Pith. sign in

REVIEW 5 major objections 5 minor 53 references

Semantic segmentation with reward

T0 review · 5 major / 5 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 This paper claims that semantic segmentation can be trained from a single per-image score, reaching within 1.88 mIoU of pixel-supervised training on Pascal Context and beating weakly supervised methods on VOC2012.

desk verdict The paper's central 'beyond labels' claim collapses because the image-level reward is mIoU computed from dense ground truth, yet the PSR/PSD heuristics for training a segmentation network with a scalar reward are a real, if incomplete, contribution. read the letter →

arxiv 2505.17905 v1 pith:B33NR5XQ submitted 2025-05-23 cs.CV

Xie Ting , Ye Huang , Zhilin Liu , Lixin Duan This is my paper · ORCID
classification cs.CV
keywords semanticsegmentationreinforcementlearningimage-levelrewardweaklysupervisedadvantagenormalizationprogressivescalerewardspairwisespatialdifference
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

RSS (Reward in Semantic Segmentation) is an attempt to replace pixel-level labels with reward signals in semantic segmentation, at two granularities: per-pixel rewards and a single global image score. At the pixel level, the network samples actions from its predicted class distributions and rewards correct samples with 1 and wrong ones with 0, using synchronized advantage normalization. At the image level, the reward is the mIoU (mean intersection over union) of the whole predicted map; the authors report that naive training fails, while progressive scale rewards (PSR) and pairwise spatial difference (PSD) make it converge to 57.93 mIoU on Pascal Context, within 1.88 of pixel-supervised training, and to 82.6/84.7 mIoU on VOC2012, above recent weak-supervision methods. The contribution is a practical recipe that turns a sparse global score into a dense training signal; if it holds, segmentation (and visual encoders trained through it) could be supervised without dense annotations, using any scalar quality feedback.

What carries the argument

The central mechanism is the conversion of a global scalar reward into per-pixel advantages via paired stochastic action maps. A sampled action map is a per-pixel class choice drawn from the predicted probability distribution; comparing two such maps, PSD gives the score difference between the two maps as the advantage to pixels that differ between them and zero to pixels that agree, while PSR performs the sampling at several downsampled resolutions so that early training receives rewards from a small action space. SyncAN is a cross-accelerator normalization of those advantages, inspired by synchronized batch normalization, and cold start provides pretrained initialization; together these parts are what make an image-level score usable for convergence.

What would settle it

Run RSS image-level on Pascal Context with the mIoU reward replaced by a scalar that is available without ground-truth maps, such as a pre-trained no-reference quality score or a simulator's feedback, keeping PSR, PSD, and cold start fixed; if convergence or the 57.93 mIoU collapses, the paper's central claim that global feedback can replace dense labels is not supported by the current experiments.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims to be the first practical application of reward-based reinforcement learning to pure semantic segmentation at two granular levels. In the pixel-level setting, categorical sampling of actions per pixel plus GRPO-style advantage normalization (with SyncAN for multi-device training) yields mIoU close to cross-entropy training, 58.92 with cold start versus 59.81 for supervised learning. In the image-level setting, the designed reward is a single scalar per image, computed as mIoU in the experiments; the discovery is that this globally sparse reward becomes trainable when action maps are sampled at multiple downscaled resolutions (PSR) and when the difference in global scores between two sampled maps is assigned as advantage to the pixels where the maps disagree, with zero advantage where they agree (PSD). With these mechanisms, cold-start pretraining, and a large learning-rate multiplier on the last layer, image-level RSS reaches 57.93 mIoU on Pascal Context and, on VOC2012, 82.6/84.7 mIoU, which the paper reads as outperforming weakly supervised methods that use only image-level class labels.

Load-bearing premise

The whole argument depends on having a score that measures segmentation quality without per-pixel labels, but the experiments use mIoU, which is computed from the full ground-truth map; if no such label-free score is supplied, the claim of training beyond labels collapses.

