Mitigating Spurious Correlations in Weakly Supervised Semantic Segmentation via Cross-architecture Consistency Regularization

Yen-Chia Hsu; Zheyuan Zhang

REVIEW 2 major objections 6 minor 1 cited by

Mitigating Spurious Correlations in Weakly Supervised Semantic Segmentation via Cross-architecture Consistency Regularization

T0 review · 2 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Forcing a ResNet teacher and a Vision Transformer student to align their internal feature representations suppresses co-occurrence bias in weakly supervised smoke segmentation, with no external supervision or prior knowledge.

desk verdict Plausible mechanism and systematic ablations, but the test protocol never measures the claimed co-occurrence suppression, and test-set hyperparameter selection inflates the headline numbers. read the letter →

arxiv 2507.21959 v1 pith:XWBV44MR submitted 2025-07-29 cs.CV

Zheyuan Zhang , Yen-chia Hsu This is my paper · ORCID

classification cs.CV

keywords weaklysupervisedsemanticsegmentationspuriouscorrelationsco-occurrencebiascross-architectureconsistencyteacher-studentknowledgetransferclassactivationmapsvisiontransformerindustrialsmoke

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that spurious correlations — a weakly supervised model's habit of treating a chimney as part of the smoke that always appears with it — can be suppressed with no extra labels, no external priors, and no human knowledge. In weakly supervised semantic segmentation (WSSS), only image-level labels are available, so pixel masks for training a segmentation model are derived from class activation maps (CAMs), the heatmaps of which regions drive the classifier's decision; these maps are prone to firing on co-occurring context like chimneys. The authors' remedy is a teacher-student setup in which a pretrained ResNet teacher guides a Vision Transformer student via a consistency loss that aligns their internal feature representations, exploiting the observation that the two architectures make different mistakes: the CNN localizes smoke but blurs its boundaries, while the ViT draws sharp boundaries but also activates co-occurring context. On a custom industrial-smoke dataset, the method raises the seed-mask mean intersection-over-union (mIoU), a standard overlap score, from 33.56 for the strongest comparison baseline (PCM, a pixel-correlation method) to 47.37, and to 52.93 after adding post-processing that combines the Segment Anything Model (SAM) with a conditional random field (CRF). The broader claim is that the agreement enforced between heterogeneous architectures acts as an internal supervisor, so a pair of biased classifiers can correct each other's bias without any outside knowledge.

What carries the argument

The load-bearing mechanism is a knowledge transfer consistency loss inside a teacher-student framework. A frozen, pretrained ResNet-50 acts as teacher and a trainable ViT-B as student; each network's feature map first passes through a learnable linear projection (a 1×1 convolution) into a shared two-channel space suited to the binary smoke/background task, and after global average pooling the two resulting vectors are aligned by cosine similarity, $S_{global} = \frac{f^{(v)} \cdot f^{(r)}}{\| f^{(v)} \| \| f^{(r)} \|}$. The training objective is $\mathcal{L} = \mathcal{L}_{cls} + \lambda \mathcal{L}_{global}$ with $\mathcal{L}_{global} = 1 - S_{global}$, where $\lambda$ is a tuned weight. This loss forces the ViT to behave internally more like the CNN, transferring the CNN's ability to ignore co-occurring context while keeping the ViT's sharper boundaries; ablations show the teacher-student configuration with global alignment and cosine similarity outperforms co-training, spatial or channel-wise alignment, and $\ell^1$ or $\ell^2$ distances.

What would settle it

Score the trained student model on chimney-only patches — chimney present, no visible smoke — and measure the smoke-class CAM activation: if a meaningful fraction of such patches light up, the spurious correlation is not suppressed and the reported mIoU, computed only on patches with visible smoke, would be hiding it. A companion check is to re-tune $\lambda$, the CRF scaling, the SAM points-per-side, and the SAM fusion strategy on a held-out split of the same dataset; if the 47.37 and 52.93 figures shrink substantially, a portion of the gain reflects selection rather than the consistency mechanism.

Watch

Extended reading notes

Core claim

The paper's central claim is that feature-level consistency between two architectures, enforced during training, suppresses co-occurrence bias in weakly supervised segmentation without any external supervision. Concretely, a fixed pretrained ResNet-50 teacher and a trainable ViT-B student are trained together with a binary cross-entropy classification loss plus a knowledge transfer loss that maximizes the cosine similarity of their global average-pooled feature vectors, after each network's features pass through a learnable linear projection into a shared space. The resulting CAMs activate foreground smoke rather than the co-occurring chimney, raising seed pseudo-mask mIoU to 47.37 on the custom smoke dataset, compared with 33.56 for PCM, 26.10 for a plain ResNet50, and 13.18 for a ViT-S trained alone; after CRF and SAM-enhanced post-processing, the pseudo masks reach 52.93. The authors state that they are the first to report that cross-architecture consistency alone can suppress co-occurrence bias during representation learning, and they read the result as evidence that heterogeneous inductive biases supply complementary supervision that needs no external source.

Load-bearing premise

The load-bearing premise is the evaluation protocol: because only patches that contain visible smoke are scored, chimney-without-smoke patches — where residual spurious activation would show up — are never measured, so the reported mIoU gains do not by themselves prove that co-occurrence bias was suppressed.

Editorial extensions

If this is right

Seed pseudo-mask quality for weakly supervised smoke segmentation rises to 47.37 mIoU with only image-level labels, up from 33.56 for PCM, 26.10 for a plain ResNet50, and 13.18 for a ViT-S trained alone.
Appending CRF and SAM-enhanced post-processing raises the pseudo masks to 52.93 mIoU, within reach of fully supervised references (54.68 for a fine-tuned SAM, 68.27 for a transformer segmentation model).
Because the co-occurrence bias is removed during representation learning rather than at refinement time, the approach targets what the paper calls the root cause of the problem, and strong seeds no longer need heavy post-processing to be usable.
Since the framework requires only image-level labels and a pair of heterogeneous architectures, it inherits none of the scalability costs of external priors such as saliency maps, causal inference, or language supervision.

