semantic segmentation methods against individual attacks. TOP: PGD, MIDDLE: SegPGD and BOTTOM: CosPGD, constrained under ℓ∞-norm (y-axis) and the ℓ2-norm (x-axis). 42 [PITH_FULL_IMAGE:figures/full_fig_p042_21.png]
Figure 22
Figure 22. Figure 22: Dataset used: PASCAL VOC2012. The correlation in the performance of semantic segmentation methods against different 2D Common Corruptions. The respective axis shows the name of the common corruption used. Colors are used to show different architectures and marker styl…
Figure 23
Figure 23. Figure 23: Dataset used: Cityscapes. The correlation in the performance of semantic segmentation methods against different 2D Common Corruptions. The respective axis shows the name of the common corruption used. Colors are used to show different architectures and marker styles a…
Figure 24
Figure 24. Figure 24: Dataset used: ADE20K. The correlation in the performance of semantic segmentation methods against different 2D Common Corruptions. The respective axis shows the name of the common corruption used. Colors are used to show different architectures and marker styles are u…
Figure 25
Figure 25. Figure 25: Using DeFormable-DETR with a ConvNeXt-small Backbone and MS-COCO dataset we show for an example input image the changes in prediction under PGD attack, under Zoom Blur corruption from 2D Common Corruptions and under Fog corruption from 3D Common Corruptions. In each i…
Figure 26
Figure 26. Figure 26: Using the PASCAL VOC dataset for object detection, we benchmark the i.i.d. performance [PITH_FULL_IMAGE:figures/full_fig_p053_26.png]
Figure 27
Figure 27. Figure 27: Using the PASCAL VOC dataset for object detection, we benchmark the OOD performance [PITH_FULL_IMAGE:figures/full_fig_p054_27.png]
Figure 28
Figure 28. Figure 28: Using the MS-COCO dataset, left: correlation between the performance against FGSM [PITH_FULL_IMAGE:figures/full_fig_p054_28.png]
Figure 29
Figure 29. Figure 29: Using the MS-COCO dataset, here we report the mAP performance of all the object [PITH_FULL_IMAGE:figures/full_fig_p055_29.png]
Figure 30
Figure 30. Figure 30: Using the MS-COCO dataset, we report the mAP performance of all considered object [PITH_FULL_IMAGE:figures/full_fig_p056_30.png]
Figure 31
Figure 31. Figure 31: Using the MS-COCO dataset, we report the mAP performance of all considered object [PITH_FULL_IMAGE:figures/full_fig_p056_31.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

145 extracted references · 53 canonical work pages

  1. [96]

    Are Synthetic Corruptions A Reliable Proxy For Real-World Corruptions? In CVPR Workshop On Synthetic Data for Computer Vision, 2025

    Shashank Agnihotri, David Schader, Nico Sharei, Mehmet Ege Kaçar, and Margret Keuper. Are Synthetic Corruptions A Reliable Proxy For Real-World Corruptions? In CVPR Workshop On Synthetic Data for Computer Vision, 2025

    work page 2025

  2. [1]

    Semantic understanding of scenes through the ade20k dataset

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

    2019

  3. [2]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pages 740–755. Springer, 2014

    2014

  4. [3]

    Benchmarking neural network robustness to common corruptions and perturbations

    Dan Hendrycks and Thomas Dietterich. Benchmarking neural network robustness to common corruptions and perturbations. In Proc. International Conference on Learning Representations (ICLR), 2019

    2019

  5. [4]

    3d common corruptions and data augmentation

    O˘guzhan Fatih Kar, Teresa Yeo, Andrei Atanov, and Amir Zamir. 3d common corruptions and data augmentation. In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 18963–18974, 2022

    2022

  6. [5]

    What do compressed deep neural networks forget? arXiv preprint arXiv:1911.05248, 2019

    Sara Hooker, Aaron Courville, Gregory Clark, Yann Dauphin, and Andrea Frome. What do compressed deep neural networks forget? arXiv preprint arXiv:1911.05248, 2019

    arXiv 1911

  7. [6]

    Goodfellow, and Samy Bengio

    Alexey Kurakin, Ian J. Goodfellow, and Samy Bengio. Adversarial machine learning at scale. In Proc. International Conference on Learning Representations (ICLR), 2017

    2017

  8. [7]

    Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks

    Francesco Croce and Matthias Hein. Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks. In ICML, 2020

    2020

Show all 145 references
  1. [8]

    Attacking motion estimation with adver- sarial snow

    Jenny Schmalfuss, Lukas Mehl, and Andrés Bruhn. Attacking motion estimation with adver- sarial snow. In Proc. ECCV Workshop on Adversarial Robustness in the Real World (AROW), 2022

    2022

  2. [9]

    Medyolo: A medical image object detection framework

    Joseph Sobek, Jose R Medina Inojosa, Betsy J Medina Inojosa, SM Rassoulinejad-Mousavi, Gian Marco Conte, Francisco Lopez-Jimenez, and Bradley J Erickson. Medyolo: A medical image object detection framework. Journal of Imaging Informatics in Medicine, pages 1–9, 2024