Editorial extensions

If this is right

  • If the image-level results hold, a semantic segmentation network, and the visual encoder inside it, can be trained from a single score per image instead of pixel-level annotations.
  • The PSR and PSD pairing is the key to convergence: with the naive image-level pipeline the model stays near 0.32 mIoU, while adding PSR, PSD, and cold start brings it to 57.93 on Pascal Context.
  • The recipe transfers across architectures: DeepLabv3 with ResNet-101 reaches 44.16 mIoU and Segmenter with EVA02-L reaches 62.37 mIoU using only image-level rewards.
  • Cold starting from a related dataset has a large effect: on Cityscapes, a Mapillary Vistas cold start raises image-level RSS from about 70.48 to 75.92 mIoU under single-scale evaluation.
  • Because the reward is a scalar, the paper's stated path is to replace mIoU with real-world feedback or scores produced by a large language model, which would keep the same advantage machinery.

Reading between the lines

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

  • Extension not in the paper: the PSD trick of turning paired global scores into dense pixel advantages is a general recipe for sparse-reward dense-prediction tasks, so the same idea could apply to depth estimation, optical flow, or surface-normal prediction when only a global quality score is available.
  • Editorial: the reported image-level reward is mIoU computed from full ground-truth maps, so the comparison with weakly supervised methods is not strictly apples-to-apples; those baselines receive image-level class labels, while RSS receives a dense-label-derived score. A cleaner test would use a reward that is genuinely obtainable without any segmentation labels.
  • Testable next step the authors mention but do not run: replace the mIoU reward with a learned no-reference quality model or a simulator's scalar feedback and measure how much of the 57.93 mIoU on Pascal Context survives.
  • Because the method's convergence depends on paired sampled action maps, it would be interesting to see whether increasing the number of sampled maps per image beyond pairs improves the advantage estimates and reduces the variance of the image-level signal.
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

5 major / 5 minor

Summary. The paper proposes Reward in Semantic Segmentation (RSS), a reinforcement-learning-based framework for training semantic segmentation networks. Two reward granularities are considered: a pixel-level reward that gives each sampled action a per-pixel correctness signal, and an image-level reward that scores an entire sampled action map. To make image-level rewards trainable, the paper introduces progressive scale rewards (PSR) and pair-wise spatial difference (PSD), along with a synchronized advantage normalization (SyncAN) and cold start. Experiments on Pascal Context, VOC2012, and Cityscapes are reported; the authors claim that the image-level variant reaches mIoU within 1.88 points of pixel-level supervised learning on Pascal Context and that it outperforms state-of-the-art weakly supervised segmentation methods that also rely on image-level signals.

Significance. If the central claims held, the paper would offer a meaningful step toward training dense predictors from scalar feedback rather than per-pixel labels, and the two-level reward formulation would be a useful addition to the RL-for-vision toolbox. The empirical results on Pascal Context, in particular the jump from 0.32 to 57.93 mIoU when PSR, PSD, and cold start are combined, suggest that the proposed techniques do address a real optimization difficulty. However, the significance is seriously undercut by two facts: the image-level reward is the mIoU computed from full pixel-level ground truth, so the method is not actually trained from 'beyond labels' feedback; and the comparison with weakly supervised baselines is therefore apples-to-oranges. The paper does not release code or training details, making the claims difficult to verify.