Reading between the lines

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

The recipe is architecture-agnostic: any two networks with different inductive biases and different failure modes could be aligned the same way, so the method is directly testable in other co-occurrence-heavy domains, such as medical imaging where a lesion always appears with an organ, or satellite imagery where a target always appears with fixed infrastructure.
The cleanest test the paper does not run is to score chimney-only patches — chimney present, no visible smoke — where a model that truly suppresses co-occurrence bias should produce near-zero smoke activation; that false-positive rate would settle the claim more directly than mIoU on the smoke-only test patches.
Two caveats bound the reported numbers: the loss weight, CRF scaling, SAM points per side, and SAM fusion strategy were all selected by maximizing mIoU on the same test set that produced the gains, and the authors' own ViT-B baseline trained with image-level labels on a portion of the test set reaches 47.99 mIoU, within 0.62 of the proposed method's 47.37.
The frozen, pretrained teacher sets a ceiling on the student: because the ViT is pulled toward the ResNet's representation, the method inherits the teacher's blind spots, so a stronger teacher or one that keeps updating is the most direct route to further seed improvement.

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.

Desk Editor's Note

read the letter

Honest take: the paper has a plausible mechanism and genuine systematic ablations, but the headline claim about suppressing co-occurrence is not actually measured by its test protocol. The stress-test is right: Section 4.1 says only patches with visible smoke are evaluated, so the cleanest probe for chimney false positives—chimney without smoke—is excluded. The mIoU gains could come from better foreground coverage or boundary sharpness, not from reducing co-occurrence activation. Even if every number is honest, the experiments do not isolate the claimed mechanism.

What is genuinely new: teacher-student ResNet-to-ViT feature consistency with a 1x1-conv projection and global cosine loss, targeted at co-occurrence in WSSS. The ablation set is thorough—paradigm, metric, alignment level, projection, loss weight—and the consistency loss improves seed mIoU across a wide lambda range (38.55 to 47.37 vs 33.56 baseline). That is real evidence the regularizer does something. The teacher-student vs co-training comparison is informative: the frozen pretrained teacher wins, which is a useful data point.

Soft spots, in order of severity. First, the evaluation-gap issue above. Second, no validation split: lambda, CRF scaling, SAM points/fusion are all chosen on the same 900-image test set that produces the headline numbers. Third, no error bars or multiple seeds. Fourth, missing comparisons to the closest cross-architecture baselines the paper itself cites ([40], [41], [34]). The "no external supervision" claim is overstated given ImageNet pretraining, frozen SAM, and out-of-distribution training images. These are constraints, not refutations: the method is demonstrated on a single custom dataset, not confirmed.

The paper reads like a master's thesis, which it likely is: honest about limitations (Section 6.2), unpolished in places, but clear about what was tried. For a reader in WSSS or applied smoke monitoring, it is a useful case study in how test-set selection can inflate gains. I would send it to a serious referee, asking for a validation split, a chimney-only evaluation, and the missing baselines. Releasing the dataset and code would make it much easier to assess.

Referee Report

2 major / 6 minor

Summary. Zhang and Hsu propose a weakly supervised semantic segmentation (WSSS) method for industrial smoke that uses only image-level labels. The core idea is a teacher-student framework in which a pretrained ResNet teacher guides a ViT student through a feature-level consistency loss (global cosine similarity after a linear projection), motivated by the observation that CNNs and ViTs produce complementary CAMs. They also evaluate a co-training variant and a set of post-processing steps (CRF, SAM-enhanced, CAM fusion, multi-scale testing). On a custom IJmond smoke dataset, the method reports seed pseudo-mask mIoU of 47.37 versus 33.56 for PCM, and 52.93 after post-processing, and claims to be the first to suppress co-occurrence bias via cross-architecture consistency without external supervision.

Significance. The central idea is worth taking seriously: using two heterogeneous architectures as mutual regularizers is a plausible way to reduce spurious correlations in WSSS, and the ablations are more systematic than is typical for this area. The paper provides code, compares teacher-student versus co-training, three alignment levels, several similarity metrics, feature projection, and a lambda sweep; the consistency loss improves mIoU over a wide lambda range (38.55 at 0.3 to 47.37 at 1.0 versus the 33.56 baseline), which is genuine evidence that the regularizer helps in the current evaluation. However, the evaluation protocol cannot currently support the headline claim of co-occurrence suppression: the test set is restricted to patches containing visible smoke, and hyperparameters are selected on the same test set that produces the final numbers. These issues are fixable, but until fixed the claimed significance is not established.

major comments (2)

[Section 4.1; Tables 4.1 and 4.3] The evaluation protocol cannot support the paper's central claim that cross-architecture consistency suppresses co-occurrence bias. Section 4.1 states that 'Only patches contains visible smoke will be used for evaluation.' Because the co-occurrence problem is precisely that chimney regions activate even when no smoke is present, excluding chimney-without-smoke patches removes the direct probe of the phenomenon. The reported mIoU gains (47.37 versus 33.56 for PCM) could be produced by improved foreground coverage or boundary sharpness without any reduction in spurious chimney activation. Please evaluate on all test patches, including smoke-free chimney patches, and report a false-positive metric for chimney regions, or explicitly report per-pixel precision on non-smoke areas in addition to smoke-class IoU.
[Section 4.4, Tables 4.10 and 4.11; Section 4.1] No validation split is described, and all tunable hyperparameters are selected by maximizing mIoU on the same 900-image test set used to report the final numbers. Specifically, lambda in Table 4.10 is set to 1.0 because it gives the maximum mIoU there; SAM points per side (Table 4.11a), SAM fusion strategy (Table 4.11b), and CRF scaling (Table 4.11c) are likewise chosen by test-set performance. The 'Optimal threshold' rows in Table 4.4 (53.92 and 57.15) are ground-truth oracle numbers and should not be presented alongside achievable methods. Please introduce a held-out validation split or nested cross-validation, select hyperparameters on it, and report final test numbers only once; remove or clearly label oracle rows.

minor comments (6)

