Pith. sign in

REVIEW 3 major objections 4 minor 47 references

Generative Semantic Segmentation via an Observable Semantic-Image Interface and Hierarchical Generator Evidence Alignment

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A generated semantic image can be decoded into a full per-pixel probability distribution, and refining that distribution in its own logit space lifts Cityscapes mIoU from 60.68% to 72.07%.

desk verdict Novel observable interface and careful evaluation, but the UC-Head control weakens the accuracy story; worth peer review with a request for multi-seed evidence. read the letter →

arxiv 2608.11537 v1 pith:J6XA6H5K submitted 2026-08-12 cs.CV cs.AIcs.LGeess.IV

Weize Cai , Yongqi Dong , Zhida Shao , Zixin Fu This is my paper · ORCID
classification cs.CVcs.AIcs.LGeess.IV
keywords generativesemanticsegmentationsemantic-imageinterfacecodebookdecodinghierarchicalfeaturealignmentone-stepdiffusiondistillationpixel-errorrankingprobabilitycalibrationCityscapes
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 Prism proposes that a one-step diffusion-distilled generator can do semantic segmentation by rendering a semantic RGB image, provided a fixed color-codebook decoder turns each rendered pixel into a full class distribution before any learned refinement happens. The core claim is that this image-defined interface should be the reference for the final prediction, and the generator's multi-level features should contribute only an additive residual in the same interface logit space. Direct decoding alone reaches 60.68% mIoU on Cityscapes val500; hierarchical refinement lifts it to 72.07% while cutting expected calibration error from 5.69% to 0.41%. The pre- and post-refinement distributions also feed a fixed readout, Contextual Interface-Hierarchy Disagreement, that ranks remaining pixel errors better than maximum softmax probability on all three datasets without changing the segmentation or adding a forward pass.

What carries the argument

The central object is the fixed distance-based codebook decoder $\Pi_C^{(\tau)}$, which maps a rendered RGB value $v$ to class probabilities by the softmax of negative squared Euclidean distances to class prototypes. This one identity makes the semantic image observable as a probability interface, because every quantity in $p^I$, including top-1 label, confidence, and pairwise log-odds, is recoverable from the rendered image alone. The refinement machinery is Hierarchical Generator Evidence Alignment (HGEA): three generator feature maps of 128, 256, and 512 channels are projected by one-by-one convolutions, group-normalized, bilinearly resampled to the output lattice, concatenated with the input image, the rendered image, and the interface distribution, and passed through two convolutional blocks followed by a zero-initialized one-by-one projection that outputs the logit residual $\Delta z^H$. The zero initialization guarantees that the refined distribution coincides with the interface at the start of training, and the additive form guarantees that refinement acts on the same log-odds space rather than opening a separate prediction path.

What would settle it

Retrain the refiner with those three feature maps replaced by noise or by the same maps after shuffling their spatial positions; if mIoU stays at the flat-refiner level rather than dropping below it, the alignment story is false.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a generated semantic image can be a genuine probabilistic prediction interface, not merely a visualization: with a fixed per-class RGB codebook $C=\{c_k\}$, the rendered color $s(u)$ defines interface logits $z^I_k(u) = -\|s(u)-c_k\|_2^2/\tau^I$ and hence a full distribution $p^I(u)$, whose top-1 label is the Euclidean Voronoi cell and whose pairwise log-odds are closed-form differences of squared distances. The hierarchy of frozen generator features is then used to predict an additive logit residual $\Delta z^H(u)$ through a zero-initialized projection, so the refined distribution $p^H = \mathrm{softmax}(z^I + \Delta z^H)$ starts equal to $p^I$ and remains anchored to the image-defined interface. This yields 72.07% mIoU on Cityscapes val500, 11.39 points above direct-interface decoding, and 0.41% expected calibration error; a matched-capacity three-seed ablation attributes part of the gain specifically to joint multi-level alignment. A second fixed readout, C-IHD, combines pointwise MSP uncertainty, a local average of it, and interface-versus-refinement disagreement to improve AUPR over MSP on frozen predictions in all three evaluated datasets.

Load-bearing premise

The load-bearing premise is that three selected internal feature maps of the generator, after being resized to the output grid and lightly projected, carry spatially aligned semantic evidence, and that this evidence, rather than the extra input images or extra capacity, drives the refinement gain.