major comments (5)
  1. [3.2.1 and Section 6] The image-level reward is defined as mIoU ('we utilize the mean Intersection over Union (mIoU) as the image-level reward'). Computing mIoU requires the full ground-truth segmentation map for every training image. The abstract and Section 6 state that RSS 'outperforms existing weakly supervised methods that also rely solely on image-level signals during training'; however, the weakly supervised baselines (CLIMS++, MoRe, ExCEL, DOEI) receive only image-level class labels, with no pixel annotations. RSS therefore trains with a scalar derived from dense pixel-level supervision, while the baselines do not. This is an invalid comparison, and the 'beyond labels' framing collapses. Appendix A.3 acknowledges that mIoU was used instead of real-world feedback, but that does not resolve the issue that mIoU is a deterministic function of the dense label map.
  2. [4 (Implementation)] No training configuration is reported: the optimizer, learning-rate schedule, base learning rate, number of iterations, batch size, input resolution, data splits, number of sampled actions α, the scale list for PSR, and random seeds are all unspecified. Table 3 shows that the last-layer learning-rate factor is critical, but without the base learning rate the result cannot be reproduced. The paper also does not provide code. The empirical results are therefore not independently verifiable, which is a load-bearing problem for a paper whose main contribution is empirical.
  3. [Table 5 and Section 6] The comparison in Table 5 is not 'under similar conditions': RSS uses ConvNeXt-v2-Nano, a recent strong backbone, while the baselines use ResNet-101, WideResNet-38, or ViT-B. The table does not state which training split was used for RSS or for the baselines, nor does it report the training protocol for RSS on VOC2012. Given that the supervision signal is fundamentally different (dense-label-derived mIoU vs. image-level class labels) and the backbone is different, the claim of 'significantly outperforms' is not a controlled comparison and should be removed or re-framed.
  4. [3.2.3 (Pair-wise spatial difference)] PSD assigns the full reward difference between two sampled action maps as the advantage to pixels where the maps differ, and zero advantage elsewhere. No derivation or bias/variance analysis is provided; the method is presented as a heuristic. Since PSD is one of the two key techniques (with PSR) that make image-level training converge, the paper should at least discuss the conditions under which this advantage estimator is consistent, or explicitly frame it as an empirical trick. The current treatment leaves the reader unable to judge whether the reported convergence is a property of the method or of the particular heuristic.
  5. [1 and 2 (Introduction and Related work)] The claim that RSS is 'the first practical application of reward-based reinforcement learning on pure semantic segmentation' is not supported by the related-work section, which discusses weakly supervised segmentation and RLHF/GRPO but does not survey prior RL-based approaches to segmentation. If such prior work exists, the claim needs qualification; if not, the survey is incomplete. This is a novelty claim that should be substantiated.
minor comments (5)
  1. [Table 3] The table title contains a typo: 'Leanring rate factor' should be 'Learning rate factor'.
  2. [2.3] The section title 'rainforment learning' should be 'reinforcement learning'.
  3. [7] The text refers to 'Fig. [51]' but should refer to 'Fig. 6'.
  4. [7] The text cites 71.4% mIoU for RSS on Cityscapes without specifying that this is the multi-scale no-cold-start result from Table 6 in the appendix; please clarify the exact configuration.
  5. [General] There are numerous informal and ungrammatical phrases (e.g., 'Let's do the warmup', 'the model nearly does not converge at all'); the paper needs careful editing.

Circularity Check

3 steps flagged · score 7.0 of 10

The image-level 'beyond labels' contribution reduces to optimizing mIoU computed from dense ground-truth labels, and the weakly-supervised comparison is invalid because baselines receive only image-level class tags while RSS receives a dense-label-derived scalar.

  1. self definitional [Section 3.2.1 (Naive pipeline)]
    "In this research, we utilize the mean Intersection over Union (mIoU) as the image-level reward, as we are focusing on challenges that occur after receiving feedback. In the future, it can be replaced with real-world feedback or scores generated by a large language model."

    mIoU is defined from the dense ground-truth segmentation map, i.e., from the exact pixel-level labels that the paper claims to avoid. Calling this an 'image-level reward' is self-definitional: the scalar is not an independent low-dimensional feedback signal but a deterministic aggregate of dense per-pixel labels. The abstract and Section 1 claim training with 'feedback beyond traditional labels' and 'only global feedback'; with mIoU as the reward, the supervision still originates from pixel-level ground truth, so the 'beyond labels' premise is true only in the trivial sense that the dense labels have been summed into one number.