[Section 4.1] Please fix the sentence 'Only patches contains visible smoke will be used for evaluation' and specify how many patches contain smoke, how many do not, and whether non-smoke patches are entirely absent from the test evaluation.
[Table 4.1] The row 'image-level + limited pixel-level T1 ViT-B✗' has no reported mIoU and an unexplained symbol; either report the value or remove the row.
[Table 4.8] The 'Seed' column is constant 33.56 in all rows; clarify that this is the PCM baseline seed and show the seed value for each variant, since seed quality depends on the training paradigm and projection choice.
[Section 4.2] The claim of 'no external supervision' should be qualified: the backbones are initialized with ImageNet classifier weights, which is external pretraining even if no additional task-specific labels or priors are used.
[Section 2.2] The text 'citeshu2021channel proposed' contains an unresolved citation key; please replace it with the proper reference.
[Section 6.2] The limitations paragraph lists underexplored research directions but does not mention the evaluation protocol limitations, specifically the smoke-only patch evaluation and the absence of a validation split; these should be acknowledged.

Circularity Check

2 steps flagged · score 5.0 of 10

Headline mIoU numbers are the argmax of sweeps run on the same test set, and the co-occurrence-suppression claim is evaluated only on smoke-containing patches, so the claimed mechanism's direct probe (chimney without smoke) is excluded by definition.

fitted input called prediction [Section 4.4 (Ablation Studies), Tables 4.10 and 4.11a-c; evaluation protocol in Section 4.1]
"Table 4.10: The Effect of the Consistency Loss Coefficient λ. We choose the max mIOU from baseline model (ResNet50+PCM) for comparison. ... Table 4.11: Impact of hyper-parameters and different strategies. Gray rows indicate best settings. ... To evaluate the quality of pseudo masks, we generate them for every image in the test set and then use the ground truth to compute mIoU."

The hyperparameters that produce the headline numbers are selected by maximizing mIoU on the same 900-image test set whose mIoU is then reported as the method's performance; no validation split is described in Sections 4.1-4.2. The reported 47.37 is the argmax of the λ sweep (λ=1.0 in Table 4.10, where the curve peaks and falls on both sides: 0.3->38.55, 0.8->45.04, 1.3->46.62), and the reported 52.93 uses the argmax CRF scaling (16, Table 4.11c: 52.52) and argmax SAM points-per-side (32, Table 4.11a: 51.03). Because each configuration is chosen as the maximum of the very curve being reported, the headline values are statistically forced upward: the reported number equals the test-set optimum of the evaluated grid by construction, not an unbiased prediction.
other [Section 2.1.2 (claim) vs. Section 4.1 (Dataset and evaluation metrics)]
"To the best of our knowledge, we are the first to report that leveraging cross-architecture consistency can effectively suppress co-occurrence bias during the representation without the need of external supervision. ... Only patches contains visible smoke will be used for evaluation."

The paper defines its target problem as chimney-smoke co-occurrence ('industrial emission are always spatially coupled with chimneys'), so the direct behavioral probe of suppression is activation on chimney regions without smoke. Section 4.1 defines the evaluation set to exclude exactly that case ('Only patches contains visible smoke will be used for evaluation'). The reported mIoU is a single score that conflates foreground coverage, boundary quality, and false positives on smoke-containing patches; it cannot detect whether chimney false positives persist, since chimneys outside smoke regions are never scored.

full rationale

The core derivation is not circular: the training objective (Eqs. 3.2-3.4: global cosine-similarity consistency loss L_global = 1 - S_global appended to binary cross-entropy with a learnable feature-projection layer) is an original construction, and its contribution is tested in ablations across paradigms, alignment levels, and metrics (Table 4.7), with the directional improvement over the 33.56 PCM baseline holding at every λ tested. No load-bearing self-citation chain exists: the architecture-difference premise is supported by external works ([43], [41], [51]), no uniqueness theorem is imported from the authors' prior work, the feature-projection design is attributed to external [55], and the cosine metric is selected by the paper's own ablation rather than by citation. The custom IJmondCAM dataset and its annotation tool are self-made, but datasets are evidence rather than derivation, so this alone does not raise circularity beyond a generalizability caveat. The two flagged steps are genuinely load-bearing and partial: (1) λ, CRF scaling, SAM points-per-side, and SAM fusion strategy are all selected by maximizing mIoU on the test set whose mIoU is then reported, so the headline magnitudes 47.37/52.93 are test-set argmax values by construction; (2) the test set excludes chimney-without-smoke patches, so the paper's central claim of co-occurrence suppression is never measured on its direct probe. Additional correctness risks, noted here rather than as circularity: a plain ViT-B trained with part of the test data (47.99, Table 4.1) exceeds 'Ours' (47.37), and the 'Optimal threshold' rows (53.92, 57.15) in Table 4.4 are ground-truth oracles listed alongside real methods. On balance, the central mechanism has independent empirical content, but the headline evidence for it is partly self-confirming, giving a partial-circularity score of 5.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The framework's effectiveness is measured through mIoU on the private IJmond test set, so the load-bearing inputs are the dataset's annotation quality and the evaluation protocol, plus a handful of test-set-chosen constants (lambda, CRF scaling, SAM points, fusion strategy). There are no invented entities. The largest ledger cost is the absence of a held-out validation split: the reported optimum is selected against the evaluation labels.

free parameters (6)

lambda (consistency loss weight) = 1.0
Selected as the value with maximum test-set mIoU in Table 4.10; the caption states the max mIoU from the baseline is chosen for comparison, so the headline 47.37 is test-set-selected.
CAM-to-mask threshold = 0.3
Fixed low threshold for assigning background, set by hand in Section 4.2; the paper itself notes optimal thresholds vary across images.
CRF scaling factor = 16
Best value from the test-set sweep in Table 4.11(c), where mIoU ranges from 33.47 (scale 2) to 52.52 (scale 16).
SAM points per side = 32
Best value from the test-set sweep in Table 4.11(a), giving 51.03 versus 50.50 for 16 points.
SAM fusion strategy = COPY (direct copy)
Chosen from test-set comparison in Table 4.11(b); AND (41.03) and OR (35.97) are worse than COPY (51.03).
Feature projection output channels = 2
Equation (3.1) projects both feature maps to a 2-channel space; the dimension is a design choice tied to the binary smoke/non-smoke task.

