Pith. sign in

REVIEW 4 major objections 6 minor 56 references

Reconciling Semantic Controllability and Diversity for Remote Sensing Image Synthesis with Hybrid Semantic Embedding

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that injecting per-pixel geometric shape descriptors extracted from the semantic mask itself lets a GAN synthesize remote sensing images that are simultaneously faithful to the mask, visually diverse, and good enough to…

desk verdict Useful remote sensing image synthesis work with a mechanism that needs clarification, a diversity claim that outruns the numbers, and no code release yet — still worth a careful peer review. read the letter →

arxiv 2411.14781 v1 pith:JII7XGMB submitted 2024-11-22 cs.CV

Junde Liu , Danpei Zhao , Bo Yuan , Wentao Li , Tian Li This is my paper · ORCID
classification cs.CV
keywords semanticimagesynthesisremotesensinggenerativeadversarialnetworkgeometric-informedspatialdescriptorhybridembeddingcontrollabilitydiversitydataaugmentation
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 claims that remote sensing image synthesis can simultaneously achieve high semantic controllability (synthesized images match their semantic masks) and high visual diversity (many distinct images per mask) by injecting geometric shape information extracted from the semantic masks themselves. The proposed HySEGGAN builds a geometric-informed spatial descriptor from mask contours, fuses it with the one-hot mask into a hybrid embedding, and adds a semantic refinement network that gives the generator pixel-level feedback. On GID-15 and ISPRS, the paper reports lower FID and higher mIoU and accuracy than six published baselines while retaining diversity, and shows that adding the synthesized images to training data improves downstream segmentation by 1.4 to 2.3 mIoU points. The significance is that no extra annotations, sketch maps, or paired data are needed for these gains.

What carries the argument

The load-bearing mechanism is the geometric-informed spatial descriptor (GSD), a feature-descriptor-style shape representation computed from the input semantic mask alone. For a pixel inside an object, GSD looks at contour pixels of its instance, converts their positions to polar coordinates relative to the pixel, bins them into $n_\rho \times n_\theta$ radial and angular cells ($n_\rho=6$, $n_\theta=12$), normalizes, and standardizes across the image. This per-pixel histogram describes how the object's boundary extends around each location, which is then combined with the class layout in the hybrid semantic feature modulation block to predict spatially adaptive normalization parameters. The supporting network is the semantic refinement network, a lightweight segmentation encoder-decoder that feeds cross-entropy segmentation losses back to the generator, providing pixel-level semantic feedback without burdening the discriminator.

What would settle it

Compare HySEGGAN's full GSD against a version where each descriptor is computed over the entire class boundary instead of per-instance contours, measuring FID, mIoU, and downstream d-mIoU on GID-15; if the two versions perform identically, the claimed part-level mechanism is not load-bearing, while a clear win for instance-level GSD on a dataset with true instance masks would confirm it.

Watch

Extended reading notes

Core claim

HySEGGAN reconciles semantic controllability and diversity for remote sensing scenes by treating each pixel's local geometry as a first-class conditioning signal. For every pixel, a polar-coordinate histogram of nearby contour points (the geometric-informed spatial descriptor) encodes how the instance shape extends around that location; this descriptor is concatenated with the semantic layout and used to predict per-pixel modulation parameters in every residual block. A semantic refinement network, a small segmentation encoder-decoder, then gives the generator local feedback through three cross-entropy losses, so the discriminator handles global fidelity while a separate network handles semantic consistency. The paper reports state-of-the-art FID, mIoU, and accuracy on both datasets and consistent downstream segmentation gains, attributing the results to part-level geometry plus fine-grained semantic feedback rather than global diversity or extra conditioning.

Load-bearing premise

The method's central mechanism assumes that meaningful object-instance or part-level contours can be extracted from the semantic mask, but the datasets used in the experiments contain only class-level land-cover labels without instance masks, so the geometric descriptor may be capturing class-level contour statistics instead of true part-level geometry.