  2. fitted input called prediction [Section 3.2.3 and Section 5.2.1 (Table 2)]
    "PSD directly utilizes the score reward difference between the two action maps as the advantage of different parts of the first action map over the second. For the same parts, the advantages are set to 0. ... After applying PSR, PSD, and Cold Start, the mIoU reached 57.93%, which is only 1.88% lower than that of pixel-level supervised learning, even though RSS only has global image-level feedback during training."

    The reward is mIoU and PSD uses mIoU differences as advantages, so the optimization objective is directly the reported metric. The reported 57.93% mIoU is therefore not an independent prediction of generalization; it is the value of the training objective being maximized with dense ground truth. In RL, reporting the reward itself is normal, but here the reward is exactly the evaluation metric and the paper presents the resulting mIoU as evidence that 'a global image-level feedback' can supervise segmentation. The high mIoU is forced by construction because the training signal is the metric itself, not a label-free external feedback.

1 more flagged steps
  1. other [Section 6 (Compare with weakly supervised methods on VOC2012)]
    "As shown in Tab. 5, our proposed RSS significantly outperforms state-of-the-art image-level weakly supervised methods under similar conditions using only image-level signals."

    The weakly supervised baselines (CLIMS++, MoRe, ExCEL, DOEI) use only image-level class labels and no pixel-level ground truth, whereas RSS computes its image-level reward from the dense ground-truth segmentation map via mIoU. The phrase 'similar conditions using only image-level signals' is therefore misleading: the two settings are not equivalent, because RSS has access to dense pixel-level supervision aggregated into a scalar, while the baselines do not. The claimed superiority over weakly supervised methods is not an apples-to-apples comparison; it is a consequence of the asymmetric supervision, so the central comparative claim is unsupported by construction.

full rationale

The paper's central claim is that RSS can train semantic segmentation from 'global feedback' and that 'the RSS, which utilizes an image-level reward, outperforms existing weakly supervised methods that also rely solely on image-level signals.' However, Section 3.2.1 states the image-level reward is mIoU, which requires the dense ground-truth segmentation map for every training image. Computing mIoU uses the same dense labels that the paper claims to avoid, so the 'beyond labels' contribution reduces by definition: the method is not trained from a label-free scalar but from a deterministic function of the pixel-level annotation. Consequently, the reported mIoU values, including the 57.93% on Pascal Context and the VOC2012 comparison in Table 5, are not independent predictions: they are the training objective being optimized, and the weakly-supervised baselines are at an information disadvantage because they receive only image-level class tags while RSS receives dense-label-derived rewards. The appendix's limitation statement acknowledges that mIoU was used instead of real-world feedback, but the abstract and introduction still assert the first practical application and the outperformance claim. The paper's PSR and PSD components are legitimate technical contributions for RL convergence, but they do not rescue the central 'label-free' framing. The overall circularity score is 7 because the core image-level reward and the headline comparison both reduce to using dense ground truth, even though some sub-components (SyncAN, PSR, PSD) are non-circular engineering details.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central machinery is a set of heuristics (SyncAN, PSR, PSD) layered onto GRPO. The most consequential unstated dependence is the mIoU reward, which requires dense labels and makes the 'beyond labels' framing misleading. PSR, PSD, and SyncAN are algorithmic components, not postulated entities.