assumptions (5)

standard math Cosine similarity and 1x1-conv linear projections are sufficient alignment operators for heterogeneous CNN/ViT feature maps.
Invoked in Section 3.2.1 (Eqs. 3.1 to 3.3); the paper tests alternatives (L1, L2, KL on logits) and cosine wins empirically, but the space of alignments is not exhaustively explored.
domain assumption Industrial smoke emissions are always spatially coupled with chimneys, making co-occurrence the dominant failure mode.
Stated in the abstract and Section 1; this justifies treating mIoU gains as co-occurrence mitigation.
domain assumption ImageNet-pretrained weights plus only 3 epochs of image-level fine-tuning produce CAMs whose ResNet/ViT errors are complementary in the way the method exploits.
Core observation in Sections 1 and 3.1; complementarity is demonstrated on a few examples (Figures 1.1 and 1.2) but not quantified.
domain assumption SAM masks with CAM overlap of at least 0.3 IoU are reliable smoke foreground for fusion.
Selection rule in Section 3.2.2; the 0.3 IoU cutoff is a design choice with no sensitivity analysis.
domain assumption Evaluating only patches that contain visible smoke captures segmentation quality for this task.
Section 4.1 states 'Only patches contains visible smoke will be used for evaluation'; this excludes chimney-only patches where spurious activation would be directly observable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Mitigating Spurious Correlations in Weakly Supervised Semantic Segmentation via Cross-architecture Consistency Regularization." pith.science (2026). https://pith.science/paper/XWBV44MR

@misc{pith2026250721959,
  author       = {Pith},
  title        = {Pith review of: Mitigating Spurious Correlations in Weakly Supervised Semantic Segmentation via Cross-architecture Consistency Regularization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XWBV44MR}},
  note         = {Machine review of arXiv:2507.21959}
}

read the original abstract

Scarcity of pixel-level labels is a significant challenge in practical scenarios. In specific domains like industrial smoke, acquiring such detailed annotations is particularly difficult and often requires expert knowledge. To alleviate this, weakly supervised semantic segmentation (WSSS) has emerged as a promising approach. However, due to the supervision gap and inherent bias in models trained with only image level labels, existing WSSS methods suffer from limitations such as incomplete foreground coverage, inaccurate object boundaries, and spurious correlations, especially in our domain, where emissions are always spatially coupled with chimneys. Previous solutions typically rely on additional priors or external knowledge to mitigate these issues, but they often lack scalability and fail to address the model's inherent bias toward co-occurring context. To address this, we propose a novel WSSS framework that directly targets the co-occurrence problem without relying on external supervision. Unlike prior methods that adopt a single network, we employ a teacher-student framework that combines CNNs and ViTs. We introduce a knowledge transfer loss that enforces cross-architecture consistency by aligning internal representations. Additionally, we incorporate post-processing techniques to address partial coverage and further improve pseudo mask quality.

Figures

Figures reproduced from arXiv: 2507.21959 by the authors.

**Figure 1.1.** CAMs supervised with image-level labels often suffer from issues like co-occurrence and partial activation. Basically, there are mainly two directions to alleviate these shortcomings: one is to improve the classifier to enhance representation learning [71, 44, 53, 29, 58, 60, 9, 20]. The other one is to use post-processing techniques [6, 61, 1] to refine the cam and pseudo label. Post-processing techniques are propo… view at source ↗

**Figure 1.2.** CAMs generated from different backbone. Collectively, the main contributions can be summarized as follows: (1) Our empirical analysis reveals complementary strengths between CNNs and ViT in CAMs generation. Building on this insight, we propose a straightforward yet effective WSSS framework based on the teacher-student paradigm to reduce knowledge bias and mitigate spurious correlations. By enforcing cross-architectu… view at source ↗

**Figure 2.1.** The pipeline of multi stage and end to end method in WSSS. In our framework,we adopt multi stage pipeline. 2.1.1. WSSS In General Various WSSS methods are proposed to get rid of expensive pixel-level annotations, which can be categorized as multi-stage method [44, 60, 1] and end to end method [21, 69, 45, 2, 62, 71]. As shown in [PITH_FULL_IMAGE:figures/full_fig_p010_2_1.png] view at source ↗

Figures from the paper (7 more)

**Figure 3.1.** Figure 3.1: The overall architecture of our proposed framework. previous works relying on a single architecture model, our knowledge transfer scheme enriches the classifier’s representation by providing more diverse internal knowledge without additional external knowledge or sup…

**Figure 3.2.** Figure 3.2: Different knowledge transfer paradigms feature map provides richer information than logits. Another reason is that we want the ViT to match ResNet more in internal behavior, as better output does not mean better CAM. Cross-Architecture Consistency Regularization Modu…

**Figure 3.3.** Figure 3.3: Distribution of Optimal Iou Thresholds on Testing Set. SAM-enhanced This post-processing method is first proposed in [6]. SAM is very powerful and tends to provide masks of all the distinguishable instances in the image. In our framework, we leverage SAM’s boundary-a…

**Figure 3.4.** Figure 3.4: CAM Fusion Conditional Random Field Conditional random field (CRF) [4] is a graphical model used for refining CAM to pseudo masks by using color and position information as features [48]. CRF tries to smooth and regularize the input probability maps using image cues,…

**Figure 3.5.** Figure 3.5: Use CLIP to generate CAMs. Multi scale input Given that single-scale image inputs often fail to capture features effectively across varying object sizes and spatial contexts, leveraging multi-scale features has become a widely adopted strategy [3]. The underlying rea…

**Figure 4.1.** Figure 4.1: Comparison of CAMs generated from our method and baseline. Our method accurately highlights the foreground regions without co-occurrence issue. 4.1. Experimental setup We design a series of experiments to evaluate the effectiveness of our proposed framework in variou…

**Figure 4.2.** Figure 4.2: Comparison of CAMs generated from our method and baseline. Training: First, we compare our final result with baseline, indicting substantial improvement in the quality (mIoU) of generated seed masks as shown in [PITH_FULL_IMAGE:figures/full_fig_p025_4_2.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Right Regions, Wrong Labels: Semantic Label Flips in Segmentation under Correlation Shift
cs.CV 2026-04 unverdicted novelty 7.0 of 10

Under category–scene correlation shift, segmentation models often preserve foreground extent but swap confusable class identities; Flip, FG-Corr/Flip/Miss, and entropy flip-risk make that failure measurable and monitorable.

