Contrastive Mask Fidelity: Reference-Free Auditing of Ground-Truth Masks in Remote Sensing Semantic Segmentation

Jie Chen; Jing Ouyang; Meng Tang; Min Huang; Shuaishuai Cao; Shuwei Peng; Youjin Wang; Zhiwei Zhai

REVIEW 3 major objections 4 minor 34 references

Contrastive Mask Fidelity: Reference-Free Auditing of Ground-Truth Masks in Remote Sensing Semantic Segmentation

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A contrastive keep–erase score lets a frozen vision-language judge decide which of two competing segmentation masks better matches the image, without designating either as ground truth.

desk verdict Fresh, well-executed reference-free mask audit with a real area-sensitivity flaw and a shared-CLIMB confound, but the evidence chain is strong enough to warrant a serious referee. read the letter →

arxiv 2608.09101 v1 pith:KJZVTC7J submitted 2026-08-10 cs.CV cs.LG

Shuaishuai Cao , Shuwei Peng , Meng Tang , Min Huang , Youjin Wang , Jie Chen , Jing Ouyang , Zhiwei Zhai This is my paper · ORCID

classification cs.CVcs.LG

keywords semanticsegmentationremotesensingmaskfidelityreference-freeevaluationvision-languagemodelannotationauditground-truthqualityopen-vocabulary

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

Semantic segmentation benchmarks are scored against human-drawn masks, but in remote sensing those masks are often coarse or misaligned, so a high overlap score can mean agreement with a flawed label rather than fidelity to the image. This paper tries to break that evaluation paradox with Contrastive Mask Fidelity (CMF): a training-free, reference-free score that asks a frozen vision-language judge whether class evidence is concentrated inside a mask and absent outside it. The paper claims CMF can decide which of two competing masks is more faithful without trusting either, and supports that claim by matching a blinded three-annotator expert consensus on 81% of adjudicated pairs and by exposing systematic annotation distortion, with man-made classes like buildings, roads, and cars favoring the model mask on most pairs while ambiguous land cover favors human annotations. If the claim holds, overlap-based evaluation can be complemented by an image-grounded audit that reports annotation-noise headroom per class and supplies safer supervision for cross-domain training.

What carries the argument

The load-bearing mechanism is the keep–erase contrast: for each binary class mask $M_c$, the paper composites a keep view $x_k^c=M_c\odot x+(1-M_c)\odot B(x)$ and an erase view $x_e^c=(1-M_c)\odot x+M_c\odot B(x)$, with $B(\cdot)$ a Gaussian blur that preserves local color statistics, then scores both views with a frozen remote-sensing vision-language judge through a temperature-scaled softmax over the dataset's class texts. The fidelity score $F(M_c)$ is the difference of the judge's class probability on the two views, so purity and completeness are tested in one number, and the contrast is designed to reduce area sensitivity, with the paper reporting the area correlation dropping from $\rho=0.61$ to $\rho=0.08$. The same frozen judge scores the human mask and the candidate mask symmetrically, which is what makes the arbitration reference-free.

What would settle it

Take a set of images with independently verified high-resolution ground truth, produce deliberately distorted masks, and check whether CMF ranks the correct mask first on a large majority of pairs; also replace the CLIP-style judge with a judge trained under a different alignment objective and see whether the per-class win-rate profile and the 81% expert agreement survive. If a non-CLIP judge reverses the verdicts, or if CMF prefers geometrically loose but semantically plausible masks over boundary-correct ones in the verified set, the reference-free fidelity claim would fail.

Watch

Extended reading notes

Core claim

CMF is defined per class as $F(M_c)=\mathbb{P}(c\mid x_k)-\mathbb{P}(c\mid x_e)$, where $x_k$ keeps only the region inside mask $M_c$ with the rest blurred and $x_e$ erases it by blurring the mask area, and $\mathbb{P}$ is a frozen remote-sensing vision-language judge's class probability over the dataset's class texts. A faithful mask must score high on both purity, meaning the kept region depicts the class, and completeness, meaning no class evidence remains outside, and the contrast between the two views cancels trivial area inflation because enlarging a mask pushes the two terms in opposite directions. The paper's empirical claim is that this score arbitrates real disagreements: on 10,731 image-class pairs across ten remote-sensing benchmarks, CMF's verdicts match expert consensus more often than keep-only scoring, model confidence, or a trained label-quality baseline, and conservative class-wise arbitration of its verdicts improves cross-domain transfer over raw annotations, pure predictions, and matched random- or confidence-based replacements.