free parameters (3)
  • last layer learning rate multiplier = 400.0
    Table 3 sweeps multipliers {1, 4, 40, 400}; only 400 with cold start yields 57.93 mIoU, showing a strong dependence on this tuned value.
  • number of action maps alpha = not reported
    The method samples alpha action maps per image (Section 3.2, Figure 2), but alpha is never given in the paper. It controls the advantage statistics and is a hand-chosen hyperparameter.
  • progressive scale reward scale list = not reported
    The PSR scales S1..SN (Section 3.2.2, Figure 3) are not specified; the ablation shows multi-scale is critical (50.37 vs 32.45 mIoU), so the exact list affects results.
assumptions (4)
  • domain assumption mIoU reward is computable during training
    Section 3.2.1 states 'we utilize the mean Intersection over Union (mIoU) as the image-level reward.' Computing mIoU requires ground-truth pixel labels, so the method is not label-free.
  • domain assumption GRPO-style advantage normalization without a value model is valid for segmentation
    The method adopts GRPO-style normalization (Section 3.1.1) designed for token-level language-model MDPs; no argument is given for its validity on spatial pixel actions.
  • ad hoc to paper Pairwise spatial difference provides a valid advantage signal
    Section 3.2.3 assigns advantage equal to the reward difference on differing pixels and 0 elsewhere; this heuristic is untested against alternatives and has no theoretical justification.
  • ad hoc to paper Multi-scale downsampling preserves the optimal policy
    PSR reduces the action space by resizing action maps (Figure 3); the paper does not show that lower-resolution actions are sufficient for fine-grained segmentation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semantic segmentation with reward." pith.science (2026). https://pith.science/paper/B33NR5XQ

@misc{pith2026250517905,
  author       = {Pith},
  title        = {Pith review of: Semantic segmentation with reward},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B33NR5XQ}},
  note         = {Machine review of arXiv:2505.17905}
}
read the original abstract

In real-world scenarios, pixel-level labeling is not always available. Sometimes, we need a semantic segmentation network, and even a visual encoder can have a high compatibility, and can be trained using various types of feedback beyond traditional labels, such as feedback that indicates the quality of the parsing results. To tackle this issue, we proposed RSS (Reward in Semantic Segmentation), the first practical application of reward-based reinforcement learning on pure semantic segmentation offered in two granular levels (pixel-level and image-level). RSS incorporates various novel technologies, such as progressive scale rewards (PSR) and pair-wise spatial difference (PSD), to ensure that the reward facilitates the convergence of the semantic segmentation network, especially under image-level rewards. Experiments and visualizations on benchmark datasets demonstrate that the proposed RSS can successfully ensure the convergence of the semantic segmentation network on two levels of rewards. Additionally, the RSS, which utilizes an image-level reward, outperforms existing weakly supervised methods that also rely solely on image-level signals during training.

Figures

Figures reproduced from arXiv: 2505.17905 by the authors.

Figure 1
Figure 1. Architecture comparison between pixel-level supervised learning and reinforcement learn [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The image-level sampling process is similar to pixel-level sampling. However, the reward is [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Scaling multi-resolution (S1 to SN ) action maps helps to reduce the action sampling space. Zoom in to see better. AM: Action map. H: Height. W: Width. α: Number of actions. a future where we can train a semantic segmentation network, or a general visual encoder, in real￾world scenarios or video games. Instead of traditional labeling, all we need is a score at each time step. Additionally, for multi-modal applicatio… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Utilizing the Pairwise Spatial Difference (PSD) between action maps [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Visualization of RSS (image-level reward) on Pascal Context dataset. Although image-level [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison of weakly supervised methods CARB and our proposed method [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Visualization of applying Mapillary Vistas cold start and doubling the size of the RSS on [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 43 canonical work pages

  1. [1]

    J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in: IEEE Conference on Computer Vision and Pattern Recognition, 2015

    2015

  2. [2]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: MICCAI, 2015

    2015

  3. [3]

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, A. L. Yuille, Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs, IEEE Transactions on Pattern Analysis and Machine Intelligence (2017)

    work page 2017

  4. [4]

    L.-C. Chen, Y . Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder-decoder with atrous separable convolution for semantic image segmentation, in: European Conference on Computer Vision, 2018

    work page 2018

  5. [5]

    Huang, Q

    Y . Huang, Q. Wang, W. Jia, X. He, See more than once–kernel-sharing atrous convolution for semantic segmentation, Neuro Computing (2021)

    work page 2021

  6. [6]

    Y . Yuan, L. Huang, J. Guo, C. Zhang, X. Chen, J. Wang, Ocnet: Object context network for scene parsing, International Journal of Computer Vision (2021)

    work page 2021

  7. [7]