Reference graph

Works this paper leans on

82 extracted references · 75 canonical work pages · cited by 1 Pith paper

[40]

Precision matters: Precision-aware ensemble for weakly supervised semantic segmentation

Junsung Park and Hyunjung Shim. “Precision matters: Precision-aware ensemble for weakly supervised semantic segmentation”. In:arXiv preprint arXiv:2406.19638(2024)

work page Pith review arXiv 2024
[41]

Usage:Aunifiedseedareagenerationparadigmforweaklysupervised semantic segmentation

ZelinPengetal.“Usage:Aunifiedseedareagenerationparadigmforweaklysupervised semantic segmentation”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 624–634

work page 2023
[34]

Pdseg: Patch-wise distillation and controllable image generation for weakly-supervised histopathology tissue segmentation

Wei-Hua Li et al. “Pdseg: Patch-wise distillation and controllable image generation for weakly-supervised histopathology tissue segmentation”. In:ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2025, pp. 1–5

work page 2025
[1]

Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation

Jiwoon Ahn and Suha Kwak.Learning Pixel-level Semantic Affinity with Image-level SupervisionforWeaklySupervisedSemanticSegmentation.2018.arXiv: 1803.10464 [cs.CV]. url:https://arxiv.org/abs/1803.10464

work page Pith review arXiv 2018
[2]

Single-stage semantic segmentation from image labels

Nikita Araslanov and Stefan Roth. “Single-stage semantic segmentation from image labels”.In:ProceedingsoftheIEEE/CVFconferenceoncomputervisionandpatternrecognition. 2020, pp. 4253–4262

2020
[3]

Attentiontoscale:Scale-awaresemanticimagesegmentation

Liang-ChiehChenetal.“Attentiontoscale:Scale-awaresemanticimagesegmentation”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 3640–3649

2016
[4]

Deeplab: Semantic image segmentation with deep convolu- tionalnets,atrousconvolution,andfullyconnectedcrfs

Liang-Chieh Chen et al. “Deeplab: Semantic image segmentation with deep convolu- tionalnets,atrousconvolution,andfullyconnectedcrfs”.In:IEEEtransactionsonpattern analysis and machine intelligence40.4 (2017), pp. 834–848

work page 2017
[5]

Weaklysupervisedsemanticsegmentationwithboundaryexploration

LiyiChenetal.“Weaklysupervisedsemanticsegmentationwithboundaryexploration”. In:Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXVI 16. Springer. 2020, pp. 347–362

work page 2020

Show all 82 references

[6]

Segment anything model (sam) enhanced pseudo labels for weakly supervised semantic segmentation

Tianle Chen et al. “Segment anything model (sam) enhanced pseudo labels for weakly supervised semantic segmentation”. In:arXiv preprint arXiv:2305.05803(2023)

2023 arXiv
[7]

Semi-supervised semantic segmentation with cross pseudo supervision

Xiaokang Chen et al. “Semi-supervised semantic segmentation with cross pseudo supervision”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021, pp. 2613–2622

2021
[8]

C-cam: Causal cam for weakly supervised semantic segmentation on medical image

Zhang Chen et al. “C-cam: Causal cam for weakly supervised semantic segmentation on medical image”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 11676–11685

2022
[9]

Extracting class activation maps from non- discriminative features as well

Zhaozheng Chen and Qianru Sun. “Extracting class activation maps from non- discriminative features as well”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023, pp. 3135–3144

2023
[10]

Class re-activation maps for weakly-supervised semantic segmentation

Zhaozheng Chen et al. “Class re-activation maps for weakly-supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 969–978

2022
[11]

Attention-based dropout layer for weakly super- vised object localization

Junsuk Choe and Hyunjung Shim. “Attention-based dropout layer for weakly super- vised object localization”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019, pp. 2219–2228. 31 References 32

2019
[12]

Convit: Improving vision transformers with soft convolutional inductive biases

Stéphane d’Ascoli et al. “Convit: Improving vision transformers with soft convolutional inductive biases”. In:International conference on machine learning. PMLR. 2021, pp. 2286– 2296

2021
[13]

Boxsup: Exploiting bounding boxes to super- vise convolutional networks for semantic segmentation

Jifeng Dai, Kaiming He, and Jian Sun. “Boxsup: Exploiting bounding boxes to super- vise convolutional networks for semantic segmentation”. In:Proceedings of the IEEE international conference on computer vision. 2015, pp. 1635–1643

2015
[14]

Imagenet: A large-scale hierarchical image database

Jia Deng et al. “Imagenet: A large-scale hierarchical image database”. In:2009 IEEE conference on computer vision and pattern recognition. Ieee. 2009, pp. 248–255

2009
[15]

Weaklysupervisedsemanticsegmentationbypixel-to-prototypecontrast

YeDuetal.“Weaklysupervisedsemanticsegmentationbypixel-to-prototypecontrast”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 4320–4329

2022
[16]

Toward practical weakly supervised semantic segmentation via point-level supervision

Junsong Fan and Zhaoxiang Zhang. “Toward practical weakly supervised semantic segmentation via point-level supervision”. In:International Journal of Computer Vision 131.12 (2023), pp. 3252–3271

2023
[17]

Learning integral objects with intra-class discriminator for weakly- supervised semantic segmentation

Junsong Fan et al. “Learning integral objects with intra-class discriminator for weakly- supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020, pp. 4283–4292

2020
[18]