Editorial extensions

If this is right

  • If the claim holds, image-form generative segmentation can keep the rendered image as a directly decodable and calibrated probability output, so the image can be inspected and evaluated without access to the model's internal features.
  • The 11.39-point mIoU gain implies that aligning several generator scales inside a single logit space is worth more than a hierarchy-free refiner of identical capacity, which trails by 1.68 +/- 0.11 mIoU points over three seeds.
  • The C-IHD result implies that pixel-error ranking can be improved on already-fixed predictions by comparing a model's own pre- and post-refinement views, with no auxiliary network and no extra forward pass.
  • Source-frozen ACDC results imply that the interface, the refinement, and the readout transfer to adverse weather conditions without adaptation, even though mIoU falls to 46.89% and the method trails the strongest discriminative baseline there.

Reading between the lines

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

  • I would extend the interface idea to any dense prediction that can be color-mapped with a fixed palette, such as depth, surface normals, or panoptic instances, where the same squared-distance decoder could expose an inspectable probability field before refinement.
  • A testable extension is to vary the codebook's minimum inter-prototype distance and measure boundary calibration: the paper's own boundary-versus-interior audit predicts that boundary ECE should degrade faster than interior ECE as the palette becomes denser.
  • Because HGEA starts as the identity and the generator is frozen, one could retrain only the residual head on a new domain and ask whether the interface remains a stable anchor; the paper's ACDC transfer hints this is possible but does not test continuous adaptation.
  • The matched ablations leave open whether multi-level alignment helps through semantic correspondence or through extra spatial context; a refiner fed the same three levels at a single common resolution would separate those explanations.
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.

Referee Report

3 major / 4 minor

Summary. Semantic Prism is a one-step generative semantic segmentation framework with deterministic inference. A distilled pix2pix-Turbo generator renders a semantic RGB image; a fixed class-color codebook and a squared-Euclidean-distance decoder (Eqs. 1–3) convert that image into an explicit per-pixel probability distribution p_I, the observable interface. A lightweight HGEA module aligns three frozen VAE-encoder feature maps and, through a zero-initialized 1x1 projection, predicts an additive logit residual, giving the final distribution p_H = softmax(z_I + Δz_H) (Eqs. 7–9); the rendered image and the fixed decoder therefore remain a reference for the final prediction. A third component, C-IHD, combines MSP uncertainty, its 5x5 local average, and a rho-weighted Jensen-Shannon disagreement between p_I and p_H into a fixed readout for ranking pixel errors (Eqs. 10–12). On Cityscapes val500 the paper reports 72.07% mIoU (11.39 points over direct-interface decoding), 0.41% ECE, and C-IHD AUPR gains over MSP on fixed predictions; a separately trained BDD100K model reaches 62.22% mIoU; and source-frozen transfer to ACDC reaches 46.89% mIoU with the lowest ECE among the methods compared.

Significance. Strengths: the paper ships matched three-seed ablations for every learned refiner variant with paired margins (Table 3, Table S4), bootstrap 95% confidence intervals for the readout gains, a metric-consistency check reproducing the headline 60.68/72.07 numbers from the released val500 predictions, a temperature sweep, a prespecified readout sensitivity grid, a source-frozen transfer audit with per-condition breakdown (Table S11), and explicit stochastic-state resume controls. If the results hold, the framework is significant in three respects: (i) Eqs. (5)–(6) make the pre-refinement distribution fully recoverable from the rendered image, giving generative segmentation an independently evaluable probabilistic interface; (ii) the zero-initialized additive-residual construction keeps the final distribution in the interface logit space, so the rendered image remains a reference rather than a byproduct; and (iii) C-IHD improves fixed-prediction pixel-error ranking without an auxiliary predictor or extra forward pass, with a large AUPR gain on ACDC (0.6580 to 0.7557).

