Pith. sign in

REVIEW 3 major objections 5 minor 34 references

GS: Generative Segmentation via Label Diffusion

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that language-driven segmentation is best formulated as a generative problem: directly generating the mask from noise with a conditioned label-diffusion model, and reports state-of-the-art Average Recall of 69.7 on…

desk verdict Direct mask diffusion for language-driven segmentation is a real idea worth refereeing, but the SOTA claim is contaminated by test-set hyperparameter selection and missing code. read the letter →

arxiv 2508.20020 v1 pith:ITQGZYVA submitted 2025-08-27 cs.CV

Yuhao Chen , Shubin Chen , Liang Lin , Guangrun Wang This is my paper · ORCID
classification cs.CV
keywords labeldiffusiongenerativesegmentationpanopticnarrativegroundinglanguage-drivenclassifier-freeguidancelatentmodelstext-to-maskgeneration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes GS, a framework that treats language-driven segmentation as a generative problem: instead of classifying pixels or extracting attention maps from an image diffusion model, a U-Net denoises a segmentation mask from Gaussian noise, conditioned on the input image and a noun-phrase description. The authors claim that this 'label diffusion' reversal is not just viable but superior, reporting 69.7 Average Recall on Panoptic Narrative Grounding against 67.1 for the best prior method, EIPA+MLMA, without panoptic segmentation pretraining. If correct, the result would establish that dense structured prediction can be achieved by modeling the label distribution itself with a conditioned diffusion process, with the label space rather than the image as the primary generative target. The paper also shows that performance improves with larger generation resolution and more DDIM sampling steps, and that classifier-free guidance strengthens text control.

What carries the argument

The load-bearing object is the conditional label-space diffusion process: a forward noising schedule applied to a one-channel latent mask and a reverse denoising U-Net $\Phi$ whose input is the channel-wise concatenation of the noisy label $x_t$ and the image latent $\varepsilon(I)$. This machinery makes the mask itself the generative variable and lets the same VAE geometry that encodes images carry the label signal, with text injected through cross-attention and multi-scale adapter tokens. Classifier-free guidance combines conditional and unconditional noise predictions at inference to sharpen textual control.

What would settle it

Encode ground-truth PNG masks with the SDXL encoder and immediately decode them, then measure pixel-wise IoU against the originals; if autoencoding visibly distorts the masks, the reported Average Recall is not measuring faithful segmentation and the central claim weakens. A complementary check is to replace VAE decoding of the denoised latent with direct bilinear upsampling of a soft prediction and quantify the Average Recall gap, separating the generative signal from the decode step.

Watch

Extended reading notes

Core claim

GS directly generates segmentation masks from noise in the latent space of a variational autoencoder, conditioned jointly on the image and on text. The clean signal $x_0$ is a one-channel latent encoding of the ground-truth mask, and the noisy label map is concatenated channel-wise with the four-channel image latent and fed to a Stable Diffusion XL denoising U-Net; text enters twice, through global CLIP embeddings with cross-attention and through token-level adapter features injected into self-attention. Training uses the standard noise-prediction objective with conditional dropout, and inference applies classifier-free guidance. The paper reports that this end-to-end generative pipeline achieves state-of-the-art Average Recall on the PNG benchmark, 69.7 overall, with leading scores on stuff (76.5), singulars (71.3), and plurals (63.7), while the things subcategory lags and is attributed to limited latent resolution.

Load-bearing premise

The load-bearing premise is that a binary segmentation mask can be faithfully represented as a one-channel clean signal in the Stable Diffusion XL variational autoencoder's latent space, so that decoding the denoised latent with the VAE decoder yields a usable mask; the paper adopts this without validating the reconstruction fidelity.

Editorial extensions

If this is right

  • Language-driven segmentation can be cast as conditional mask synthesis, so neither panoptic pretraining on COCO nor handcrafted post-processing is required to reach competitive performance.
  • Resolution is a first-order factor: in the paper's ablation, raising generation size from 512 to 1024 lifts overall Average Recall from 32.7 to 71.0.
  • More DDIM sampling steps give finer masks (50 steps: 71.0; 20 steps: 68.7), so inference cost can be traded against accuracy.
  • Classifier-free guidance with text dropout transfers from image generation to mask generation and is needed for strong conditioning.
  • Because training is end-to-end on masks, the formulation can in principle apply to other language-driven segmentation tasks beyond the PNG benchmark.