Weakly-supervised semantic segmentation for histopathology images based on dataset synthesis and feature consistency constraint

Zĳie Fang et al. “Weakly-supervised semantic segmentation for histopathology images based on dataset synthesis and feature consistency constraint”. In:Proceedings of the AAAI conference on artificial intelligence. Vol. 37. 1. 2023, pp. 606–613

2023
[19]

Shortcut learning in deep neural networks

Robert Geirhos et al. “Shortcut learning in deep neural networks”. In:Nature Machine Intelligence2.11 (2020), pp. 665–673

2020
[20]

Mitigating undisciplined over-smoothing in transformer for weakly supervised semantic segmentation

Jingxuan He et al. “Mitigating undisciplined over-smoothing in transformer for weakly supervised semantic segmentation”. In:arXiv preprint arXiv:2305.03112(2023)

2023 arXiv
[21]

Progressive feature self-reinforcement for weakly supervised semantic segmentation

Jingxuan He et al. “Progressive feature self-reinforcement for weakly supervised semantic segmentation”. In:Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 38. 3. 2024, pp. 2085–2093

2024
[22]

Distilling the knowledge in a neural network

Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. “Distilling the knowledge in a neural network”. In:arXiv preprint arXiv:1503.02531(2015)

2015 arXiv
[23]

Weakly supervised instance segmentation using the bounding box tightness prior

Cheng-Chun Hsu et al. “Weakly supervised instance segmentation using the bounding box tightness prior”. In:Advances in neural information processing systems32 (2019)

2019
[24]

Project RISE: recognizing industrial smoke emissions

Yen-Chia Hsu et al. “Project RISE: recognizing industrial smoke emissions”. In:Pro- ceedings of the AAAI conference on artificial intelligence. Vol. 35. 17. 2021, pp. 14813– 14821

2021
[25]

L2g: A simple local-to-global knowledge transfer framework for weakly supervised semantic segmentation

Peng-Tao Jiang et al. “L2g: A simple local-to-global knowledge transfer framework for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 16886–16896. References 33

2022
[26]

Layercam: Exploring hierarchical class activation maps for localization

Peng-Tao Jiang et al. “Layercam: Exploring hierarchical class activation maps for localization”. In:IEEE Transactions on Image Processing30 (2021), pp. 5875–5888

2021
[27]

Segmentanything

AlexanderKirillovetal.“Segmentanything”.In:ProceedingsoftheIEEE/CVFinternational conference on computer vision. 2023, pp. 4015–4026

2023
[28]

From sam to cams: Exploring segment anything model for weakly supervised semantic segmentation

Hyeokjun Kweon and Kuk-Jin Yoon. “From sam to cams: Exploring segment anything model for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 19499–19509

2024
[29]

Weakly Supervised Semantic Segmen- tation via Dual-Stream Contrastive Learning of Cross-Image Contextual Information

Qi Lai, Chi-Man Vong, and Chuangquan Chen. “Weakly Supervised Semantic Segmen- tation via Dual-Stream Contrastive Learning of Cross-Image Contextual Information”. In:IEEE Transactions on Industrial Informatics(2024)

2024
[30]

Ficklenet: Weakly and semi-supervised semantic image segmenta- tion using stochastic inference

Jungbeom Lee et al. “Ficklenet: Weakly and semi-supervised semantic image segmenta- tion using stochastic inference”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019, pp. 5267–5276

2019
[31]

Weaklysupervisedsemanticsegmentationusingout-of-distribution data

JungbeomLeeetal.“Weaklysupervisedsemanticsegmentationusingout-of-distribution data”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 16897–16906

2022
[32]

Railroad is not a train: Saliency as pseudo-pixel supervision for weakly supervised semantic segmentation

Seungho Lee et al. “Railroad is not a train: Saliency as pseudo-pixel supervision for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021, pp. 5495–5505

2021
[33]

Transcam: Transformer attention-based cam refinement for weakly supervised semantic segmentation

Ruiwen Li et al. “Transcam: Transformer attention-based cam refinement for weakly supervised semantic segmentation”. In:Journal of Visual Communication and Image Representation92 (2023), p. 103800

2023
[35]

Understanding the role of the projector in knowledge distillation

Roy Miles and Krystian Mikolajczyk. “Understanding the role of the projector in knowledge distillation”. In:Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 38. 5. 2024, pp. 4233–4241

2024
[36]

Learning cnn on vit: A hybrid model to explicitly class-specific boundaries for domain adaptation

Ba Hung Ngo et al. “Learning cnn on vit: A hybrid model to explicitly class-specific boundaries for domain adaptation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 28545–28554

2024
[37]

Background-aware pooling and noise-aware loss for weakly-supervised semantic segmentation

Youngmin Oh, Beomjun Kim, and Bumsub Ham. “Background-aware pooling and noise-aware loss for weakly-supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021, pp. 6913–6922

2021
[38]

Semi-supervised semantic segmen- tation with cross-consistency training

Yassine Ouali, Céline Hudelot, and Myriam Tami. “Semi-supervised semantic segmen- tation with cross-consistency training”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020, pp. 12674–12684. References 34

2020
[39]

Weakly-andsemi-supervisedlearningofadeepconvolutional network for semantic image segmentation

GeorgePapandreouetal.“Weakly-andsemi-supervisedlearningofadeepconvolutional network for semantic image segmentation”. In:Proceedings of the IEEE international conference on computer vision. 2015, pp. 1742–1750

2015
[42]

Learning transferable visual models from natural language supervision

Alec Radford et al. “Learning transferable visual models from natural language supervision”. In:International conference on machine learning. PmLR. 2021, pp. 8748–8763

2021
[43]

Do vision transformers see like convolutional neural networks?