major comments (3)
  1. [Supplementary Table S5; §3.3 (Eqs. 2 and 9); Introduction] The capacity-matched UC-Head control in Table S5 predicts final logits directly from the same three hierarchy levels with the same 190k decoder; on seed 18313 it reaches 73.06 mIoU versus 71.97 for ML-HGEA, and the accompanying text concedes that residual anchoring preserves an explicit prototype-logit reference but is not, by itself, the source of the accuracy gain in this control. This control is the direct test of the paper's central framing, namely that additive refinement in the interface logit space is what reconciles an observable interface with fine-grained accuracy (Introduction, §3.3, Figure 1, Conclusion). Because UC-Head was run for a single seed, the 1.09-point gap has no uncertainty estimate, yet it is the same order of magnitude as the 1.68±0.11 paired margin that Table 3 uses to attribute the gain to multi-level alignment. As reported, the evidence supports the value of hierarchical generator features and of learned refinement per se, but it does not support the claim that the interface-anchored same-logit-space design contributes to the 11.39-point headline gain over direct decoding; the design may even carry an accuracy cost. Please run UC-Head under the full three-seed 36k-step protocol and, if feasible, the canonical 48k protocol, report the paired margin against ML-HGEA with the same bootstrap treatment as Table 3, and align the abstract, contribution (2), and conclusion so that the accuracy gain is attributed to hierarchical refinement of generator features, with interface anchoring presented as an inspectability and transparency property whose quantifiable cost is given by this control.
  2. [Experiments (Experimental Setup); Tables 1 and 3] The headline 72.07% mIoU and the 11.39-point gain over the interface rest on a single checkpoint from the 48k-step curriculum (seed 38001), whereas the matched three-seed ML-HGEA runs use a 36k-step protocol and give 71.43±0.47. The main text reports no uncertainty for the canonical result, and the closest discriminative baseline in Table 1 (SegFormer-B0, 71.13) lies within about two within-method sample standard deviations; several baselines in Tables 1, 2, and 4 are reported without any variance information. Please report the per-seed distribution of the full-protocol result or state explicitly that 72.07 is a single endpoint, and provide error bars or bootstrap intervals for at least the closest baselines so that the primary comparative claims are held to the same evidentiary standard as the internal three-seed ablations.
  3. [Table 5; Table S9; Table S10] The component analysis in the supplement shows that the novel IHD term is, on its own, worse than MSP on Cityscapes (AUROC 0.92958 versus 0.94502; AURC 5.080 versus 4.281×10^-3) and that the in-domain C-IHD gain over MSP is almost entirely carried by Local-MSP, which is a standard 5×5 box average of MSP uncertainty (+0.0031 AUPR absolute on Cityscapes). On ACDC the same fixed functional form behaves very differently: the full C-IHD readout jumps to AUPR 0.7557, far above MSP+Local-MSP (0.6657), while MSP+IHD alone is flat or negative in AUPR (0.6554). This large cross-dataset non-additivity is unexplained, and as written contribution (3) overstates the role of the interface–hierarchy disagreement in the claimed readout gains. Please provide per-dataset component attribution and a mechanistic explanation for the interaction (for example, the effect of reusing source-training standardization statistics on the shifted target), or restrict the C-IHD claim to what the decompositions support.
minor comments (4)
  1. [Computational Cost; Table S12] The main text reports 1.57 FPS for Semantic Prism, while Table S12 reports 1.92 FPS for what appears to be the same three-window path; the stated 635.04 ms full-path timing implies approximately 1.57 FPS. Please reconcile the two tables or state the measurement difference.
  2. [Supplementary Material, Evaluation Metrics] The failure-ranking metrics (AUROC, AUPR, AURC) are computed on a fixed per-image budget of pixels selected by a token-seeded hash, but the budget size is never stated. Please specify the per-image pixel budget and confirm that it is identical across checkpoints, readouts, and datasets.
  3. [Table 1] The Direct Interface row reports no AUROC, AUPR, or AURC entries. Reporting the MSP-based error-ranking values for the 60.68% interface predictions would complete the readout isolation in Table 5 and quantify how much of the final ranking quality already exists at the interface.
  4. [Experimental Setup] The claim that training endpoints, interface temperature, and C-IHD coefficients were fixed without validation-set model selection is important for credibility, but the selection history of hyperparameters such as rho=0.8, the 5×5 window, the weights (1,0.5,0.2), and the curriculum step counts is not documented. Please add a short reproducibility appendix listing which hyperparameters were committed before the study and which were adjusted during development.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's interface is definitional by design, its refinement is trained on held-out benchmarks, and its readouts are fixed before evaluation.