Reading between the lines

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

  • Beyond the reported experiments, the same label-diffusion formulation could be pointed at other dense prediction targets such as depth, edges, or saliency by swapping the one-channel target while keeping image-text conditioning; the paper remarks on generality but does not test it.
  • The unvalidated use of the SDXL VAE for masks is independently testable: if mask autoencoding is lossy, some of the measured gain may come from the bilinear post-processing rather than from the generative formulation itself.
  • The supplement's failure cases, such as segmenting nearly the whole scene as 'bench', look like hallucination modes of generative sampling; stronger spatial conditioning or consistency terms may be needed before the approach is reliable for small or ambiguous phrases.
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 / 5 minor

Summary. The paper proposes GS (Generative Segmentation), a framework that treats language-driven segmentation as a generative task by diffusing directly in the label space. Given an input image and a noun phrase, a U-Net conditioned on both the image latent (via channel-wise concatenation) and text features (via token-level adapter injection and cross-attention) denoises a latent representation of the segmentation mask. Training uses a standard noise-prediction objective with conditional dropout for classifier-free guidance; inference applies CFG and then decodes/interpolates the predicted latent. The method is evaluated on Panoptic Narrative Grounding (PNG), where Table 2 reports an overall Average Recall of 69.7, claimed to be a new state of the art over the previous best EIPA+MLMA at 67.1. The paper also includes ablations on image size, DDIM steps, and guidance scale, plus qualitative visualizations.

Significance. If the reported numbers withstand a clean evaluation, the paper would make a useful conceptual contribution: it reframes segmentation as a primary generative objective rather than an auxiliary use of image diffusion, and it demonstrates end-to-end training without panoptic segmentation pretraining. The dual-branch conditioning mechanism (spatial concatenation plus token-level text injection) is plausible, and the idea of generating masks directly from noise is interesting and clearly presented. The main strength is the novelty of the formulation; the main weakness is that the central empirical claim is not yet supported by a valid evaluation protocol, and one load-bearing architectural assumption about the VAE latent space is not validated. The paper does not provide machine-checked proofs or reproducible code within the manuscript, so the reported performance numbers are the only support for the SOTA claim.

major comments (3)
  1. [§4.4, Tables 3–5] The ablation study is run on a subset of the test set comprising approximately 7% of its size, and the settings selected there (1024×1024 image size, 50 DDIM steps, guidance 7.5) are exactly the settings used for the headline result in Table 2. Selecting hyperparameters on the test set invalidates the 69.7 overall AR as an unbiased estimate of test performance, and the 2.6-point margin over EIPA+MLMA is within the range of plausible selection effects: Table 3 alone shows a 6.4-point drop when moving from 1024 to 768 resolution, and a 38.3-point drop from 1024 to 512. Please re-select hyperparameters on a validation split (or pre-register a fixed configuration before touching the test set), report the resulting test-set AR, and if the margin changes materially, temper the SOTA claim accordingly.
  2. [§3.3, Eq. (2), §3.5, Fig. 4] The clean signal x0 is assumed to be a 1-channel segmentation mask in the latent space of the SDXL VAE, which was trained on RGB images with a 4-channel latent. No experiment or reconstruction analysis demonstrates that binary masks are faithfully representable in this latent geometry. The paper itself notes in Fig. 4 that VAE decoding 'introduces undesired artifacts' and that bilinear interpolation of the latent is used for the final mask, which conflicts with §3.5's statement that x0 is decoded and then resized. Please either provide round-trip reconstruction statistics for masks through the SDXL VAE, or clearly state that final masks are obtained by interpolating the denoised latent without VAE decoding, and update the inference description in §3.5 accordingly.
  3. [§4 and Appendix A] The supplementary text says that source code and deployment instructions are provided, but the manuscript as submitted contains no code, checkpoints, or explicit definition of which split is used for the reported test numbers. Given that the central claim is an empirical performance number and the evaluation protocol uses test-data tuning, release of code and a precise statement of the validation/test split are necessary for independent verification. Please make the code available and describe the split and evaluation pipeline in enough detail that the reported AR can be recomputed.
minor comments (5)
  1. [§4.5] The heading 'Appliable Efficiency' contains a typo; it should be 'Applicable Efficiency' or simply 'Efficiency'.
  2. [§3.5 and Fig. 4] The relationship between the predicted latent and the final binary mask should be specified precisely: the text mentions decoding x0, while Fig. 4 compares bilinear interpolation, nearest-neighbor interpolation, and VAE decoding, with the caption indicating that bilinear interpolation is preferred. Please clarify which operation is actually used in the reported quantitative results.
  3. [§4.1] The dataset description says there are 133,103 training images and 8,533 validation images, but the paper repeatedly refers to a 'test set' and says images in the figures 'belong to the test set.' Please state explicitly whether the official PNG validation split is used as the test set, and whether any separate held-out test split exists.
  4. [Table 2 and §4.3] The phrase 'our method without pretraining outperforms other pretrained counterparts' is potentially misleading because the model is initialized from a pretrained SDXL inpainting backbone; what is meant is likely 'without panoptic segmentation pretraining on COCO.' Please clarify this in the text and table caption.
  5. [§4.3 and Table 3] The claim that higher resolution will 'substantially enhance' the things metric is speculative, since only one resolution (1024×1024) is tested as the maximum and the trend could saturate; please soften this statement or add experiments at a larger resolution.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: GS's headline result is an empirical benchmark score from a standard conditional diffusion pipeline; test-set hyperparameter tuning is an evaluation concern, not a derivation-level circularity.