    J. Fu, J. Liu, H. Tian, Y . Li, Y . Bao, Z. Fang, H. Lu, Dual attention network for scene segmenta- tion, in: IEEE Conference on Computer Vision and Pattern Recognition, 2019

    work page 2019

  8. [8]

    Y . Yuan, X. Chen, J. Wang, Object-contextual representations for semantic segmentation, in: European Conference on Computer Vision, 2020

    work page 2020

Show all 53 references
  1. [9]

    Zhang, Y

    F. Zhang, Y . Chen, Z. Li, Z. Hong, J. Liu, F. Ma, J. Han, E. Ding, Acfnet: Attentional class feature network for semantic segmentation, in: International Conference on Computer Vision, 2019

    2019

  2. [10]

    Huang, D

    Y . Huang, D. Kang, L. Chen, X. Zhe, W. Jia, L. Bao, X. He, Car: Class-aware regularizations for semantic segmentation, in: European Conference on Computer Vision, 2022

    2022

  3. [11]

    Cheng, A

    B. Cheng, A. G. Schwing, A. Kirillov, Per-pixel classification is not all you need for semantic segmentation, in: Conference on Neural Information Processing Systems, 2021

    2021

  4. [12]

    Huang, D

    Y . Huang, D. Kang, S. Gao, W. Li, L. Duan, Ieee tcsvt, IEEE Transactions on Circuits and Systems for Video Technology (2024) 1–1doi:10.1109/TCSVT.2024.3393632

    2024

  5. [13]

    Huang, D

    Y . Huang, D. Kang, L. Chen, W. Jia, X. He, L. Duan, X. Zhe, L. Bao, Card: Semantic segmentation with efficient class-aware regularized decoder, IEEE TCSVT (2024) 1–1 doi: 10.1109/TCSVT.2024.3395132

    2024

  6. [14]

    Cheng, I

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, R. Girdhar, Masked-attention mask transformer for universal image segmentation, in: IEEE Conference on Computer Vision and Pattern Recognition, 2022

    2022

  7. [15]

    Zhang, Z

    B. Zhang, Z. Tian, Q. Tang, X. Chu, X. Wei, C. Shen, Y . Liu, Segvit: Semantic segmentation with plain vision transformers, in: Conference on Neural Information Processing Systems, 2022

    2022

  8. [16]

    Y . Ge, Q. Nie, Y . Huang, Y . Liu, C. Wang, F. Zheng, W. Li, L. Duan, Beyond prototypes: Semantic anchor regularization for better representation learning, in: AAAI, 2024

    2024

  9. [17]

    Z. Chen, Y . Duan, W. Wang, J. He, T. Lu, J. Dai, Y . Qiao, Vision transformer adapter for dense predictions, in: International Conference on Learning Representations, 2023

    2023

  10. [18]

    X. Tang, Y . Huang, G. Yin, L. Duan, Vpnext – rethinking dense decoding for plain vision transformer (2025).arXiv:2502.16654. URLhttps://arxiv.org/abs/2502.16654

    2025

  11. [19]

    Q. Wang, W. Jia, X. He, Y . Lu, M. Blumenstein, Y . Huang, S. Lyu, Reelfa: A scene text recog- nizer with encoded location and focused attention, in: International Conference on Document Analysis and Recognition Workshops (ICDARW), 2019. 12

    2019

  12. [20]

    Ranftl, A

    R. Ranftl, A. Bochkovskiy, V . Koltun, Vision transformers for dense prediction, in: ICCV , 2021

    2021

  13. [21]

    Huang, B

    W. Huang, B. Jia, Z. Zhai, S. Cao, Z. Ye, F. Zhao, Z. Xu, Y . Hu, S. Lin, Vision-r1: Incentivizing reasoning capability in multimodal large language models (2025).arXiv:2503.06749. URLhttps://arxiv.org/abs/2503.06749

    2025 arXiv

  14. [22]

    DeepSeek-AI, D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi, X. Zhang, X. Yu, Y . Wu, Z. F. Wu, Z. Gou, Z. Shao, Z. Li, Z. Gao, A. Liu, B. Xue, B. Wang, B. Wu, B. Feng, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan, D. Dai, D. Chen, D. Ji, E. L...