    2024

  3. [10]

    Surgical instrument detection algorithm based on improved yolov7x

    Boping Ran, Bo Huang, Shunpan Liang, and Yulei Hou. Surgical instrument detection algorithm based on improved yolov7x. Sensors, 23(11):5037, 2023

    2023

  4. [11]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, pages 234–241. Springer, 2015

    2015

  5. [12]

    Using duck-net for polyp image segmentation

    Razvan-Gabriel Dumitru, Darius Peteleaza, and Catalin Craciun. Using duck-net for polyp image segmentation. Scientific reports, 13(1):9803, 2023

    2023

  6. [13]

    Object detection in autonomous vehicles: Status and open challenges

    Abhishek Balasubramaniam and Sudeep Pasricha. Object detection in autonomous vehicles: Status and open challenges. arXiv preprint arXiv:2201.07706, 2022

    2022 arXiv

  7. [14]

    Enhancing object detection in self-driving cars using a hybrid approach

    Sajjad Ahmad Khan, Hyun Jun Lee, and Huhnkuk Lim. Enhancing object detection in self-driving cars using a hybrid approach. Electronics, 12(13), 2023

    2023

  8. [15]

    Object scene flow for autonomous vehicles

    Moritz Menze and Andreas Geiger. Object scene flow for autonomous vehicles. In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3061– 3070, 2015

    2015

  9. [16]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In Proceedings of the IEEE conference on computer vision and pattern reco...

    2016

  10. [17]

    Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness

    Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A Wichmann, and Wieland Brendel. Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness. In International Conference on Learning Representations, 2018. 10

    2018

  11. [18]

    Towards Robust and Resilient Machine Learning

    Adarsh Prasad. Towards Robust and Resilient Machine Learning. PhD thesis, Carnegie Mellon University, 2022

    2022

  12. [19]

    Wichmann

    Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A. Wichmann. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2(11):665–673, nov 2020

    2020

  13. [20]

    RobustBench: a standardized adversarial robustness benchmark

    Francesco Croce, Maksym Andriushchenko, Vikash Sehwag, Edoardo Debenedetti, Nicolas Flammarion, Mung Chiang, Prateek Mittal, and Matthias Hein. RobustBench: a standardized adversarial robustness benchmark. In Advances in Neural Information Processing Systems (NeurIPS), 2021

    2021

  14. [21]

    Neural architecture design and robustness: A dataset

    Steffen Jung, Jovita Lukasik, and Margret Keuper. Neural architecture design and robustness: A dataset. In ICLR. OpenReview. net, 2023

    2023

  15. [22]

    Towards understanding adversarial robustness of optical flow networks

    Simon Schrodi, Tonmoy Saikia, and Thomas Brox. Towards understanding adversarial robustness of optical flow networks. In CVPR, pages 8916–8924, 2022

    2022

  16. [23]

    Improving feature stability during upsampling–spectral artifacts and the importance of spatial context

    Shashank Agnihotri, Julia Grabinski, and Margret Keuper. Improving feature stability during upsampling–spectral artifacts and the importance of spatial context. In European Conference on Computer Vision, pages 357–376. Springer, 2024

    2024

  17. [24]

    Beware of Aliases– Signal Preservation is Crucial for Robust Image Restoration

    Shashank Agnihotri, Julia Grabinski, Janis Keuper, and Margret Keuper. Beware of Aliases– Signal Preservation is Crucial for Robust Image Restoration. arXiv preprint arXiv:2304.14736, 2024

    2024 arXiv

  18. [25]

    Roll the dice: Monte carlo downsampling as a low-cost adversarial defence, 2024

    Shashank Agnihotri, Shashank Priyadarshi, Hendrik Sommerhoff, Julia Grabinski, Andreas Kolb, and Margret Keuper. Roll the dice: Monte carlo downsampling as a low-cost adversarial defence, 2024

    2024

  19. [26]

    Improving native CNN robustness with filter frequency regularization

    Jovita Lukasik, Paul Gavrikov, Janis Keuper, and Margret Keuper. Improving native CNN robustness with filter frequency regularization. Transactions on Machine Learning Research, 2023

    2023

  20. [27]

    Frequencylowcut pooling- plug and play against catastrophic overfitting

    Julia Grabinski, Steffen Jung, Janis Keuper, and Margret Keuper. Frequencylowcut pooling- plug and play against catastrophic overfitting. In European Conference on Computer Vision, pages 36–57. Springer, 2022

    2022

  21. [28]

    How do training methods influence the utilization of vision models? InNeurIPS 2024 workshop on Interpretable AI: Past, Present and Future, 2024

    Paul Gavrikov, Shashank Agnihotri, Margret Keuper, and Janis Keuper. How do training methods influence the utilization of vision models? InNeurIPS 2024 workshop on Interpretable AI: Past, Present and Future, 2024

    2024

  22. [29]

    MMSegmentation: Openmmlab semantic segmentation toolbox and benchmark

    MMSegmentation Contributors. MMSegmentation: Openmmlab semantic segmentation toolbox and benchmark. https://github.com/open-mmlab/mmsegmentation, 2020

    2020

  23. [30]

    MMDetection: Open mmlab detection toolbox and benchmark

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng, Chenchen Zhu, Tianheng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu, Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang,...