full rationale

The paper's derivation chain is self-contained: Equations (1)-(11) define a standard conditional denoising diffusion process in which x0 is the latent segmentation mask, the image latent and text embeddings are conditioning inputs, and the training loss is the usual noise-prediction objective. The reported state-of-the-art figure (69.7 Average Recall in Table 2) is a measured benchmark outcome, not a quantity obtained by substituting fitted constants back into the defining equations. No load-bearing step reduces to its own input by construction, and no fitted parameter is renamed as a prediction. The only author self-citation (Wang et al. 2022) appears in related work as an example of generative representations and is not used to justify GS's architecture or results. Section 4.4 does reveal that the final hyperparameters (image size 1024, 50 DDIM steps, guidance scale 7.5) were selected on a roughly 7% slice of the test set; this is a legitimate evaluation-integrity concern that may bias the reported margin, but it is selection bias rather than circularity because the full-test score is not algebraically forced by the subset results and no parameter fitted on the subset is presented as an independent prediction. The paper's unvalidated assumption that binary masks are faithfully representable as 1-channel SDXL latents is a correctness risk, but it does not make the derivation circular.

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

The central claim rests on the transferability of a pretrained image diffusion backbone (SDXL inpainting) and CLIP text encoder to the task of mask generation. The most fragile assumption is that a single-channel binary mask can be treated as a latent image in the SDXL VAE space. Hyperparameters such as guidance scale, sampling steps, and image size were selected using a portion of the test set, so the reported result is partly tuned to the evaluation data. No new physical entities are introduced.