full rationale

The claimed derivation chain is not circular. The 'observable interface' is explicitly constructed: Equation (3) defines a fixed distance-based decoder that maps rendered RGB values to a distribution, so the statement that the distribution is recoverable from the rendered image is a transparent design property rather than a prediction derived from an equivalent input. HGEA's residual formulation in Equation (2) starts from this interface and is trained with held-out validation benchmarks; the zero-initialized projection guarantees pH = pI at initialization, but the reported accuracy gains come from trained residual predictions, not from re-inserting the interface as the answer. The ablation in Table 3 uses matched-capacity, multi-seed controls, and the supplementary UC-Head control (Table S5) explicitly tests the residual-anchoring mechanism, which is the appropriate non-circular way to interrogate a design hypothesis. C-IHD is a fixed readout whose normalization statistics are estimated once from source training predictions and then applied unchanged to validation sets; the AUPR comparisons are evaluated on the same fixed pH, so the readout is not fitted to the metric it claims to improve. References to pix2pix-Turbo, DDPS, GSS, and other baselines are external prior work, not a self-citation chain that supplies the central load-bearing premise. The metric consistency check in the supplement re-derives the reported 60.68% and 72.07% from the same prediction files, which is an implementation audit rather than a circular derivation. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no known result is repackaged under new coordinates as a derivation.

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

The paper is an empirical machine-learning systems paper, so its central claim rests on trained components and design constants rather than mathematical axioms. The most consequential free choices are the codebook prototypes, interface temperature, generator loss weighting, and C-IHD coefficients.

free parameters (5)
  • Interface temperature tau_I = 900 (8-bit RGB squared-distance units)
    Sets the scale of the softmax over prototype distances in Eq. (3) and Eq. (S3); controls p_I confidence and ECE; the authors fixed it without validation selection, but it is a hand-chosen design constant.
  • Class-color codebook prototypes C={c_k} = 19 RGB triplets in Table S2
    Built by greedy max-min plus confusion-aware class assignment; defines the observable interface and Voronoi decoding. Choosing different prototypes changes direct-interface mIoU and boundary recovery (Table S3).
  • Generator loss weights and temperatures = 0.5 L_rgb, 3.0 L_proto, 1.5 L_margin, 0.3 L_boundary, tau_g=0.03, margin m=0.02
    Supervision terms used to train the one-step generator; not tuned on validation, but determine whether rendered colors land near prototypes and hence how useful the observable interface is.
  • C-IHD readout weights and rho = w=(1,0.5,0.2), rho=0.8
    Manual coefficients in Eq. (12); fixed before evaluation and sensitivity-gridded, but still chosen coefficients rather than derived values.
  • C-IHD normalization statistics mu_tr, sigma_tr = mu=(0.041874,0.042058,0.051223); sigma=(0.115402,0.106081,0.173971)
    Estimated once from source-training predictions and used to standardize the readout. Not fitted to validation, but derived from the same model family and therefore carry training-set information.
assumptions (5)
  • ad hoc to paper Squared Euclidean distance in RGB space is a valid proxy for semantic class likelihood.
    The observable interface assumes that closeness to a class prototype in 8-bit RGB space is meaningful; this is a design choice not derived from data.
  • domain assumption The frozen one-step generator produces semantic RGB images whose errors are concentrated at boundaries and thin structures, and are correctable by fine-grained features.
    The refinement story depends on this failure-mode distribution; Table S7 supports it on Cityscapes, but it is assumed for BDD100K and ACDC.
  • domain assumption VAE encoder levels with 128, 256, and 512 channels provide aligned multi-scale semantic evidence.
    HGEA design assumes these features are useful; supported only indirectly by the ML-HGEA versus CM-Flat ablation.
  • domain assumption Dataset annotations and splits are ground truth.
    All metrics use Cityscapes, BDD100K, and ACDC labels as ground truth, and ACDC annotations are used only for evaluation.
  • ad hoc to paper No validation-based model selection occurred.
    The authors state that endpoints, temperatures, and C-IHD coefficients are fixed without validation-set model selection; external readers cannot verify this from the paper alone.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative Semantic Segmentation via an Observable Semantic-Image Interface and Hierarchical Generator Evidence Alignment." pith.science (2026). https://pith.science/paper/J6XA6H5K