    1906 arXiv

  24. [31]

    Benchmarking the robustness of semantic segmen- tation models

    Christoph Kamann and Carsten Rother. Benchmarking the robustness of semantic segmen- tation models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8828–8838, 2020

    2020

  25. [32]

    Robust object detection in challenging weather conditions

    Himanshu Gupta, Oleksandr Kotlyar, Henrik Andreasson, and Achim J Lilienthal. Robust object detection in challenging weather conditions. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 7523–7532, 2024

    2024

  26. [33]

    Benchmarking robustness in object detection: Autonomous driving when winter is coming

    Claudio Michaelis, Benjamin Mitzkus, Robert Geirhos, Evgenia Rusak, Oliver Bringmann, Alexander S Ecker, Matthias Bethge, and Wieland Brendel. Benchmarking robustness in object detection: Autonomous driving when winter is coming. arXiv preprint arXiv:1907.07484, 2019. 11

    1907 arXiv

  27. [34]

    Robust object detection in extreme construction conditions

    Yuexiong Ding, Ming Zhang, Jia Pan, Jinxing Hu, and Xiaowei Luo. Robust object detection in extreme construction conditions. Automation in Construction, 165:105487, 2024

    2024

  28. [35]

    On the robustness of semantic segmentation models to adversarial attacks

    Anurag Arnab, Ondrej Miksik, and Philip HS Torr. On the robustness of semantic segmentation models to adversarial attacks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 888–897, 2018

    2018

  29. [36]

    Towards reliable evaluation and fast training of robust semantic segmentation models

    Francesco Croce, Naman D Singh, and Matthias Hein. Towards reliable evaluation and fast training of robust semantic segmentation models. In ECCV, 2024

    2024

  30. [37]

    Frod: Robust object detection for free

    Muhammad Awais, Weiming Zhuang, Lingjuan Lyu, and Sung-Ho Bae. Frod: Robust object detection for free. CoRR, 2023

    2023

  31. [38]

    Adversarially-aware robust object detector

    Ziyi Dong, Pengxu Wei, and Liang Lin. Adversarially-aware robust object detector. In European Conference on Computer Vision, pages 297–313. Springer, 2022

    2022

  32. [39]

    Adver- sarial examples for semantic segmentation and object detection

    Cihang Xie, Jianyu Wang, Zhishuai Zhang, Yuyin Zhou, Lingxi Xie, and Alan Yuille. Adver- sarial examples for semantic segmentation and object detection. In Proceedings of the IEEE international conference on computer vision, pages 1369–1378, 2017

    2017

  33. [40]

    Detection defenses: An empty promise against adversarial patch attacks on optical flow

    Erik Scheurer, Jenny Schmalfuss, Alexander Lis, and Andrés Bruhn. Detection defenses: An empty promise against adversarial patch attacks on optical flow. In Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024

    2024

  34. [41]

    A perturbation-constrained adversarial attack for evaluating the robustness of optical flow

    Jenny Schmalfuss, Philipp Scholze, and Andrés Bruhn. A perturbation-constrained adversarial attack for evaluating the robustness of optical flow. InProc. European Conference on Computer Vision (ECCV), LNCS 13682, pages 183–200, 2022

    2022

  35. [42]

    Distracting downpour: Adversarial weather attacks for motion estimation

    Jenny Schmalfuss, Lukas Mehl, and Andrés Bruhn. Distracting downpour: Adversarial weather attacks for motion estimation. In Proc. IEEE/CVF International Conference on Computer Vision (ICCV), pages 10106–10116, 2023

    2023

  36. [43]

    Towards class-wise robustness analysis

    Tejaswini Medi, Julia Grabinski, and Margret Keuper. Towards class-wise robustness analysis. arXiv preprint arXiv:2411.19853, 2024

    2024 arXiv

  37. [44]

    Explaining and harnessing adver- sarial examples

    Ian Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adver- sarial examples. In Proc. International Conference on Learning Representations (ICLR) , 2015

    2015

  38. [45]

    Adversarial examples in the physical world

    Alexey Kurakin, Ian J Goodfellow, and Samy Bengio. Adversarial examples in the physical world. In Artificial Intelligence Safety and Security, pages 99–112. Chapman and Hall/CRC, 2018

    2018

  39. [46]

    CosPGD: an efficient white-box adversarial attack for pixel-wise prediction tasks

    Shashank Agnihotri, Steffen Jung, and Margret Keuper. CosPGD: an efficient white-box adversarial attack for pixel-wise prediction tasks. In Proc. International Conference on Machine Learning (ICML), 2024

    2024

  40. [47]

    Segpgd: An effective and efficient adversarial attack for evaluating and boosting segmentation robustness

    Jindong Gu, Hengshuang Zhao, V olker Tresp, and Philip HS Torr. Segpgd: An effective and efficient adversarial attack for evaluating and boosting segmentation robustness. In ECCV, pages 308–325. Springer, 2022

    2022

  41. [48]

    Transferable adversarial attacks for image and video object detection

    Xingxing Wei, Siyuan Liang, Ning Chen, and Xiaochun Cao. Transferable adversarial attacks for image and video object detection. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19 , pages 954–960. International Joint Conferen...