free parameters (5)
  • Classifier-free guidance scale w = 7.5
    Chosen by ablations on a 7% subset of the test set (Table 5); directly affects the trade-off between text fidelity and mask quality.
  • DDIM sampling steps = 50
    Ablated on the test subset (Table 4); 50 steps used for final results.
  • Generated image size = 1024x1024
    Ablated on the test subset (Table 3); larger size improves AR, particularly for 'things' and 'singulars'.
  • Learning rate = 1e-7
    Reported in Sec 4.2; very low, indicating a fine-tuning regime on the frozen SDXL backbone.
  • Conditioning dropout probability p_drop = not reported
    Required for classifier-free guidance (Sec 3.5), but its value is never given in the paper.
assumptions (5)
  • domain assumption PNG dataset annotations (COCO panoptic masks + Localized Narratives captions) are correct ground truth.
    All training and evaluation rely on these labels as ground truth (Sec 4.1).
  • ad hoc to paper A binary mask can be represented as a 1-channel map in the SDXL VAE latent space and decoded back to a valid mask.
    Introduced in Sec 3.3 (Eq. 2, Fig 2) without justification; the VAE was trained on RGB images with a 4-channel latent.
  • domain assumption CLIP text embeddings of noun phrases contain sufficient semantic information for grounding when injected via adapters and cross-attention.
    The method assumes text conditioning through CLIP is adequate for the grounding task (Sec 3.4).
  • domain assumption The DDPM noise schedule and Gaussian noise assumptions apply to segmentation masks.
    Masks are categorical but treated as continuous values in the noising process (Sec 3.2, Eq. 1).
  • domain assumption The SDXL inpainting U-Net can be adapted to accept the concatenated image latent and noisy mask latent with minimal fine-tuning.
    The paper uses SDXL inpainting as the backbone (Sec 4.2) but does not describe how the input channel count is adjusted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GS: Generative Segmentation via Label Diffusion." pith.science (2026). https://pith.science/paper/ITQGZYVA

@misc{pith2026250820020,
  author       = {Pith},
  title        = {Pith review of: GS: Generative Segmentation via Label Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ITQGZYVA}},
  note         = {Machine review of arXiv:2508.20020}
}
read the original abstract

Language-driven image segmentation is a fundamental task in vision-language understanding, requiring models to segment regions of an image corresponding to natural language expressions. Traditional methods approach this as a discriminative problem, assigning each pixel to foreground or background based on semantic alignment. Recently, diffusion models have been introduced to this domain, but existing approaches remain image-centric: they either (i) use image diffusion models as visual feature extractors, (ii) synthesize segmentation data via image generation to train discriminative models, or (iii) perform diffusion inversion to extract attention cues from pre-trained image diffusion models-thereby treating segmentation as an auxiliary process. In this paper, we propose GS (Generative Segmentation), a novel framework that formulates segmentation itself as a generative task via label diffusion. Instead of generating images conditioned on label maps and text, GS reverses the generative process: it directly generates segmentation masks from noise, conditioned on both the input image and the accompanying language description. This paradigm makes label generation the primary modeling target, enabling end-to-end training with explicit control over spatial and semantic fidelity. To demonstrate the effectiveness of our approach, we evaluate GS on Panoptic Narrative Grounding (PNG), a representative and challenging benchmark for multimodal segmentation that requires panoptic-level reasoning guided by narrative captions. Experimental results show that GS significantly outperforms existing discriminative and diffusion-based methods, setting a new state-of-the-art for language-driven segmentation.

Figures

Figures reproduced from arXiv: 2508.20020 by the authors.

Figure 1
Figure 1. Comparison of GS with prior segmentation paradigms. (a) Discriminative models encode image and text into feature space and predict pixel-wise segmentation; this includes both traditional methods and diffusion-assisted approaches that rely on external discriminators. (b) Diffusion inversion methods extract attention maps from pre-trained image diffusion models for zero-shot grounding. (c) Our proposed GS directly per… view at source ↗
Figure 2
Figure 2. provides an overview of our framework. We evaluate GS on the Panoptic Narrative Ground￾ing (PNG) task, a comprehensive benchmark involving language-conditioned segmentation. However, GS is general and applicable to a wide range of language-driven segmen￾tation settings. 3.1 Notation We summarize the key notations used throughout this paper in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization of GS on representational samples on the PNG dataset. All visualized images belong to the test set. Image Groundtruth Bilinear(Ours) Nearest VAE w/o cfg [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparison with different post-processing techniques on final mask. Notably, our bilinear interpolation method with classifier-free guidance (cfg) achieves superior results compared to gt. Furthermore, naive linear interpolation causes serrated edges, VAE decoding intr…
Figure 5
Figure 5. Figure 5: Time-series Visualization of Images and Ground Truth for GS Samples on the PNG Dataset. All visualized images belong to the test set. The figures are divided into two parts (a) and (b). Each part contains “Image” and “GT” columns, with the time steps t=0, t=279, t=519,…
Figure 6
Figure 6. Figure 6: Visualization of Images and Ground Truth for Correct GS Samples on the PNG Dataset test split [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Visualization of Images and Ground Truth for Failed GS Samples on the PNG Dataset test split [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 19 canonical work pages