Load-bearing premise

The whole result rests on the assumption that the frozen vision-language judge's preference for one mask over another reflects genuine visual evidence in the image rather than shared pretraining biases, since the judge and the candidate generator both come from the CLIP-style model family and could therefore agree for reasons unrelated to true fidelity.

Editorial extensions

If this is right

Benchmarks can be audited without a trusted reference: any two competing masks, including a human annotation and an open-vocabulary prediction, can be scored directly against the image.
Annotation noise becomes reportable per class, so the audit can show on which classes human labels are systematically distorted and on which they carry information the image alone does not.
CMF verdicts provide usable supervision: conservatively replacing the human label on pairs with $\Delta F>0$ improves cross-domain mIoU by about 1.2 to 1.6 points over raw labels and beats coverage-matched confidence replacement, indicating the gain comes from selection rather than pseudo-label distillation.
Because CMF is training-free and needs only four judge passes per image-class pair, roughly 16 minutes for the full 10,731-pair audit on one GPU, the audit can scale to large datasets.
The arbitration margin acts as a precision–coverage dial: raising $\delta$ trades replacement coverage for per-replacement expert agreement without losing most of the transfer gain, so CMF can serve as a triage layer in annotation pipelines.

Reading between the lines

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

If the keep–erase contrast really measures fidelity, the same construction should transfer beyond closed-set segmentation to instance- or panoptic-level claims and to open-vocabulary classes that no annotation scheme ever named, since the judge's vocabulary need not match the dataset's.
The class-structured win rates imply that annotation-production conventions, such as polygonization, minimum mapping units, and registration, drive more of the observed distortion than class difficulty or object size; a testable prediction is that re-annotating the same imagery under stricter conventions would shrink the prediction-win gap for man-made classes.
The authors' caveat that all four judges tested share the CLIP-style pretraining paradigm suggests a stress test the paper does not run: if a judge trained under a different alignment objective produced a different class structure, part of the 81% expert agreement would be representation bias rather than image fidelity.
Because trained label-quality baselines inherit the very label distribution they audit, CMF could be used as a pre-filter to clean training sets before robust training, decoupling the audit from the model under test.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

You should read this one. The core idea is new and useful: instead of trusting one mask as reference, score two competing masks by asking a frozen VLM whether class evidence sits inside or outside each mask. The keep/erase construction is clever, and the authors back it with a serious evidence chain: controlled corruptions, a 10,731-pair audit over ten datasets, a blinded expert study, and matched transfer experiments that separate arbitration from pseudo-label distillation. The matched controls in §4.5 are particularly good—random replacement does nothing, confidence replacement does less than CMF, so the effect is attributable to the arbitration signal, not just to injecting strong pseudo-labels.

The stress-test flag is real, though I'd put a softer spin on it. In §3.3 the paper claims the two terms of Eq. (5) respond to mask enlargement in opposite directions. That is wrong: if the mask is inside the true region, enlarging it raises the keep-view probability and lowers the erase-view probability, so both changes push F up. The empirical controls agree—dilation detection is 0–61% across datasets, with 0% on Massachusetts Buildings—so CMF is partially blind to over-inclusion, one of the annotation defects that motivated the audit. The paper acknowledges this in passing but doesn't link it to the class-structured win rates. That's a genuine weakness, and it needs a direct fix, not just a sentence in the limitations.

But I don't think the weakness is load-bearing. The expert validation is a real check: CMF matches the expert majority on 81% of pairs overall and 89% on man-made classes. If the verdicts were just a mask-area artifact, independent human judges would not agree with them that often. The expert study is small (279 decidable pairs), and the shared CLIP paradigm between judge and candidate generator is a legitimate worry, but the judge-swap stability across four CLIP-family models and the expert agreement provide enough support to take the empirical claim seriously. The authors are also honest about these limits in §4.7.

Main fixes I'd want before fully trusting it: correct the erroneous sentence in §3.3, report win rates split by whether the preferred mask is larger or smaller to quantify area bias, and release code and data. The supplementary material is described but not available to me.