    2019

  42. [49]

    Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshmi- narayanan

    Dan Hendrycks, Norman Mu, Ekin D. Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshmi- narayanan. AugMix: A simple data processing method to improve robustness and uncertainty. Proc. of the International Conference on Learning Representations (ICLR), 2020

    2020

  43. [50]

    Towards improving robustness of compressed cnns

    J Hoffmann, S Agnihotri, Tonmoy Saikia, and Thomas Brox. Towards improving robustness of compressed cnns. In ICML Workshop on Uncertainty and Robustness in Deep Learning (UDL), 2021. 12

    2021

  44. [51]

    Robustspring: Benchmarking robustness to image corruptions for optical flow, scene flow and stereo

    Jenny Schmalfuss, Victor Oei, Lukas Mehl, Madlen Bartsch, Shashank Agnihotri, Margret Keuper, and Andrés Bruhn. Robustspring: Benchmarking robustness to image corruptions for optical flow, scene flow and stereo. arXiv preprint arXiv:2505.09368, 2025

    2025 arXiv

  45. [52]

    ACDC: The adverse conditions dataset with correspondences for semantic driving scene understanding

    Christos Sakaridis, Dengxin Dai, and Luc Van Gool. ACDC: The adverse conditions dataset with correspondences for semantic driving scene understanding. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2021

    2021

  46. [53]

    Torr, and Dacheng Tao

    Shiyu Tang, Ruihao Gong, Yan Wang, Aishan Liu, Jiakai Wang, Xinyun Chen, Feng- wei Yu, Xianglong Liu, Dawn Song, Alan Yuille, Philip H.S. Torr, and Dacheng Tao. Robustart: Benchmarking robustness on architecture design and training techniques. https://arxiv.org/pdf/2109.05211....

    2021 arXiv

  47. [54]

    Intra-& extra-source exemplar- based style synthesis for improved domain generalization

    Yumeng Li, Dan Zhang, Margret Keuper, and Anna Khoreva. Intra-& extra-source exemplar- based style synthesis for improved domain generalization. International Journal of Computer Vision, 132(2):446–465, 2024

    2024

  48. [55]

    Implicit representations for image segmentation

    Jan Philipp Schneider, Fatima Mishal, Jovita Lukasik, Andreas Kolb, Margret Keuper, and Michael Moeller. Implicit representations for image segmentation. In First Workshop on Unifying Representations in Neural Models, pages 1–6. OpenReview. net, 2022

    2022

  49. [56]

    Implicit representations for constrained image segmentation

    Jan Philipp Schneider, Mishal Fatima, Jovita Lukasik, Andreas Kolb, Margret Keuper, and Michael Moeller. Implicit representations for constrained image segmentation. In Proceedings of the 41st International Conference on Machine Learning (ICML) , pages 43765–43790. MLResearchP...

    2024

  50. [57]

    Torchattacks: A pytorch repository for adversarial attacks

    Hoki Kim. Torchattacks: A pytorch repository for adversarial attacks. arXiv preprint arXiv:2010.01950, 2020

    2010 arXiv

  51. [58]

    Foolbox native: Fast adversarial attacks to benchmark the robustness of machine learning models in pytorch, tensorflow, and jax

    Jonas Rauber, Roland Zimmermann, Matthias Bethge, and Wieland Brendel. Foolbox native: Fast adversarial attacks to benchmark the robustness of machine learning models in pytorch, tensorflow, and jax. Journal of Open Source Software, 5(53):2607, 2020

    2020

  52. [59]

    Segmentmeifyoucan: A benchmark for anomaly segmentation

    Robin Chan, Krzysztof Lis, Svenja Uhlemeyer, Hermann Blum, Sina Honari, Roland Siegwart, Pascal Fua, Mathieu Salzmann, and Matthias Rottmann. Segmentmeifyoucan: A benchmark for anomaly segmentation. In J. Vanschoren and S. Yeung, editors, Proceedings of the Neural Information ...

    2021

  53. [60]

    DispBench

    Shashank Agnihotri, Amaan Ansari, Annika Dackermann, Fabian Rösch, and Margret Keuper. DispBench. In CVPR Workshop On Synthetic Data for Computer Vision, 2025

    2025

  54. [61]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In CVPR, pages 2881–2890, 2017

    2017

  55. [62]

    Improving 2d feature representations by 3d-aware fine-tuning

    Yuanwen Yue, Anurag Das, Francis Engelmann, Siyu Tang, and Jan Eric Lenssen. Improving 2d feature representations by 3d-aware fine-tuning. In European Conference on Computer Vision, pages 57–74. Springer, 2024

    2024

  56. [63]