Editorial extensions

If this is right

  • If the central claim holds, a single generator can produce many visually distinct, mask-faithful images for remote sensing scenes without any auxiliary input such as sketches or instance annotations.
  • Adding HySEGGAN-synthesized images to training sets would let segmentation models improve accuracy beyond what the original labeled data alone provides, on both satellite (0.8 m) and airborne (0.05 m) imagery.
  • The reported ablation results imply each component contributes: removing GSD or SRN lowers controllability metrics, while SRN trades a little LPIPS diversity for much higher mIoU.
  • The method's robustness to background classes suggests it can synthesize complex unlabeled regions, which are often excluded in remote sensing synthesis benchmarks.
  • The paper itself notes a failure mode when one class dominates the mask, such as large pavement areas, causing repeated textures and pattern collapse.

Reading between the lines

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

  • The GSD is described per instance and Algorithm 1 extracts instance indices, but both datasets provide only class-level land-cover masks without instance IDs; if the implementation uses whole-class connected regions, the claimed part-level geometry may actually reduce to class-level contour statistics, and an explicit instance-level test would settle which mechanism explains the gains.
  • A natural extension is to compute GSDs over true object instances obtained from an off-the-shelf instance segmentation model; if instance-level descriptors improve FID and mIoU over connected-component pseudo-instances, the part-level hypothesis is confirmed.
  • Because the semantic refinement network provides segmentation feedback through a learned segmentation head, the approach resembles distilling a segmentation model into the generator, so a stronger pre-trained segmenter might further improve controllability.
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

4 major / 6 minor

Summary. The paper proposes HySEGGAN, a GAN-based semantic image synthesis method for remote sensing images, with three components: a Geometric-informed Spatial Descriptor (GSD) computed from semantic masks, a Hybrid Semantic embedding Guided Network (HSGNet) that modulates features using the hybrid embedding, and a Semantic Refinement Network (SRN) that provides auxiliary segmentation-based feedback. The authors report quantitative comparisons on GID-15 and ISPRS, claiming state-of-the-art controllability (lower FID, higher mIoU and accuracy) and improved downstream segmentation when synthesized images are used for data augmentation. The paper also includes module ablations, loss ablations, hyperparameter studies, qualitative comparisons, and a discussion of failure cases.

Significance. If the reported results are reproducible, the paper makes a useful empirical contribution: it demonstrates consistent improvements in FID, mIoU, and downstream segmentation metrics over several strong baselines on two remote sensing datasets, and it proposes a descriptor-based conditioning mechanism that is relatively novel in this area. The inclusion of downstream-task evaluation is a strength, as is the attempt to decompose the contributions of GSD, HSFM, and SRN through ablation. However, the significance is tempered by the fact that no code or trained weights are released, no error bars or multiple-seed results are provided, and the core mechanism of the GSD is ambiguously specified for the class-level masks that the datasets actually provide. The central empirical claim is plausible but not yet established to the standard required for publication.