Maithra Raghu et al. “Do vision transformers see like convolutional neural networks?” In:Advances in neural information processing systems34 (2021), pp. 12116–12128

2021
[44]

Boundary-enhanced co-training for weakly supervised semantic segmentation

Shenghai Rong et al. “Boundary-enhanced co-training for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023, pp. 19574–19584

2023
[45]

Learning affinity from attention: End-to-end weakly-supervised semantic segmentation with transformers

Lixiang Ru et al. “Learning affinity from attention: End-to-end weakly-supervised semantic segmentation with transformers”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 16846–16855

2022
[46]

Mobilenetv2: Inverted residuals and linear bottlenecks

Mark Sandler et al. “Mobilenetv2: Inverted residuals and linear bottlenecks”. In: ProceedingsoftheIEEEconferenceoncomputervisionandpatternrecognition.2018,pp.4510– 4520

2018
[47]

Grad-cam: Visual explanations from deep networks via gradient-based localization

Ramprasaath R Selvaraju et al. “Grad-cam: Visual explanations from deep networks via gradient-based localization”. In:Proceedings of the IEEE international conference on computer vision. 2017, pp. 618–626

2017
[48]

Self-supervised difference detection for weakly- supervised semantic segmentation

Wataru Shimoda and Keĳi Yanai. “Self-supervised difference detection for weakly- supervised semantic segmentation”. In:Proceedings of the IEEE/CVF international confer- ence on computer vision. 2019, pp. 5208–5217

2019
[49]

Segmenter:Transformerforsemanticsegmentation

RobinStrudeletal.“Segmenter:Transformerforsemanticsegmentation”.In:Proceedings of the IEEE/CVF international conference on computer vision. 2021, pp. 7262–7272

2021
[50]

Yukun Su et al.Context Decoupling Augmentation for Weakly Supervised Semantic Segmen- tation. 2021. arXiv:2103.01795 [cs.CV].url:https://arxiv.org/abs/2103.01795

2021 arXiv
[51]

Massive activations in large language models

Mingjie Sun et al. “Massive activations in large language models”. In:arXiv preprint arXiv:2402.17762(2024)

2024 arXiv
[52]

All-pairs consistency learning forweakly supervised semantic segmentation

Weixuan Sun et al. “All-pairs consistency learning forweakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 826–837. References 35

2023
[53]

Hunting attributes: Context prototype-aware learning for weakly supervised semantic segmentation

Feilong Tang et al. “Hunting attributes: Context prototype-aware learning for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 3324–3334

2024
[54]

Training data-efficient image transformers & distillation through attention

Hugo Touvron et al. “Training data-efficient image transformers & distillation through attention”. In:International conference on machine learning. PMLR. 2021, pp. 10347–10357

2021
[55]

A hybrid resnet-vit approach to bridge the global and local features for myocardial infarction detection

Junaid Abdul Wahid et al. “A hybrid resnet-vit approach to bridge the global and local features for myocardial infarction detection”. In:Scientific Reports14.1 (2024), p. 4359

2024
[56]

Cycle-consistent learning for weakly supervised semantic segmen- tation

Bin Wang et al. “Cycle-consistent learning for weakly supervised semantic segmen- tation”. In:Proceedings of the 3rd International Workshop on Human-Centric Multimedia Analysis. 2022, pp. 7–13

2022
[57]

Score-CAM: Score-weighted visual explanations for convolutional neuralnetworks

Haofan Wang et al. “Score-CAM: Score-weighted visual explanations for convolutional neuralnetworks”.In:ProceedingsoftheIEEE/CVFconferenceoncomputervisionandpattern recognition workshops. 2020, pp. 24–25

2020
[58]

Self-supervised equivariant attention mechanism for weakly super- vised semantic segmentation

Yude Wang et al. “Self-supervised equivariant attention mechanism for weakly super- vised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020, pp. 12275–12284

2020
[59]

When cnn meet with vit: Towards semi-supervised learning for multi-class medical image semantic segmentation

Ziyang Wang et al. “When cnn meet with vit: Towards semi-supervised learning for multi-class medical image semantic segmentation”. In:European conference on computer vision. Springer. 2022, pp. 424–441

2022
[60]

Object region mining with adversarial erasing: A simple classi- fication to semantic segmentation approach

Yunchao Wei et al. “Object region mining with adversarial erasing: A simple classi- fication to semantic segmentation approach”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2017, pp. 1568–1576

2017
[61]

WPS-SAM: Towards Weakly-Supervised Part Segmentation with FoundationModels

Xin-Jian Wu et al. “WPS-SAM: Towards Weakly-Supervised Part Segmentation with FoundationModels”.In:EuropeanConferenceonComputerVision.Springer.2024,pp.314– 333

2024
[62]

DINO is Also a Semantic Guider: Exploiting Class-aware Affinity for Weakly Supervised Semantic Segmentation

Yuanchen Wu et al. “DINO is Also a Semantic Guider: Exploiting Class-aware Affinity for Weakly Supervised Semantic Segmentation”. In:Proceedings of the 32nd ACM International Conference on Multimedia. 2024, pp. 1389–1397

2024
[63]

Dupl: Dual student with trustworthy progressive learning for robust weakly supervised semantic segmentation

Yuanchen Wu et al. “Dupl: Dual student with trustworthy progressive learning for robust weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 3534–3543

2024
[64]

Wider or deeper: Revisiting the resnet model for visual recognition

Zifeng Wu, Chunhua Shen, and Anton Van Den Hengel. “Wider or deeper: Revisiting the resnet model for visual recognition”. In:Pattern recognition90 (2019), pp. 119–133

2019
[65]

SegFormer: Simple and efficient design for semantic segmentation with transformers

Enze Xie et al. “SegFormer: Simple and efficient design for semantic segmentation with transformers”. In:Advances in neural information processing systems34 (2021), pp. 12077– 12090. References 36

2021
[66]

Clims: Cross language image matching for weakly supervised semantic segmentation

Jinheng Xie et al. “Clims: Cross language image matching for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 4483–4492

2022
[67]

Boat in the sky: Background decoupling and object-aware pooling for weakly supervised semantic segmentation

Jianjun Xu et al. “Boat in the sky: Background decoupling and object-aware pooling for weakly supervised semantic segmentation”. In:Proceedings of the 30th ACM international conference on multimedia. 2022, pp. 5783–5792

2022
[68]