    Mta-clip: Language-guided semantic segmentation with mask-text alignment

    Anurag Das, Xinting Hu, Li Jiang, and Bernt Schiele. Mta-clip: Language-guided semantic segmentation with mask-text alignment. In European Conference on Computer Vision, pages 39–56. Springer, 2024

    2024

  57. [64]

    Faster R-CNN: Towards real-time object detection with region proposal networks

    Shaoqing Ren. Faster R-CNN: Towards real-time object detection with region proposal networks. arXiv preprint arXiv:1506.01497, 2015

    2015 arXiv

  58. [65]

    Focal loss for dense object detection

    T Lin. Focal loss for dense object detection. arXiv preprint arXiv:1708.02002, 2017

    2017 arXiv

  59. [66]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017

    2017

  60. [67]

    Cascade r-cnn: High quality object detection and instance segmentation

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: High quality object detection and instance segmentation. IEEE transactions on pattern analysis and machine intelligence , 43(5):1483–1498, 2019. 13

    2019

  61. [68]

    Dn-detr: Accelerate detr training by introducing query denoising

    Feng Li, Hao Zhang, Shilong Liu, Jian Guo, Lionel M Ni, and Lei Zhang. Dn-detr: Accelerate detr training by introducing query denoising. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13619–13627, 2022

    2022

  62. [69]

    Fooling the eyes of autonomous vehicles: Robust physical adversarial examples against traffic sign recognition systems

    Wei Jia, Zhaojun Lu, Haichun Zhang, Zhenglin Liu, Jie Wang, and Gang Qu. Fooling the eyes of autonomous vehicles: Robust physical adversarial examples against traffic sign recognition systems. In Network and Distributed Systems Security (NDSS) Symposium , number 1 in V olume 2...

    2022

  63. [70]

    Context-aware transfer attacks for object detection

    Zikui Cai, Xinxin Xie, Shasha Li, Mingjun Yin, Chengyu Song, Srikanth V Krishnamurthy, Amit K Roy-Chowdhury, and M Salman Asif. Context-aware transfer attacks for object detection. In Proceedings of the AAAI Conference on Artificial Intelligence , V olume 36, Number 1, pages 1...

    2022

  64. [71]

    T-sea: Transfer- based self-ensemble attack on object detection

    Hao Huang, Ziyan Chen, Huanran Chen, Yongtao Wang, and Kevin Zhang. T-sea: Transfer- based self-ensemble attack on object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 20514–20523, 2023

    2023

  65. [72]

    Towards deep learning models resistant to adversarial attacks, 2017

    Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks, 2017

    2017

  66. [73]

    Zico Kolter

    Eric Wong, Leslie Rice, and J. Zico Kolter. Fast is better than free: Revisiting adversarial training. ArXiv, abs/2001.03994, 2020

    2001 arXiv

  67. [74]

    On the unreasonable vulnerability of transformers for image restoration-and an easy fix

    Shashank Agnihotri, Kanchana Vaishnavi Gandikota, Julia Grabinski, Paramanand Chan- dramouli, and Margret Keuper. On the unreasonable vulnerability of transformers for image restoration-and an easy fix. In Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV...

    2023

  68. [75]

    Sp2 net for generalized zero-label semantic segmentation

    Anurag Das, Yongqin Xian, Yang He, Bernt Schiele, and Zeynep Akata. Sp2 net for generalized zero-label semantic segmentation. In DAGM German Conference on Pattern Recognition, pages 235–249. Springer, 2021

    2021

  69. [76]

    Dynamic divide-and-conquer adversarial train- ing for robust semantic segmentation

    Xiaogang Xu, Hengshuang Zhao, and Jiaya Jia. Dynamic divide-and-conquer adversarial train- ing for robust semantic segmentation. In IEEE/CVF International Conference on Computer Vision (ICCV), pages 7466–7475, 2021

    2021

  70. [77]

    Rethinking atrous convolution for semantic image segmentation, 2017

    Liang-Chieh Chen, George Papandreou, Florian Schroff, and Hartwig Adam. Rethinking atrous convolution for semantic image segmentation, 2017

    2017

  71. [78]

    Segformer: Simple and efficient design for semantic segmentation with transformers.Advances in neural information processing systems, 34:12077–12090, 2021

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M Alvarez, and Ping Luo. Segformer: Simple and efficient design for semantic segmentation with transformers.Advances in neural information processing systems, 34:12077–12090, 2021

    2021

  72. [79]

    Internimage: Exploring large-scale vision foundation models with deformable convolutions

    Wenhai Wang, Jifeng Dai, Zhe Chen, Zhenhang Huang, Zhiqi Li, Xizhou Zhu, Xiaowei Hu, Tong Lu, Lewei Lu, Hongsheng Li, et al. Internimage: Exploring large-scale vision foundation models with deformable convolutions. In Proceedings of the IEEE/CVF conference on computer vision a...