major comments (4)
  1. [Section III-B, Algorithm 1, Section IV-A1] The GSD computation assumes instance-level masks, but both datasets used in the experiments provide only class-level land-cover labels. Algorithm 1 extracts 'instance indices U <- unique(I)' and constructs a binary mask per index, yet GID-15 and ISPRS semantic masks (Section IV-A1) contain class labels, not instance IDs. If U contains class labels, then each binary mask is the union of all disconnected regions of that class, and the contour set P encodes class-level region statistics rather than the 'part-level layout' or 'irregular spatial distribution within instances' that the method motivates in Section III-B and Fig. 1. If, instead, a connected-component labeling or an instance-segmentation step is used, that step is not described in the paper or in Algorithm 1. Either way, the claimed mechanism behind the FID and mIoU improvements in Tables II–IV is not established, because the GSD could simply be acting as an additional class-conditioning signal. Please clarify the exact grouping used and, ideally, provide an ablation comparing instance-level, connected-component, and class-level contour descriptors.
  2. [Section IV-D5, Tables IX and X vs Tables II and IV] The hyperparameters n_rho, n_theta, and Gamma are selected by FID ablation on GID-15, and the same GID-15 evaluation split is then used to report the final results in Tables II and IV. This is a form of tuning on the evaluation set, and it makes the headline GID-15 numbers optimistic relative to a fully held-out protocol. The authors should either report results on a separate validation split, use a nested tuning procedure, or present multiple-seed statistics that account for the selection. The ISPRS results are less affected because the hyperparameters are transferred from GID-15, but the paper should state this explicitly.
  3. [Tables II, III, and IV] All quantitative results are reported as single numbers without error bars, confidence intervals, or multiple-seed runs. This is particularly concerning for the diversity metrics, where the differences between the proposed method and the next best baseline are small: in Table II, LPIPS is 0.189 versus 0.187 for CLADE, and mCSD is 0.028 versus 0.027 for CLADE; in Table III, mCSD is 0.044 versus 0.043 for CLADE. Given the known variance of GAN training, these differences are not statistically meaningful as reported. The authors should provide at least three independent runs and report means and standard deviations, or otherwise justify that the margins exceed run-to-run variability.
  4. [Abstract, Section IV-B1, and Section IV-B2] The abstract and text claim an 'excellent balance' between controllability and diversity, but the diversity results are suboptimal on the LPIPS and mCSD metrics in both datasets (Table II: LPIPS 0.189 and mCSD 0.028 are below SAFM and INADE; Table III: LPIPS 0.317 and mCSD 0.044 are below SAFM, CLADE, and INADE). Only mOCD is consistently optimal. Since the contribution is framed as reconciling controllability and diversity, the paper should discuss this trade-off quantitatively and avoid implying that the method is best on all diversity axes. The current wording overstates the diversity component of the contribution.
minor comments (6)
  1. [Section IV-A3, Eq. (23)] The FID formula appears to have a sign and square-root error: the trace term should be Tr(C_x + C_g − 2(C_x C_g)^{1/2}), not "Tr(Cx + Cg + 2(CxCg))1/2". Please correct the equation.
  2. [Section III-E5, Eq. (22)] The total loss writes min_G max_D L_D_adv plus additional generator losses. As written, the max_D applies only to L_D_adv, which is not the standard simultaneous min-max formulation used elsewhere in the paper. Please clarify the notation or reformulate.
  3. [Algorithm 1] The line "Compute(C_p, S, D[b num])" is underspecified: the function name and the role of "highlight pixels H_p" are not defined in the text. The reader cannot reconstruct the descriptor computation from the pseudocode alone.
  4. [Table III] The accuracy value 91.98 for the proposed method is formatted inconsistently (missing a trailing zero compared to other entries).
  5. [References] References [5] and [7] are identical (Yuan et al., "Birds of a feather flock together"). Please deduplicate and ensure all citations are used appropriately.
  6. [Section I and Section II-C] The phrase "Derived from by CSEBGAN [27]" is grammatically incomplete and should be rephrased.

Circularity Check

1 steps flagged · score 4.0 of 10

Mild evaluation circularity: the final GID-15 headline FID is the best value of a hyperparameter sweep on GID-15, not a held-out fixed configuration; no definitional circularity in the architecture.

  1. fitted input called prediction [Sec. IV-A2 (Experimental Details) and Sec. IV-D5 (Impact of Hyper-parameters), Tables IX, X and II]
    "The hyperparameters are set as nρ = 6, nθ = 12, and Γ = 80, according to the ablation study in Sec. IV-D5. ... Table IX illustrates the epoch at which semantic feedback starts to affect model performance. Table X demonstrates the influence of nθ and nρ on GSD feature representation and their impact on performance. Note that these hyperparameters are not designed for a specific dataset, although we only present the quantitative results for GID-15."

    The reported GID-15 headline FID (132.283 in Table II) is exactly the best value found by sweeping nθ, nρ (Table X: (12,6) -> 132.283) and Γ (Table IX: epoch 80 -> 132.283) on the same GID-15 validation set. The final configuration is therefore the argmin of the evaluation metric on the benchmark that is then used to claim state-of-the-art performance, with no separate held-out split for model selection. This is a fitted input presented as the method's fixed performance: the 'prediction' (SOTA FID) is statistically forced by the preceding selection procedure. It is mild rather than definitional because ISPRS and the downstream segmentation gains provide some independent evidence.