@misc{pith2026260811537,
  author       = {Pith},
  title        = {Pith review of: Generative Semantic Segmentation via an Observable Semantic-Image Interface and Hierarchical Generator Evidence Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6XA6H5K}},
  note         = {Machine review of arXiv:2608.11537}
}
read the original abstract

Generative semantic segmentation exposes structured predictions as images, but direct color decoding is susceptible to color drift and boundary mixing, whereas latent-feature decoders that predict a separate output distribution may relegate the rendered image to an intermediate visualization. We present Semantic Prism, a conditional semantic-image generation-and-refinement framework with deterministic inference. A diffusion-distilled one-step generator renders a semantic RGB image; per-pixel distances from the rendered colors to a fixed class-color codebook define an explicit probabilistic interface. Hierarchical Generator Evidence Alignment spatially aligns multi-level generator features and uses a zero-initialized output projection to predict an additive residual in the interface logit space, retaining the image-defined interface as the reference for the final distribution. The interface and refined distributions further enable Contextual Interface--Hierarchy Disagreement (C-IHD), a fixed readout for ranking remaining pixel errors without an auxiliary predictor or additional forward pass. On the 500-image Cityscapes validation set, Semantic Prism achieves 72.07% mean intersection over union, 11.39 mIoU points above direct-interface decoding, with 0.41% expected calibration error. Matched-capacity ablations over three seeds support the benefit of jointly aligned multi-level evidence. A separately trained model attains 62.22% mIoU on BDD100K, while the Cityscapes-trained model reaches 46.89\% mIoU under source-frozen transfer to the Adverse Conditions Dataset with Correspondences, without target-domain adaptation. Across all three datasets, C-IHD consistently improves the area under the precision--recall curve for pixel-error ranking over maximum softmax probability on the same segmentation predictions; on ACDC, it raises AUPR from 0.6580 to 0.7557.

Figures

Figures reproduced from arXiv: 2608.11537 by the authors.

Figure 1
Figure 1. Overview of Semantic Prism. A one-step pix2pix-Turbo generator renders a semantic RGB image and exposes multi [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Observable-to-final Cityscapes predictions. Fixed-codebook decoding converts the generated semantic RGB image [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Source-frozen ACDC transfer evaluation. Rows show scenarios of fog, night, rain, and snow; columns show the input, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 42 canonical work pages

  1. [1]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages =

    Marius Cordts and Mohamed Omran and Sebastian Ramos and Timo Rehfeld and Markus Enzweiler and Rodrigo Benenson and Uwe Franke and Stefan Roth and Bernt Schiele , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages =

    work page

  2. [2]

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

    Christos Sakaridis and Dengxin Dai and Luc Van Gool , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

    work page

  3. [3]

    Alvarez and Ping Luo , title =

    Enze Xie and Wenhai Wang and Zhiding Yu and Anima Anandkumar and Jose M. Alvarez and Ping Luo , title =. Advances in Neural Information Processing Systems , volume =

    work page

  4. [4]

    Schwing and Alexander Kirillov and Rohit Girdhar , title =

    Bowen Cheng and Ishan Misra and Alexander G. Schwing and Alexander Kirillov and Rohit Girdhar , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    work page

  5. [5]

    Proceedings of the European Conference on Computer Vision , pages =

    Liang-Chieh Chen and Yukun Zhu and George Papandreou and Florian Schroff and Hartwig Adam , title =. Proceedings of the European Conference on Computer Vision , pages =

    work page

  6. [6]

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

    Jun Fu and Jing Liu and Haijie Tian and Yong Li and Yongjun Bao and Zhiwei Fang and Hanqing Lu , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    work page

  7. [7]

    arXiv preprint arXiv:1904.04514 , year =

    Ke Sun and Yang Zhao and Borui Jiang and Tianheng Cheng and Bin Xiao and Dong Liu and Yadong Mu and Xinggang Wang and Wenyu Liu and Jingdong Wang , title =. arXiv preprint arXiv:1904.04514 , year =

    arXiv 1904

  8. [8]