    2023

  73. [80]

    Deep residual learning for image recognition

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

    2016

  74. [81]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proc. of the IEEE/CVF conference on computer vision and pattern recognition, pages 11976–11986, 2022

    2022

  75. [82]

    Swin 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 transformer: Hierarchical vision transformer using shifted windows. In Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021

    2021

  76. [83]

    Vision transformers are robust learners

    Sayak Paul and Pin-Yu Chen. Vision transformers are robust learners. In Proceedings of the AAAI conference on Artificial Intelligence, V olume 36, Number 2, pages 2071–2081, 2022. 14

    2022

  77. [84]

    Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9924–9935, 2022

    2022

  78. [85]

    DETRs with collaborative hybrid assignments training

    Zhuofan Zong, Guanglu Song, and Yu Liu. DETRs with collaborative hybrid assignments training. In Proceedings of the IEEE/CVF international conference on computer vision, pages 6748–6758, 2023

    2023

  79. [86]

    Dense distinct query for end-to-end object detection

    Shilong Zhang, Xinjiang Wang, Jiaqi Wang, Jiangmiao Pang, Chengqi Lyu, Wenwei Zhang, Ping Luo, and Kai Chen. Dense distinct query for end-to-end object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7329–7338, 2023

    2023

  80. [87]

    Grounded language-image pre-training

    Liunian Harold Li, Pengchuan Zhang, Haotian Zhang, Jianwei Yang, Chunyuan Li, Yiwu Zhong, Lijuan Wang, Lu Yuan, Lei Zhang, Jenq-Neng Hwang, et al. Grounded language-image pre-training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages ...

    2022

  81. [88]

    Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection

    Shifeng Zhang, Cheng Chi, Yongqiang Yao, Zhen Lei, and Stan Z Li. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9759–9768, 2020

    2020

  82. [89]

    An open and comprehensive pipeline for unified object grounding and detection

    Xiangyu Zhao, Yicheng Chen, Shilin Xu, Xiangtai Li, Xinjiang Wang, Yining Li, and Haian Huang. An open and comprehensive pipeline for unified object grounding and detection. arXiv preprint arXiv:2401.02361, 2024

    2024 arXiv

  83. [90]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the International Conference on Computer Vision (ICCV), 2021

    2021

  84. [91]

    Rtmdet: An empirical study of designing real-time object detectors

    Chengqi Lyu, Wenwei Zhang, Haian Huang, Yue Zhou, Yudong Wang, Yanyi Liu, Shilong Zhang, and Kai Chen. Rtmdet: An empirical study of designing real-time object detectors. arXiv preprint arXiv:2212.07784, 2022

    2022 arXiv

  85. [92]

    Aggregated residual transformations for deep neural networks

    Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregated residual transformations for deep neural networks. In Proc. of the IEEE conference on computer vision and pattern recognition, pages 1492–1500, 2017

    2017

  86. [93]

    Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks

    Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dryden, and Alexandra Peste. Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks. Journal of Machine Learning Research, 22(241):1–124, 2021

    2021

  87. [94]

    Classification robustness to common optical aberrations

    Patrick Müller, Alexander Braun, and Margret Keuper. Classification robustness to common optical aberrations. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3632–3643, 2023

    2023

  88. [95]

    Robustifying token attention for vision transformers

    Yong Guo, David Stutz, and Bernt Schiele. Robustifying token attention for vision transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 17557– 17568, 2023

    2023

  89. [97]

    Urban scene semantic segmentation with low-cost coarse annotation

    Anurag Das, Yongqin Xian, Yang He, Zeynep Akata, and Bernt Schiele. Urban scene semantic segmentation with low-cost coarse annotation. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 5978–5987, 2023

    2023

  90. [98]

    Everingham, L

    M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2012 (VOC2012) Results. http://www.pascal- network.org/challenges/VOC/voc2012/workshop/index.html, 2012. 15

    2012

  91. [99]

    Center- net: Keypoint triplets for object detection

    Kaiwen Duan, Song Bai, Lingxi Xie, Honggang Qi, Qingming Huang, and Qi Tian. Center- net: Keypoint triplets for object detection. In Proceedings of the IEEE/CVF international conference on computer vision, pages 6569–6578, 2019

    2019

  92. [100]

    Tood: Task- aligned one-stage object detection

    Chengjian Feng, Yujie Zhong, Yu Gao, Matthew R Scott, and Weilin Huang. Tood: Task- aligned one-stage object detection. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 3490–3499. IEEE Computer Society, 2021

    2021

  93. [101]

    Yolo9000: Better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: Better, faster, stronger. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6517–6525, 2017

    2017

  94. [102]

    Yolox: Exceeding yolo series in 2021

    Z Ge. Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430, 2021

    2021 arXiv

  95. [103]

    Yolov6: A single-stage object detection framework for industrial applications

    Chuyi Li, Lulu Li, Hongliang Jiang, Kaiheng Weng, Yifei Geng, Liang Li, Zaidan Ke, Qingyuan Li, Meng Cheng, Weiqiang Nie, et al. Yolov6: A single-stage object detection framework for industrial applications. arXiv preprint arXiv:2209.02976, 2022

    2022 arXiv

  96. [104]

    Yolov7: Trainable bag- of-freebies sets new state-of-the-art for real-time object detectors