    2025 arXiv

  15. [23]

    H. Zhao, J. Shi, X. Qi, X. Wang, J. Jia, Pyramid scene parsing network, in: IEEE Conference on Computer Vision and Pattern Recognition, 2017

    2017

  16. [24]

    M. Yang, K. Yu, C. Zhang, Z. Li, K. Yang, Denseaspp for semantic segmentation in street scenes, in: IEEE Conference on Computer Vision and Pattern Recognition, 2018

    2018

  17. [25]

    Z. Zhu, M. Xu, S. Bai, T. Huang, X. Bai, Asymmetric non-local neural networks for semantic segmentation, in: International Conference on Computer Vision, 2019

    2019

  18. [26]

    Zhang, H

    H. Zhang, H. Zhan, C. Wang, J. Xie, Semantic correlation promoted shape-variant context for segmentation, in: IEEE Conference on Computer Vision and Pattern Recognition, 2019

    2019

  19. [27]

    Huang, D

    Y . Huang, D. Kang, W. Jia, X. He, L. liu, Channelized axial attention - considering channel relation within spatial attention for semantic segmentation, in: AAAI, 2022

    2022

  20. [28]

    T.-Y . Lin, P. Dollá, R. Girshick, K. He, B. Hariharan, S. Belongie, Feature pyramid networks for object detection, in: IEEE Conference on Computer Vision and Pattern Recognition, 2017

    2017

  21. [29]

    H. Wu, J. Zhang, K. Huang, K. Liang, Y . Yizhou, Fastfcn: Rethinking dilated convolution in the backbone for semantic segmentation (2019)

    2019

  22. [30]

    T. Xiao, Y . Liu, B. Zhou, Y . Jiang, J. Sun, Unified perceptual parsing for scene understanding, in: European Conference on Computer Vision, 2018

    2018

  23. [31]

    Sixiao, L

    Z. Sixiao, L. Jiachen, Z. Hengshuang, Z. Xiatian, L. Zekun, W. Yabiao, F. Yanwei, F. Jianfeng, X. Tao, T. P. H.S., Z. Li, Rethinking semantic segmentation from a sequence-to-sequence per- spective with transformers, in: IEEE Conference on Computer Vision and Pattern Recognition, 2021

    2021

  24. [32]

    Mottaghi, X

    R. Mottaghi, X. Chen, X. Liu, N.-G. Cho, S.-W. Lee, S. Fidler, R. Urtasun, A. Yuille, The role of context for object detection and semantic segmentation in the wild, in: IEEE Conference on Computer Vision and Pattern Recognition, 2014

    2014

  25. [33]

    K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: IEEE Conference on Computer Vision and Pattern Recognition, 2016. 13

    2016

  26. [34]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, An image is worth 16x16 words: Transformers for image recognition at scale, in: International Conference on Learning Repr...

    2021

  27. [35]

    Everingham, L

    M. Everingham, L. V . Gool, C. K.l.Wiliams, J. Winn, A. Zisserman, The pascal visual object classes (voc) challenge, International Journal of Computer Vision (2009)

    2009

  28. [36]

    T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert-V oss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. M. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray, B....