    Kim and Seungryong Kim and Jaegul Choo , title =

    Sungha Choi and Sanghun Jung and Huiwon Yun and Joanne T. Kim and Seungryong Kim and Jaegul Choo , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    work page

Show all 47 references
  1. [9]

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

    Jiaqi Chen and Jiachen Lu and Xiatian Zhu and Li Zhang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  2. [10]

    arXiv preprint arXiv:2306.01721 , year =

    Zeqiang Lai and Yuchen Duan and Jifeng Dai and Ziheng Li and Ying Fu and Hongsheng Li and Yu Qiao and Wenhai Wang , title =. arXiv preprint arXiv:2306.01721 , year =

    arXiv

  3. [11]

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

    Yuanfeng Ji and Zhe Chen and Enze Xie and Lanqing Hong and Xihui Liu and Zhaoqiang Liu and Tong Lu and Zhenguo Li and Ping Luo , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  4. [12]

    Advances in Neural Information Processing Systems , volume =

    Chaoyang Wang and Xiangtai Li and Lu Qi and Henghui Ding and Yunhai Tong and Ming-Hsuan Yang , title =. Advances in Neural Information Processing Systems , volume =

  5. [13]

    arXiv preprint arXiv:2505.19422 , year =

    Jiru Deng and Tengjin Weng and Tianyu Yang and Wenhan Luo and Zhiheng Li and Wenhao Jiang , title =. arXiv preprint arXiv:2505.19422 , year =

    arXiv

  6. [14]

    arXiv preprint arXiv:2311.17042 , year =

    Axel Sauer and Dominik Lorenz and Andreas Blattmann and Robin Rombach , title =. arXiv preprint arXiv:2311.17042 , year =

    arXiv

  7. [15]

    Narasimhan and Jun

    Gaurav Parmar and Taesung Park and Srinivasa G. Narasimhan and Jun. One-Step Image Translation with Text-to-Image Models , journal =

  8. [16]

    Proceedings of the 37th International Conference on Machine Learning , series =

    Pang Wei Koh and Thao Nguyen and Yew Siang Tang and Stephen Mussmann and Emma Pierson and Been Kim and Percy Liang , title =. Proceedings of the 37th International Conference on Machine Learning , series =

  9. [17]

    International Journal of Computer Vision , volume =

    Max Losch and Mario Fritz and Bernt Schiele , title =. International Journal of Computer Vision , volume =

  10. [18]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =

    Miko. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =

  11. [19]

    Weinberger , title =

    Chuan Guo and Geoff Pleiss and Yu Sun and Kilian Q. Weinberger , title =. Proceedings of the 34th International Conference on Machine Learning , series =

  12. [20]

    International Conference on Learning Representations , year =

    Dan Hendrycks and Kevin Gimpel , title =. International Conference on Learning Representations , year =

  13. [21]

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

    Dongdong Wang and Boqing Gong and Liqiang Wang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  14. [22]

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

    Zhipeng Ding and Xu Han and Peirong Liu and Marc Niethammer , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  15. [23]

    Confidence Calibration for Object Detection and Segmentation , journal =

    Fabian K. Confidence Calibration for Object Detection and Segmentation , journal =

  16. [24]

    Prediction Error Meta Classification in Semantic Segmentation: Detection via Aggregated Dispersion Measures of Softmax Probabilities , booktitle =

    Matthias Rottmann and Pascal Colling and Thomas-Paul Hack and Robin Chan and Fabian H. Prediction Error Meta Classification in Semantic Segmentation: Detection via Aggregated Dispersion Measures of Softmax Probabilities , booktitle =

  17. [25]

    Advances in Neural Information Processing Systems , volume =

    Yonatan Geifman and Ran El-Yaniv , title =. Advances in Neural Information Processing Systems , volume =

  18. [26]

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

    Yunxiang Fu and Meng Lou and Yizhou Yu , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  19. [27]

    Proceedings of the 42nd International Conference on Machine Learning , series =

    Fan Li and Xuan Wang and Min Qi and Zhaoxiang Zhang and Yuelei Xu , title =. Proceedings of the 42nd International Conference on Machine Learning , series =

  20. [28]

    Advances in Neural Information Processing Systems , volume =

    Mengyu Wang and Henghui Ding and Jun Hao Liew and Jiajun Liu and Yao Zhao and Yunchao Wei , title =. Advances in Neural Information Processing Systems , volume =