This paper deserves a serious referee. The idea is novel, the execution is mostly careful, and the central claim is testable. I'd accept it conditionally, with the area-sensitivity analysis as a required revision.

Referee Report

3 major / 4 minor

Summary. The paper introduces Contrastive Mask Fidelity (CMF), a training-free and reference-free metric that compares two competing class masks by scoring keep and erase counterfactual views of each mask with a frozen vision-language judge (RemoteCLIP). The fidelity score F(M_c) = P(c|x_k) - P(c|x_e) in Eq. (5) is used to decide, per class, whether the human annotation or a candidate prediction is more faithful to the image. The authors validate CMF on controlled corruptions, audit 10,731 image-class pairs across ten remote-sensing datasets, report class-structured win rates (man-made classes favor the candidate, amorphous land cover favors the human label), compare CMF against a blinded three-annotator consensus (81.0% agreement), and show that training a segmenter on CMF-arbitrated labels improves cross-domain transfer over raw annotations and matched controls.

Significance. The paper is well executed and unusually transparent: it validates on controlled corruptions, includes a blinded expert study, swaps judges, sweeps the arbitration margin, and uses matched controls to isolate the arbitration effect from pseudo-label distillation. If the metric's limitations can be resolved, CMF would be a genuinely useful screening tool for segmentation datasets. The main threat is that the metric appears to be primarily a completeness/omission detector: it detects erosion and shift much better than dilation (Table 2), and its mathematical formulation rewards enlarging an under-inclusive mask through both terms of Eq. (5). This threatens the interpretation of the audit's class structure. The shared CLIP-style representation between judge and candidate is a second, acknowledged confound. Both are addressable with additional targeted analyses rather than being fundamental impossibilities.

major comments (3)

[§3.3, Eq. (5), Table 2] The assertion in §3.3 that the two terms of Eq. (5) 'respond to mask enlargement in opposite directions' is not correct for the regime relevant to the audit. If a mask is under-inclusive with respect to the true class extent T (the direction of the documented defects: incomplete tracing, polygonized simplification), enlarging it to include more of T adds true-class pixels to the keep view, increasing P(c|x_k), and removes those same pixels from the erase view, decreasing P(c|x_e). Both effects increase F(M). Conversely, when a mask is over-inclusive, the metric has limited ability to penalize the enlargement: Table 2 reports dilation detection rates of only 0-61% (0% on Massachusetts Buildings) versus 52-86% for erosion. CMF therefore does not implement the symmetric purity/completeness test it claims; it is primarily a completeness and displacement detector. Because Seg-Probe's masks for man-made classes are described as expansive, the class-structured win rates in Table 2 (e.g., buildings 75-78%, roads 76%, cars 85%) may reflect mask-enlargement bias rather than annotation distortion. Please add a formal analysis of F for under-inclusion versus over-inclusion, per-class dilation detection rates, and an area-matched or area-stratified version of the audit.
[§4.2, Table 2] The area-sensitivity evidence is insufficient to rule out a directional area bias. The reported correlation ρ=0.08 is a single Pearson coefficient pooled over the full score range and does not test the specific failure mode suggested by the dilation results, namely a monotone preference for larger masks at fixed semantic content. I request a stratified analysis: report the prediction win rate by deciles of the candidate/GT area ratio, and repeat the audit on the subset of pairs where the candidate and GT masks are area-matched (e.g., by clipping the larger mask to the smaller mask's area). If the man-made class wins persist under area matching, the class-structure claim is secure; otherwise it must be reinterpreted as a completeness preference rather than a fidelity judgment.
[§4.6/§4.7, Table 3(d)] The confound between the judge and the candidate generator is acknowledged but not resolved. All four judges tested in Table 3(d) share the CLIP pretraining paradigm, and Seg-Probe is built on SegEarth-OV3, which itself uses CLIP-style text-visual alignment; §4.7 states that 'all four judges we test share the CLIP pretraining paradigm.' The expert study is an independent anchor, but it covers only 279 decidable pairs and its aggregate 81% agreement does not rule out a shared CLIP bias on the man-made classes where the audit reports the largest effects. Please test at least one non-CLIP judge (e.g., a supervised segmentation model or a pixel-level classifier trained on a disjoint dataset) or generate candidate masks with a non-CLIP method and verify that the class-structured win rates persist. If this is not feasible, the audit's conclusion should be reframed as 'distortion as perceived by CLIP-family judges' rather than as a property of the annotations.