    2020 arXiv

  29. [37]

    Zhang, K

    H. Zhang, K. Dana, J. Shi, Z. Zhang, X. Wang, A. Tyagi, A. Agrawal, Context encoding for semantic segmentation, in: IEEE Conference on Computer Vision and Pattern Recognition, 2018

    2018

  30. [38]

    Kirillov, R

    A. Kirillov, R. Girshick, K. He, P. Dollar, Panoptic feature pyramid networks, in: IEEE Conference on Computer Vision and Pattern Recognition, 2019

    2019

  31. [39]

    X. Wang, R. Girshick, A. Gupta, K. He, Non-local neural networks, in: IEEE Conference on Computer Vision and Pattern Recognition, 2018

    2018

  32. [40]

    S. Woo, S. Debnath, R. Hu, X. Chen, Z. Liu, I. S. Kweon, S. Xie, Convnext v2: Co-designing and scaling convnets with masked autoencoders, in: IEEE Conference on Computer Vision and Pattern Recognition, 2023

    2023

  33. [41]

    L.-C. Chen, G. Papandreou, F. Schroff, H. Adam, Rethinking atrous convolution for semantic image segmentation (2017).arXiv:1706.05587

    2017 arXiv

  34. [42]

    Robin, G

    S. Robin, G. Ricardo, L. Ivan, S. Cordelia, Segmenter: Transformer for semantic segmentation, in: ICCV , 2021

    2021

  35. [43]

    Y . Fang, W. Wang, B. Xie, Q. Sun, L. Wu, X. Wang, T. Huang, X. Wang, Y . Cao, Eva: Exploring the limits of masked visual representation learning at scale, in: IEEE Conference on Computer Vision and Pattern Recognition, 2023

    2023

  36. [44]

    J. Xie, S. Deng, X. Hou, Z. Luo, L. Shen, Y . Huang, Y . Zheng, M. Z. Shou, Clims++: Cross language image matching with automatic context discovery for weakly supervised seman- tic segmentation, International Journal of Computer Vision (May 2025). doi:10.1007/ s11263-025-02442...

    2025 doi

  37. [45]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, I. Sutskever, Learning transferable visual models from natural language supervision, in: International Conference on Machine Learning, 2021

    2021

  38. [46]

    Z. Wu, C. Shen, A. van den Hengel, Wider or deeper: Revisiting the resnet model for visual recognition, Pattern Recognition 90 (2019) 119–133. doi:https://doi.org/10.1016/j. patcog.2019.01.006. URLhttps://www.sciencedirect.com/science/article/pii/S0031320319300135

    2019 doi

  39. [47]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Dollár, R. Girshick, Segment anything, in: International Conference on Computer Vision, 2023

    2023

  40. [48]

    Z. Yan, Y . Meng, K. Fu, S. Wang, Z. Song, More: Class patch attention needs regularization for weakly supervised semantic segmentation, in: AAAI, 2025

    2025

  41. [49]

    Z. Yang, Y . Meng, K. Fu, F. Tang, S. Wang, Z. Song, Exploring clip’s dense knowledge for weakly supervised semantic segmentation, in: IEEE Conference on Computer Vision and Pattern Recognition, 2025. 14

    2025

  42. [50]

    H. Zhu, Z. Zhang, G. Pang, X. Wang, S. Wen, Y . Bai, D. Ergu, Y . Cai, Y . Zhao, Doei: Dual optimization of embedding information for attention-enhanced class activation maps, arXiv preprint arXiv:2502.15885 (2025)

    2025 arXiv

  43. [51]

    Marius, O

    C. Marius, O. Mohamed, R. Sebastian, R. Timo, E. Markus, B. Rodrigo, F. Uwe, S. Roth, S. Bernt, The cityscapes dataset for semantic urban scene understanding, in: IEEE Conference on Computer Vision and Pattern Recognition, 2016

    2016

  44. [52]

    D. Kim, S. Lee, J. Choe, H. Shim, Weakly supervised semantic segmentation for driving scenes, in: AAAI, V ol. 38, 2024, pp. 2741–2749

    2024

  45. [53]

    Neuhold, T

    G. Neuhold, T. Ollmann, S. R. Bulo, P. Kontschieder, The mapillary vistas dataset for semantic understanding of street scenes., in: International Conference on Computer Vision, 2017. 15

    2017

Pith tools

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