  1. [1]

    Baranchuk, D.; Rubachev, I.; Voynov, A.; Khrulkov, V.; and Babenko, A. 2021. Label-efficient semantic segmentation with diffusion models. arXiv preprint arXiv:2112.03126

    arXiv 2021

  2. [2]

    Choi, Y.; Kwak, S.; Lee, K.; Choi, H.; and Shin, J. 2024. Improving diffusion models for authentic virtual try-on in the wild. In European Conference on Computer Vision, 206--235. Springer

    work page 2024

  3. [3]

    Dhariwal, P.; and Nichol, A. 2021. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34: 8780--8794

    2021

  4. [4]

    Ding, Z.; Ding, Z.-h.; Hui, T.; Huang, J.; Wei, X.; Wei, X.; and Liu, S. 2022. Ppmn: Pixel-phrase matching network for one-stage panoptic narrative grounding. In Proceedings of the 30th ACM International Conference on Multimedia, 5537--5546

    work page 2022

  5. [5]

    Gonz \'a lez, C.; Ayobi, N.; Hern \'a ndez, I.; Hern \'a ndez, J.; Pont-Tuset, J.; and Arbel \'a ez, P. 2021. Panoptic narrative grounding. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 1364--1373

    work page 2021

  6. [6]

    Gonz \'a lez, C.; Ayobi, N.; Hern \'a ndez, I.; Pont-Tuset, J.; and Arbel \'a ez, P. 2023. Piglet: Pixel-level grounding of language expressions with transformers. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(10): 12206--12221

    work page 2023

  7. [7]

    Hinton, G. E. 2007. To recognize shapes, first learn to generate images. Progress in brain research, 165: 535--547

    work page 2007

  8. [8]

    Ho, J.; Jain, A.; and Abbeel, P. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 6840--6851

    2020

Show all 34 references
  1. [9]

    Ho, J.; and Salimans, T. 2022. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598

    2022 arXiv

  2. [10]

    Hui, T.; Ding, Z.; Huang, J.; Wei, X.; Wei, X.; Dai, J.; Han, J.; and Liu, S. 2023. Enriching phrases with coupled pixel and object contexts for panoptic narrative grounding. arXiv preprint arXiv:2311.01091

    2023 arXiv

  3. [11]

    C.; Lo, W.-Y.; et al

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A. C.; Lo, W.-Y.; et al. 2023. Segment anything. In Proceedings of the IEEE/CVF international conference on computer vision, 4015--4026

    2023

  4. [12]

    C.; Prabhudesai, M.; Duggal, S.; Brown, E.; and Pathak, D

    Li, A. C.; Prabhudesai, M.; Duggal, S.; Brown, E.; and Pathak, D. 2023 a . Your diffusion model is secretly a zero-shot classifier. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2206--2217

    2023

  5. [13]

    Li, H.; Hui, T.; Ding, Z.; Zhang, J.; Ma, B.; Wei, X.; Han, J.; and Liu, S. 2024. Dynamic prompting of frozen text-to-image diffusion models for panoptic narrative grounding. In Proceedings of the 32nd ACM International Conference on Multimedia, 9485--9494

    2024

  6. [14]

    Li, T.; Chang, H.; Mishra, S.; Zhang, H.; Katabi, D.; and Krishnan, D. 2023 b . Mage: Masked generative encoder to unify representation learning and image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2142--2152

    2023

  7. [15]

    Li, Z.; Zhou, Q.; Zhang, X.; Zhang, Y.; Wang, Y.; and Xie, W. 2023 c . Open-vocabulary object segmentation with diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 7667--7676

    2023

  8. [16]

    Lin, Y.; Jin, X.-B.; Wang, Q.; and Huang, K. 2023. Context does matter: end-to-end panoptic narrative grounding with deformable attention refined matching network. In 2023 IEEE International Conference on Data Mining (ICDM), 1163--1168. IEEE

    2023

  9. [17]

    Liu, X.; Huang, S.; Kang, Y.; Chen, H.; and Wang, D. 2024. Vgdiffzero: Text-to-image diffusion models can be zero-shot visual grounders. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2765--2769. IEEE