minor comments (4)

[§4.6] There is a typo in §4.6: 'does nothingeonthe specific judge' should read 'does not hinge on the specific judge.'
[§4.4] The expert study excludes 21 split-vote pairs and then reports agreement on 279 pairs; please report the agreement when split votes are counted as disagreements or resolved by a pre-registered rule, and give the per-dataset and per-class breakdown of the 300 pairs, since the class imbalance of the expert sample may affect the aggregate 81%.
[§4.2, Table 2] The validity controls in Table 2 report only detection rates; please also report false-positive rates on clean pairs and bootstrap confidence intervals, especially for Massachusetts Buildings where n=130 and the dilation detection rate is 0%.
[Table 2] The 'All' row pools pairs across datasets with different class distributions and class vocabularies; report dataset-macro averages for the validity controls as well as the pooled values.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; CMF is defined independently and validated against external evidence.

full rationale

The paper's central claim is that CMF, defined by Eqs. (2)–(5) as a difference of softmax probabilities from a frozen external judge, measures semantic mask fidelity. This is a definition, not a derivation from fitted inputs. The verdict rule (Eq. 6) follows directly from the definition, and the audit results are empirical statistics over 10,731 pairs, not predictions that reduce to the metric's construction. The metric itself is not fitted to any data; the margin δ is set to 0, the temperature τ is fixed, and the judge is frozen. The validation against expert consensus (§4.4) and controlled corruptions (§4.2) is external to the metric's definition. The candidate generator Seg-Probe is a separate module with its own ablations and is explicitly not the basis of the metric. No load-bearing self-citation appears; all cited external models (RemoteCLIP, SAM3, SegEarth-OV3) are independent prior work. The paper's own acknowledged limitations (all judges share CLIP pretraining, label-conditioned audit) are threats to validity, not circularity. While the claim in §3.3 that the keep and erase terms respond to enlargement in opposite directions is algebraically incorrect, this is a correctness flaw in the description, not a circular reduction of the metric's output to its input. The derivation chain is self-contained: CMF is proposed, its properties are tested, and its utility is validated on independent benchmarks and human judgments.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the judge's softmax belief being a valid indicator of class presence, on the compositing scheme not introducing artifacts, on the expert majority being a reliable ground truth, and on the candidate generator and judge being sufficiently independent. The last assumption is only partially satisfied because both use CLIP-style encoders.

free parameters (5)

Logit scale tau = 100 (default), swept 50-200
Temperature in the softmax in Eq. (2); chosen by hand, but verdicts vary by at most 0.6 points across the sweep.
Gaussian blur sigma = 24
Standard deviation of the blur in Eqs. (3)-(4); selected for in-distribution compositing, and the ablation shows Gaussian beats black and mean-color fills.
Arbitration margin delta = 0
Threshold in Eq. (7); the default defers ties to the human label, and the sweep shows transfer performance is stable across delta.
Seg-Probe dataset-level activation settings = not specified in main text
Deferred to the supplementary material; may be tuned per dataset, which would affect candidate masks and hence the audit results.
Judge input resolution = 224x224
Fixed resize for RemoteCLIP inputs; a standard but arbitrary choice.

assumptions (5)

domain assumption Softmax over class text embeddings yields a valid relative measure of class presence in an image view.
Eq. (2) defines P(c|r) as a temperature-scaled softmax over cosine similarities; the entire metric depends on this being a faithful judge of class evidence.
domain assumption Gaussian blur fill preserves local color statistics and avoids out-of-distribution artifacts.
Eqs. (3)-(4) use B(x) as the fill for keep and erase views; the ablation supports this choice, but it is still a modeling assumption.
domain assumption The three-annotator majority vote on 279 pairs is an accurate expert consensus for mask fidelity.
Section 4.4 uses this consensus as ground truth to validate CMF; the study is small and covers only six datasets.
ad hoc to paper The candidate generator and the judge are sufficiently independent to avoid systematically favoring the candidate.
Section 4.7 admits all judges share the CLIP pretraining paradigm, and Seg-Probe also builds on CLIP-based segmentation, so this independence is incomplete and is a recognized limitation.
domain assumption The original human annotation is clean in the validity controls.
Section 4.2 ranks a corrupted version against the original GT mask; the authors note this assumes the original annotation is clean.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Contrastive Mask Fidelity: Reference-Free Auditing of Ground-Truth Masks in Remote Sensing Semantic Segmentation." pith.science (2026). https://pith.science/paper/KJZVTC7J