    Chien-Yao Wang, Alexey Bochkovskiy, and Hong-Yuan Mark Liao. Yolov7: Trainable bag- of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7464–7475, 2023

    2023

  97. [105]

    Glenn Jocher, K Nishimura, T Mineeva, and RJAM Vilariño. yolov5. Code repository, 2, 2020

    2020

  98. [106]

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

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2016

    2016

  99. [107]

    Yolov4: Optimal speed and accuracy of object detection

    Alexey Bochkovskiy, Chien-Yao Wang, and Hong-Yuan Mark Liao. Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934, 2020

    2004 arXiv

  100. [108]

    Yolov3: An incremental improvement, 2018

    Joseph Redmon and Ali Farhadi. Yolov3: An incremental improvement, 2018

    2018

  101. [109]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In European conference on computer vision, pages 213–229. Springer, 2020

    2020

  102. [110]

    Hypercolumns for object segmentation and fine-grained localization

    Bharath Hariharan, Pablo Arbeláez, Ross Girshick, and Jitendra Malik. Hypercolumns for object segmentation and fine-grained localization. In CVPR, pages 447–456, 2015

    2015

  103. [111]

    Hengshuang Zhao. semseg. https://github.com/hszhao/semseg, 2019

    2019

  104. [112]

    Semantic contours from inverse detectors

    Bharath Hariharan, Pablo Arbelaez, Lubomir Bourdev, Subhransu Maji, and Jitendra Malik. Semantic contours from inverse detectors. In International Conference on Computer Vision (ICCV), 2011

    2011

  105. [113]

    Schwing, Alexander Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. In CVPR, 2022

    2022

  106. [114]

    Flowbench: A robustness benchmark for optical flow estimation, 2025

    Shashank Agnihotri, Julian Yuya Caspary, Luca Schwarz, Xinyan Gao, Jenny Schmalfuss, Andres Bruhn, and Margret Keuper. Flowbench: A robustness benchmark for optical flow estimation, 2025

    2025

  107. [115]

    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 conference on computer vision (ECCV), pages 801–818, 2018

    2018

  108. [116]

    Beit: Bert pre-training of image transformers

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. Beit: Bert pre-training of image transformers. arXiv preprint arXiv:2106.08254, 2021

    2021 arXiv

  109. [117]

    Unified perceptual parsing for scene understanding

    Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene understanding. In Proceedings of the European conference on computer vision (ECCV), pages 418–434, 2018. 16

    2018

  110. [118]

    Unified perceptual parsing for scene understanding

    Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and Jian Sun. Unified perceptual parsing for scene understanding. In European Conference on Computer Vision. Springer, 2018

    2018

  111. [119]

    Achieving adversarial robustness via sparsity

    Ningyi Liao, Shufan Wang, Liyao Xiang, Nanyang Ye, Shuo Shao, and Pengzhi Chu. Achieving adversarial robustness via sparsity. Machine Learning, pages 1–27, 2022

    2022

  112. [120]

    Sparsity winning twice: Better robust generalization from more efficient training

    Tianlong Chen, Zhenyu Zhang, pengjun wang, Santosh Balachandra, Haoyu Ma, Zehao Wang, and Zhangyang Wang. Sparsity winning twice: Better robust generalization from more efficient training. In International Conference on Learning Representations, 2022

    2022

  113. [121]

    Theoretical explanation of activation sparsity through flat minima and adversarial robustness

    Ze Peng, Lei Qi, Yinghuan Shi, and Yang Gao. Theoretical explanation of activation sparsity through flat minima and adversarial robustness. arXiv preprint arXiv:2309.03004, 2023

    2023 arXiv

  114. [122]

    Everingham, L

    M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The pascal visual object classes (voc) challenge. International Journal of Computer Vision, 88(2):303–338, June 2010

    2010

  115. [123]

    LVIS: A dataset for large vocabulary instance segmentation

    Agrim Gupta, Piotr Dollar, and Ross Girshick. LVIS: A dataset for large vocabulary instance segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019

    2019

  116. [124]

    Conditional detr for fast training convergence

    Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, and Jingdong Wang. Conditional detr for fast training convergence. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2021

    2021

  117. [125]

    Dab-detr: Dynamic anchor boxes are better queries for detr

    Shilong Liu, Feng Li, Hao Zhang, Xiao Yang, Xianbiao Qi, Hang Su, Jun Zhu, and Lei Zhang. Dab-detr: Dynamic anchor boxes are better queries for detr. In International Conference on Learning Representations, 2022

    2022

  118. [126]

    Disen- tangle your dense object detector

    Zehui Chen, Chenhongyi Yang, Qiaofei Li, Feng Zhao, Zheng-Jun Zha, and Feng Wu. Disen- tangle your dense object detector. In Proceedings of the 29th ACM international conference on multimedia, pages 4939–4948, 2021

    2021

  119. [127]

    Deformable detr: Deformable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. In International Conference on Learning Representations, 2021

    2021

  120. [128]

    Rethinking classification and localization for object detection

    Yue Wu, Yinpeng Chen, Lu Yuan, Zicheng Liu, Lijuan Wang, Hongzhi Li, and Yun Fu. Rethinking classification and localization for object detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10186–10195, 2020

    2020

  121. [129]

    Dynamic r-cnn: Towards high quality object detection via dynamic training

    Hongkai Zhang, Hong Chang, Bingpeng Ma, Naiyan Wang, and Xilin Chen. Dynamic r-cnn: Towards high quality object detection via dynamic training. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16 , pages 260–275. Spr...