    2024

  10. [18]

    Luo, G.; Zhou, Y.; Sun, X.; Cao, L.; Wu, C.; Deng, C.; and Ji, R. 2020. Multi-task collaborative network for joint referring expression comprehension and segmentation. In Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition, 10034--10043

    2020

  11. [19]

    Ma, C.; Yang, Y.; Ju, C.; Zhang, F.; Liu, J.; Wang, Y.; Zhang, Y.; and Wang, Y. 2023. Diffusionseg: Adapting diffusion towards unsupervised object discovery. arXiv preprint arXiv:2303.09813

    2023 arXiv

  12. [20]

    Ng, A.; and Jordan, M. 2001. On discriminative vs. generative classifiers: A comparison of logistic regression and naive bayes. Advances in neural information processing systems, 14

    2001

  13. [21]

    Nguyen, Q.; Vu, T.; Tran, A.; and Nguyen, K. 2023. Dataset diffusion: Diffusion-based synthetic data generation for pixel-level semantic segmentation. Advances in Neural Information Processing Systems, 36: 76872--76892

    2023

  14. [22]

    Ni, M.; Zhang, Y.; Feng, K.; Li, X.; Guo, Y.; and Zuo, W. 2023. Ref-diff: Zero-shot referring image segmentation with generative models. arXiv preprint arXiv:2308.16777

    2023 arXiv

  15. [23]

    Q.; and Dhariwal, P

    Nichol, A. Q.; and Dhariwal, P. 2021. Improved denoising diffusion probabilistic models. In International conference on machine learning, 8162--8171. PMLR

    2021

  16. [24]

    Rombach, R.; Blattmann, A.; Lorenz, D.; Esser, P.; and Ommer, B. 2022. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 10684--10695

    2022

  17. [25]

    Sohl-Dickstein, J.; Weiss, E.; Maheswaranathan, N.; and Ganguli, S. 2015. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, 2256--2265. pmlr

    2015

  18. [26]

    Tian, J.; Aggarwal, L.; Colaco, A.; Kira, Z.; and Gonzalez-Franco, M. 2024. Diffuse attend and segment: Unsupervised zero-shot segmentation using stable diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 3554--3563

    2024

  19. [27]

    Wang, G.; Tang, Y.; Lin, L.; and Torr, P. H. 2022. Semantic-aware auto-encoders for self-supervised representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9664--9675

    2022

  20. [28]

    Wang, H.; Ji, J.; Guo, T.; Yang, Y.; Zhou, Y.; Sun, X.; and Ji, R. 2023 a . Nice: improving panoptic narrative detection and segmentation with cascading collaborative learning. arXiv preprint arXiv:2310.10975

    2023 arXiv

  21. [29]

    Wang, H.; Ji, J.; Zhou, Y.; Wu, Y.; and Sun, X. 2023 b . Towards real-time panoptic narrative grounding by an end-to-end grounding network. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, 2528--2536

    2023

  22. [30]

    Z.; Zhou, H.; and Shen, C

    Wu, W.; Zhao, Y.; Shou, M. Z.; Zhou, H.; and Shen, C. 2023. Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 1206--1217

    2023

  23. [31]

    Xu, J.; Liu, S.; Vahdat, A.; Byeon, W.; Wang, X.; and De Mello, S. 2023. Open-vocabulary panoptic segmentation with text-to-image diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2955--2966

    2023

  24. [32]

    Yang, D.; Dong, R.; Ji, J.; Ma, Y.; Wang, H.; Sun, X.; and Ji, R. 2024. Exploring phrase-level grounding with text-to-image diffusion model. In European Conference on Computer Vision, 161--180. Springer

    2024

  25. [33]

    Zhang, L.; Rao, A.; and Agrawala, M. 2023. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, 3836--3847

    2023

  26. [34]

    Zhu, Z.; Feng, X.; Chen, D.; Yuan, J.; Qiao, C.; and Hua, G. 2024. Exploring pre-trained text-to-video diffusion models for referring video object segmentation. In European Conference on Computer Vision, 452--469. Springer

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.