@misc{pith2026260809101,
  author       = {Pith},
  title        = {Pith review of: Contrastive Mask Fidelity: Reference-Free Auditing of Ground-Truth Masks in Remote Sensing Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KJZVTC7J}},
  note         = {Machine review of arXiv:2608.09101}
}

read the original abstract

Semantic segmentation models are trained and evaluated against human-drawn masks, yet remote-sensing annotations are often coarse, incomplete, or misaligned; high overlap scores may then reflect agreement with imperfect labels rather than faithfulness to the image, creating an evaluation paradox. We introduce Contrastive Mask Fidelity (CMF), a training-free, reference-free metric that scores competing class masks directly against image evidence. CMF composites keep and erase counterfactual views of each mask and asks a frozen vision-language judge whether class evidence is concentrated inside the mask and absent outside. We validate CMF on controlled mask corruptions, then audit 10,731 image-class pairs across ten remote-sensing benchmarks using candidate masks from Seg-Probe, a training-free open-vocabulary probe built on SegEarth-OV3 that outperforms prior baselines on nine of ten datasets. The audit reveals systematic, class-dependent annotation distortion: man-made classes such as buildings, roads, and cars favor the candidate mask on 62-85% of pairs, whereas ambiguous land cover more often favors human annotations. On a blinded three-annotator consensus, CMF matches expert judgment on 81% of pairs, exceeding keep-only scoring, model confidence, and a trained label-quality baseline. Finally, conservative class-wise arbitration yields supervision that improves cross-domain transfer over raw annotations and matched replacement controls, positioning CMF as a scalable tool for auditing ground truth rather than presuming it infallible.

Figures

Figures reproduced from arXiv: 2608.09101 by the authors.

**Figure 1.** Is the ground truth always the best label? On binary (top) and multi-class (bottom) benchmarks, the human ground truth misses or distorts the true extent while a training-free prediction follows the image. How do we evaluate—and exploit—mask fidelity without trusting the annotation? and Mueller 2021), and label-quality scores for segmentation confirm that pixel-accurate annotation is error-prone at scale (Lad an… view at source ↗

**Figure 2.** The GeoVeritas pipeline. Stage I: Seg-Probe generates candidate masks with a frozen SAM 3 encoder, pooling synonym prompts by a soft-OR (SEP) and fusing multi-scale and flipped views (MSE, TTA). Stage II: for each class, CMF builds keep/erase views of both the prediction and the ground truth and lets a frozen RemoteCLIP judge decide which mask is more faithful (∆F). Stage III: labels arbitrated by the verdicts train… view at source ↗

**Figure 3.** Class-evidence heatmaps of Seg-Probe and its SegEarth-OV3 backbone on the ten audited datasets. Every panel follows the same column protocol—Image / SegEarth-OV3 / Seg-Probe—and the cyan # Class tag names the queried class. Relative to the backbone, Seg-Probe concentrates evidence on the probed class and suppresses background bleed, so its disagreements with the human masks are worth adjudicating rather than dismiss… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Accuracy–efficiency trade-off of training-free open-vocabulary methods. Average mIoU over the ten audited datasets (top) and fine-GSD urban mIoU, the mean of Potsdam and Vaihingen (bottom), against per-image runtime (log scale); bubble area encodes peak inference VRA…

**Figure 5.** Figure 5: The CMF judging process. Left: single-class audits on five datasets—the judge scores the keep/erase views of the ground truth and of the prediction, and the fidelity gap ∆Fc decides the verdict. Right: multi-class audit of one LoveDA image—each annotated class is audit…

**Figure 6.** Figure 6: Cross-domain qualitative comparison on three target domains. On unseen target images, the model trained on CMF-arbitrated labels recovers buildings and roads that the GT-trained model misses (boxes), consistent with the audit’s class structure. benefit (+0.3, within on…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages

[1]

International Conference on Learning Representations (ICLR) , year =

Carion, Nicolas and Gustafson, Laura and Hu, Yuan-Ting and Debnath, Shoubhik and Hu, Ronghang and Suris Coll-Vinent, Didac and Ryali, Chaitanya and Alwala, Kalyan Vasudev and Khedr, Haitham and Huang, Andrew and Lei, Jie and Ma, Tengyu and Guo, Baishan and Kalla, Arpit and Marks, Markus and Greer, Joseph and Wang, Meng and Sun, Peize and R. International ...

arXiv
[2]

Zhang, Dengke and Liu, Fagui and Tang, Quan , booktitle =

work page
[3]

Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

Segment Anything , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

work page
[4]

International Conference on Learning Representations (ICLR) , year =

Ravi, Nikhila and Gabeur, Valentin and Hu, Yuan-Ting and Hu, Ronghang and Ryali, Chaitanya and Ma, Tengyu and Khedr, Haitham and R. International Conference on Learning Representations (ICLR) , year =

work page
[5]

Proceedings of the 38th International Conference on Machine Learning , series =

Learning Transferable Visual Models From Natural Language Supervision , author =. Proceedings of the 38th International Conference on Machine Learning , series =
[6]

2024 , note =

Zhang, Zilun and Zhao, Tiancheng and Guo, Yulong and Yin, Jianwei , journal =. 2024 , note =

work page 2024
[7]

2024 , note =

Liu, Fan and Chen, Delong and Guan, Zhangqingyun and Zhou, Xiaocong and Zhu, Jiale and Ye, Qiaolin and Fu, Liyong and Zhou, Jun , journal =. 2024 , note =

work page 2024
[8]

2021 , doi =

Hessel, Jack and Holtzman, Ari and Forbes, Maxwell and Le Bras, Ronan and Choi, Yejin , booktitle =. 2021 , doi =

work page 2021

Show all 34 references

[9]

Li, Kaiyu and Liu, Ruixun and Cao, Xiangyong and Bai, Xueru and Zhou, Feng and Meng, Deyu and Wang, Zhi , booktitle =
[10]

2025 , eprint =

Li, Kaiyu and Zhang, Shengqi and Wang, Yujie and Deng, Yupeng and Wang, Zhi and Meng, Deyu and Cao, Xiangyong , journal =. 2025 , eprint =

2025
[11]

Niu, Xin and Zhao, Manqi and Jiang, Dongsheng and Wu, Yingying and Su, Bing , booktitle =
[12]

Chen, Wenyang and Hu, Zhanxuan and Zhang, Yaping and Ning, Hailong and Tai, Yonghang , booktitle =
[13]

IEEE Transactions on Geoscience and Remote Sensing , volume =

Open-Vocabulary High-Resolution Remote Sensing Image Semantic Segmentation , author =. IEEE Transactions on Geoscience and Remote Sensing , volume =. 2025 , note =

2025
[14]

Proceedings of the AAAI Conference on Artificial Intelligence , volume =

Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2025 , doi =

2025
[15]

arXiv preprint arXiv:2307.05080 , year =

Estimating Label Quality and Errors in Semantic Segmentation Data via Any Model , author =. arXiv preprint arXiv:2307.05080 , year =. 2307.05080 , archivePrefix=

arXiv
[16]

and Wang, Yi and Li, Qingyu and Zhu, Xiao Xiang , journal =

Liu, Chenying and Albrecht, Conrad M. and Wang, Yi and Li, Qingyu and Zhu, Xiao Xiang , journal =. 2024 , doi =

2024
[17]

2025 , doi =

Yang, Bingnan and Zhang, Mi and Zhang, Zhili and Zhang, Zhan and Zhao, Yuanxin and Hu, Xiangyun and Gong, Jianya , journal =. 2025 , doi =

2025
[18]

Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

Semi-Supervised Semantic Segmentation under Label Noise via Diverse Learning Groups , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =
[19]

2024 , doi =

Lee, Seongyun and Kim, Seungone and Park, Sue and Kim, Geewook and Seo, Minjoon , booktitle =. 2024 , doi =

2024
[20]

Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , volume =

Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks , author =. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , volume =
[21]