    2020

  122. [130]

    Fully convolu- tional one-stage 3d object detection on lidar range images

    Zhi Tian, Xiangxiang Chu, Xiaoming Wang, Xiaolin Wei, and Chunhua Shen. Fully convolu- tional one-stage 3d object detection on lidar range images. In Advances in Neural Information Processing Systems, volume 35, pages 34899–34911, 2022

    2022

  123. [131]

    Foveabox: Beyound anchor-based object detection

    Tao Kong, Fuchun Sun, Huaping Liu, Yuning Jiang, Lei Li, and Jianbo Shi. Foveabox: Beyound anchor-based object detection. IEEE Transactions on Image Processing, 29:7389– 7398, 2020

    2020

  124. [132]

    Freeanchor: Learning to match anchors for visual object detection

    Xiaosong Zhang, Fang Wan, Chang Liu, Rongrong Ji, and Qixiang Ye. Freeanchor: Learning to match anchors for visual object detection. Advances in neural information processing systems, 32, 2019

    2019

  125. [133]

    Feature selective anchor-free module for single-shot object detection

    Chenchen Zhu, Yihui He, and Marios Savvides. Feature selective anchor-free module for single-shot object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 840–849, 2019

    2019

  126. [134]

    Region proposal by guided anchoring

    Jiaqi Wang, Kai Chen, Shuo Yang, Chen Change Loy, and Dahua Lin. Region proposal by guided anchoring. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2965–2974, 2019. 17

    2019

  127. [135]

    Grid r-cnn

    Xin Lu, Buyu Li, Yuxin Yue, Quanquan Li, and Junjie Yan. Grid r-cnn. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7363–7372, 2019

    2019

  128. [136]

    Libra r-cnn: Towards balanced learning for object detection

    Jiangmiao Pang, Kai Chen, Jianping Shi, Huajun Feng, Wanli Ouyang, and Dahua Lin. Libra r-cnn: Towards balanced learning for object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 821–830, 2019

    2019

  129. [137]

    Probabilistic anchor assignment with iou prediction for object detection

    Kang Kim and Hee Seok Lee. Probabilistic anchor assignment with iou prediction for object detection. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXV 16, pages 355–371. Springer, 2020

    2020

  130. [138]

    Reppoints: Point set repre- sentation for object detection

    Ze Yang, Shaohui Liu, Han Hu, Liwei Wang, and Stephen Lin. Reppoints: Point set repre- sentation for object detection. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9657–9666, 2019

    2019

  131. [139]

    Side-aware boundary localization for more precise object detection

    Jiaqi Wang, Wenwei Zhang, Yuhang Cao, Kai Chen, Jiangmiao Pang, Tao Gong, Jianping Shi, Chen Change Loy, and Dahua Lin. Side-aware boundary localization for more precise object detection. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, ...

    2020

  132. [140]

    Sparse r-cnn: End-to-end object detection with learnable proposals

    Peize Sun, Rufeng Zhang, Yi Jiang, Tao Kong, Chenfeng Xu, Wei Zhan, Masayoshi Tomizuka, Lei Li, Zehuan Yuan, Changhu Wang, et al. Sparse r-cnn: End-to-end object detection with learnable proposals. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogn...

    2021

  133. [141]

    Varifocalnet: An iou-aware dense object detector

    Haoyang Zhang, Ying Wang, Feras Dayoub, and Niko Sunderhauf. Varifocalnet: An iou-aware dense object detector. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8514–8523, 2021. 18 SEMSEGBENCH & DETEC BENCH : Benchmarking Reliability a...

    2021

  134. [142]

    Each model is trained on the training dataset of the Cityscapes dataset

    truly represent the distribution and domain shifts in the real world we try to find correlations in evaluations on ACDC and 2D Common Corruptions. Each model is trained on the training dataset of the Cityscapes dataset. The y-axis represents values from evaluations on the ACDC...

  135. [144]

    The third row is when using ℓ2-norm bounded CosPGD attack with ϵ = 8

  136. [145]

    threat_model

    The fourth row is when using 2D Common Corruptions, here we calculate the mean for each metric across all 2D Common Corruptions. Colors are used to show different architectures and marker styles are used to show different backbones used by the semantic segmentation methods. B ...

  137. [255]

    The fourth row is when using 2D Common Corruptions, here we calculate the mean for each metric across all 2D Common Corruptions

    The third row is when using ℓ2-norm bounded CosPGD attack with ϵ = 8 255. The fourth row is when using 2D Common Corruptions, here we calculate the mean for each metric across all 2D Common Corruptions. Colors are used to show different architectures and marker styles are used...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.