Multi-class token transformer for weakly supervised semantic segmenta- tion

Lian Xu et al. “Multi-class token transformer for weakly supervised semantic segmenta- tion”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 4310–4319

2022
[69]

Self correspondence distillation for end-to-end weakly-supervised semantic segmentation

Rongtao Xu et al. “Self correspondence distillation for end-to-end weakly-supervised semantic segmentation”. In:Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. 3. 2023, pp. 3045–3053

2023
[70]

Transmission-guided bayesian generative model for smoke segmentation

Siyuan Yan, Jing Zhang, and Nick Barnes. “Transmission-guided bayesian generative model for smoke segmentation”. In:Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 36. 3. 2022, pp. 3009–3017

2022
[71]

More: Class patch attention needs regularization for weakly supervised semantic segmentation

Zhiwei Yang et al. “More: Class patch attention needs regularization for weakly supervised semantic segmentation”. In:Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 39. 9. 2025, pp. 9400–9408

2025
[72]

Separateandconquer:Decouplingco-occurrenceviadecomposition and representation for weakly supervised semantic segmentation

ZhiweiYangetal.“Separateandconquer:Decouplingco-occurrenceviadecomposition and representation for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 3606–3615

2024
[73]

Fine-grainedBackgroundRepresentationforWeaklySupervisedSemantic Segmentation

XuYinetal.“Fine-grainedBackgroundRepresentationforWeaklySupervisedSemantic Segmentation”. In:IEEE Transactions on Circuits and Systems for Video Technology(2024)

2024
[74]

Dynamic feature regularized loss for weakly supervised semantic segmentation

Bingfeng Zhang, Jimin Xiao, and Yao Zhao. “Dynamic feature regularized loss for weakly supervised semantic segmentation”. In:arXiv preprint arXiv:2108.01296(2021)

2021 arXiv
[75]

Frozen clip: A strong backbone for weakly supervised semantic segmentation

Bingfeng Zhang et al. “Frozen clip: A strong backbone for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 3796–3806

2024
[76]

Weaklysupervisedsemanticsegmentationviaalternateself-dual teaching

DingwenZhangetal.“Weaklysupervisedsemanticsegmentationviaalternateself-dual teaching”. In:IEEE Transactions on Image Processing(2023)

2023
[77]

Cumulative spatial knowledge distillation for vision transformers

Borui Zhao, Renjie Song, and Jiajun Liang. “Cumulative spatial knowledge distillation for vision transformers”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 6146–6155

2023
[78]

Transformer-cnn cohort: Semi-supervised semantic segmentation by the best of both students

Xu Zheng et al. “Transformer-cnn cohort: Semi-supervised semantic segmentation by the best of both students”. In:2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE. 2024, pp. 11147–11154

2024
[79]

Learningdeepfeaturesfordiscriminativelocalization

BoleiZhouetal.“Learningdeepfeaturesfordiscriminativelocalization”.In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 2921–2929. References 37

2016
[80]

Regional semantic contrast and aggregation for weakly supervised semantic segmentation

Tianfei Zhou et al. “Regional semantic contrast and aggregation for weakly supervised semantic segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022, pp. 4299–4309

2022
[81]

A good student is cooperative and reliable: Cnn-transformer collabo- rative learning for semantic segmentation

Jinjing Zhu et al. “A good student is cooperative and reliable: Cnn-transformer collabo- rative learning for semantic segmentation”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 11720–11730

2023
[82]

Weaktr: Exploring plain vision transformer for weakly-supervised semantic segmentation

Lianghui Zhu et al. “Weaktr: Exploring plain vision transformer for weakly-supervised semantic segmentation”. In:arXiv preprint arXiv:2304.01184(2023)

2023

Pith tools

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

[1] [40]

Precision matters: Precision-aware ensemble for weakly supervised semantic segmentation

Junsung Park and Hyunjung Shim. “Precision matters: Precision-aware ensemble for weakly supervised semantic segmentation”. In:arXiv preprint arXiv:2406.19638(2024)

work page Pith review arXiv 2024

[2] [41]

Usage:Aunifiedseedareagenerationparadigmforweaklysupervised semantic segmentation

ZelinPengetal.“Usage:Aunifiedseedareagenerationparadigmforweaklysupervised semantic segmentation”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 624–634

work page 2023

[3] [34]

Pdseg: Patch-wise distillation and controllable image generation for weakly-supervised histopathology tissue segmentation

Wei-Hua Li et al. “Pdseg: Patch-wise distillation and controllable image generation for weakly-supervised histopathology tissue segmentation”. In:ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2025, pp. 1–5

work page 2025

[4] [1]

Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation

Jiwoon Ahn and Suha Kwak.Learning Pixel-level Semantic Affinity with Image-level SupervisionforWeaklySupervisedSemanticSegmentation.2018.arXiv: 1803.10464 [cs.CV]. url:https://arxiv.org/abs/1803.10464

work page Pith review arXiv 2018

[5] [2]

Single-stage semantic segmentation from image labels

Nikita Araslanov and Stefan Roth. “Single-stage semantic segmentation from image labels”.In:ProceedingsoftheIEEE/CVFconferenceoncomputervisionandpatternrecognition. 2020, pp. 4253–4262

2020

[6] [3]

Attentiontoscale:Scale-awaresemanticimagesegmentation

Liang-ChiehChenetal.“Attentiontoscale:Scale-awaresemanticimagesegmentation”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 3640–3649

2016

[7] [4]

Deeplab: Semantic image segmentation with deep convolu- tionalnets,atrousconvolution,andfullyconnectedcrfs

Liang-Chieh Chen et al. “Deeplab: Semantic image segmentation with deep convolu- tionalnets,atrousconvolution,andfullyconnectedcrfs”.In:IEEEtransactionsonpattern analysis and machine intelligence40.4 (2017), pp. 834–848

work page 2017

[8] [5]

Weaklysupervisedsemanticsegmentationwithboundaryexploration

LiyiChenetal.“Weaklysupervisedsemanticsegmentationwithboundaryexploration”. In:Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXVI 16. Springer. 2020, pp. 347–362

work page 2020