  21. [29]

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

    Xinlong Wang and Xiaosong Zhang and Yue Cao and Wen Wang and Chunhua Shen and Tiejun Huang , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  22. [30]

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

    Lu Qi and Lehan Yang and Weidong Guo and Yu Xu and Bo Du and Varun Jampani and Ming-Hsuan Yang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  23. [31]

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

    Zigang Geng and Binxin Yang and Tiankai Hang and Chen Li and Shuyang Gu and Ting Zhang and Jianmin Bao and Zheng Zhang and Houqiang Li and Han Hu and Dong Chen and Baining Guo , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  24. [32]

    Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops , pages =

    Masud Ahmed and Zahid Hasan and Syed Arefinul Haque and Abu-Zaher Faridee and Sanjay Purushotham and Suya You and Nirmalya Roy , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops , pages =

  25. [33]

    Advances in Neural Information Processing Systems , volume =

    Canyu Zhao and Yanlong Sun and Mingyu Liu and Huanyi Zheng and Muzhi Zhu and Zhiyue Zhao and Hao Chen and Tong He and Chunhua Shen , title =. Advances in Neural Information Processing Systems , volume =

  26. [34]

    Valentin Gabeur and Shangbang Long and Songyou Peng and Paul Voigtlaender and Shuyang Sun and Yanan Bao and Karen Truong and Zhicheng Wang and Wenlei Zhou and Jonathan T. Barron and Kyle Genova and Nithish Kannen and Sherry Ben and Yandong Li and Mandy Guo and Suhas Yogin and ...

    arXiv

  27. [35]

    Francisco Caetano and Christiaan Viviers and Peter H. N. de With and Fons van der Sommen , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =

  28. [36]

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

    Yuhuan Yang and Xianwei Zhuang and Yuxuan Cai and Chaofan Ma and Shuai Bai and Jiangchao Yao and Ya Zhang and Junyang Lin and Yanfeng Wang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  29. [37]

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

    Wenliang Zhao and Yongming Rao and Zuyan Liu and Benlin Liu and Jie Zhou and Jiwen Lu , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  30. [38]

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

    Jiarui Xu and Sifei Liu and Arash Vahdat and Wonmin Byeon and Xiaolong Wang and Shalini De Mello , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  31. [39]

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

    Benyuan Meng and Qianqian Xu and Zitai Wang and Xiaochun Cao and Longtao Huang and Qingming Huang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  32. [40]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =

    Hugo Porta and Emanuele Dalsasso and Diego Marcos and Devis Tuia , title =. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =

  33. [41]

    IEEE Robotics and Automation Letters , volume =

    Quazi Marufur Rahman and Niko S. IEEE Robotics and Automation Letters , volume =

  34. [42]

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

    Donghyeon Kwon and Suha Kwak , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  35. [43]

    Vanessa Emanuela Guarino and Claudia Winklmayr and Jannik Franzen and Josef Lorenz Rumberger and Manuel Pfeuffer and Sonja Greven and Klaus Maier-Hein and Dagmar Kainmueller and Christoph Karg and Carsten T. L. Better than Average: Spatially-Aware Aggregation of Segmentation U...

  36. [44]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

    Fisher Yu and Haofeng Chen and Xin Wang and Wenqi Xian and Yingying Chen and Fangchen Liu and Vashisht Madhavan and Trevor Darrell , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

  37. [45]

    arXiv preprint arXiv:2406.03702 , year =

    Zilu Guo and Liuyang Bian and Xuan Huang and Hu Wei and Jingyu Li and Huasheng Ni , title =. arXiv preprint arXiv:2406.03702 , year =

    arXiv

  38. [46]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

    John Lambert and Zhuang Liu and Ozan Sener and James Hays and Vladlen Koltun , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

  39. [47]

    Neighbor-Aware Calibration of Segmentation Networks with Penalty-Based Constraints , journal =

    Balamurali Murugesan and Sukesh Adiga Vasudeva and Bingyuan Liu and Herv. Neighbor-Aware Calibration of Segmentation Networks with Penalty-Based Constraints , journal =. 2025 , doi =

    2025

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.