Wang, Junjue and Zheng, Zhuo and Ma, Ailong and Lu, Xiaoyan and Zhong, Yanfei , booktitle =
[22]

Xia, Junshi and Yokoya, Naoto and Adriano, Bruno and Broni-Bediako, Clifford , booktitle =
[23]

Rottensteiner, Franz and Sohn, Gunho and Jung, Jaewook and Gerke, Markus and Baillard, Caroline and Benitez, Sebastien and Breitkopf, Uwe , journal =. The. 2012 , doi =

2012
[24]

Boguszewski, Adrian and Batorski, Dominik and Ziemba-Jankowska, Natalia and Dziedzic, Tomasz and Zambrzycka, Anna , booktitle =
[25]

2018 , doi =

Demir, Ilke and Koperski, Krzysztof and Lindenbaum, David and Pang, Guan and Huang, Jing and Basu, Saikat and Hughes, Forest and Tuia, Devis and Raskar, Ramesh , booktitle =. 2018 , doi =

2018
[26]

2021 , doi =

Rahnemoonfar, Maryam and Chowdhury, Tashnim and Sarkar, Argho and Varshney, Debvrat and Yari, Masoud and Murphy, Robin Roberson , journal =. 2021 , doi =

2021
[27]

Machine Learning for Aerial Image Labeling , author =
[28]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

Single-Stage Semantic Segmentation From Image Labels , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =
[29]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

Reproducible Scaling Laws for Contrastive Language-Image Learning , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =
[30]

Proceedings of the European Conference on Computer Vision (ECCV) , pages =

Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation , author =. Proceedings of the European Conference on Computer Vision (ECCV) , pages =
[31]

Extract Free Dense Labels from

Zhou, Chong and Loy, Chen Change and Dai, Bo , booktitle =. Extract Free Dense Labels from. 2022 , doi =

2022
[32]

, booktitle =

Wang, Feng and Mei, Jieru and Yuille, Alan L. , booktitle =. 2024 , doi =

2024
[33]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

Grounding Everything: Emerging Localization Properties in Vision-Language Transformers , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =
[34]

2024 , doi =

Lan, Mengcheng and Chen, Chaofeng and Ke, Yiping and Wang, Xinjiang and Feng, Litong and Zhang, Wayne , booktitle =. 2024 , doi =

2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

International Conference on Learning Representations (ICLR) , year =

Carion, Nicolas and Gustafson, Laura and Hu, Yuan-Ting and Debnath, Shoubhik and Hu, Ronghang and Suris Coll-Vinent, Didac and Ryali, Chaitanya and Alwala, Kalyan Vasudev and Khedr, Haitham and Huang, Andrew and Lei, Jie and Ma, Tengyu and Guo, Baishan and Kalla, Arpit and Marks, Markus and Greer, Joseph and Wang, Meng and Sun, Peize and R. International ...

arXiv

[2] [2]

Zhang, Dengke and Liu, Fagui and Tang, Quan , booktitle =

work page

[3] [3]

Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

Segment Anything , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

work page

[4] [4]

International Conference on Learning Representations (ICLR) , year =

Ravi, Nikhila and Gabeur, Valentin and Hu, Yuan-Ting and Hu, Ronghang and Ryali, Chaitanya and Ma, Tengyu and Khedr, Haitham and R. International Conference on Learning Representations (ICLR) , year =

work page

[5] [5]

Proceedings of the 38th International Conference on Machine Learning , series =

Learning Transferable Visual Models From Natural Language Supervision , author =. Proceedings of the 38th International Conference on Machine Learning , series =

[6] [6]

2024 , note =

Zhang, Zilun and Zhao, Tiancheng and Guo, Yulong and Yin, Jianwei , journal =. 2024 , note =

work page 2024

[7] [7]

2024 , note =

Liu, Fan and Chen, Delong and Guan, Zhangqingyun and Zhou, Xiaocong and Zhu, Jiale and Ye, Qiaolin and Fu, Liyong and Zhou, Jun , journal =. 2024 , note =

work page 2024

[8] [8]

2021 , doi =

Hessel, Jack and Holtzman, Ari and Forbes, Maxwell and Le Bras, Ronan and Choi, Yejin , booktitle =. 2021 , doi =

work page 2021