full rationale

Aside from the GID-15 hyperparameter selection, the paper's derivation chain is self-contained. GSD is a conventional shape-context-like histogram computed from the input mask; HSGNet and SRN are architectural components with standard losses (adversarial, feature-matching, perceptual, cross-entropy refinement) whose targets are the same semantic masks used as conditioning, which is the intended objective rather than a hidden circularity. The final comparisons use external DeeplabV3+ segmentation pretrained on real data, and downstream augmentation gains are measured against a source-only baseline, so those results are not equivalent to the method's own inputs. References to the authors' prior segmentation work are contextual and not load-bearing for the synthesis claims; CSEBGAN is cited for the 'extensibility' concept but is not by the current author set, and the method does not rely on a self-citation chain. One genuine concern is that Algorithm 1's 'unique(I)' on class-level GID-15/ISPRS masks means the descriptor is computed per class rather than per object instance, so the 'part-level' mechanism is not demonstrated; however, this is a reproducibility/interpretability gap, not a circular reduction of a prediction to an input. Overall circularity score 4 reflects the mild evaluation-tunnelling from selecting nρ, nθ, Γ on the same GID-15 validation set that later reports the headline FID.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central claim is an empirical engineering claim, so the ledger is dominated by tuned hyperparameters and domain assumptions rather than derived constants. The free parameters are GSD bin counts, the SRN warm-up epoch, unstated radial range values, and loss weights. The most load-bearing domain assumption is that class-level semantic masks can supply per-object geometric contours; this is not explicitly verified. GSD and SRN are new engineered modules without independent evidence outside this paper.

free parameters (5)
  • GSD radial bin count n_rho = 6
    Selected by FID ablation on GID-15 (Table X); determines the histogram dimension of the geometric descriptor.
  • GSD angular bin count n_theta = 12
    Selected by FID ablation on GID-15 (Table X); determines the histogram dimension of the geometric descriptor.
  • SRN warm-up epoch Gamma = 80
    Selected by FID ablation on GID-15 (Table IX); controls when the semantic refinement losses are activated.
  • GSD radial range r_inner and r_outer = not specified
    Called preset values in Eq. (5) but never numerically defined; directly affects descriptor construction.
  • loss weights lambda_adv, lambda_fm, lambda_perc, lambda_ref = 1, 10, 10, 1
    Fixed in Sec. IV-A2 without sensitivity analysis; the total loss in Eq. (22) depends on these weights.
assumptions (4)
  • domain assumption Semantic masks provide extractable object instances or contours that capture part-level geometry.
    Section III-B and Algorithm 1 assume instance-level contours exist, while GID-15 and ISPRS supply class-level land-cover labels without instance IDs.
  • domain assumption Shape-context-style histograms computed from mask contours are a useful geometric prior that does not leak image appearance.
    This is the core motivation of Sec. III-B; the paper provides only its own empirical results as evidence, with no external verification of the descriptor's semantic meaning.
  • domain assumption The Semantic Refinement Network trained on real images gives reliable pixel-level semantic feedback on synthesized images.
    Section III-D/E4 relies on SRN outputs for fake images being meaningful training signals; if SRN fails on out-of-distribution synthetic images, L_z_ref and L_S_ref become unreliable.
  • standard math Standard GAN training assumptions: hinge adversarial loss, spectral normalization, and TTUR stabilize training.
    Section III-E1 and Sec. IV-A2 adopt standard practices from the cited adversarial training literature.
invented entities (2)
  • Geometric-informed Spatial Descriptor (GSD)
    purpose: Encodes part-level geometry of semantic objects as a polar-coordinate histogram to guide synthesis.
    It is an engineered feature representation introduced by this paper; its benefit is supported only by the paper's own experiments, with no external falsifiable benchmark.
  • Semantic Refinement Network (SRN)
    purpose: Provides local, fine-grained semantic feedback to the generator through segmentation losses.
    A new network module introduced to improve controllability; its effectiveness is demonstrated only through internal ablations and downstream evaluations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reconciling Semantic Controllability and Diversity for Remote Sensing Image Synthesis with Hybrid Semantic Embedding." pith.science (2026). https://pith.science/paper/JII7XGMB

@misc{pith2026241114781,
  author       = {Pith},
  title        = {Pith review of: Reconciling Semantic Controllability and Diversity for Remote Sensing Image Synthesis with Hybrid Semantic Embedding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JII7XGMB}},
  note         = {Machine review of arXiv:2411.14781}
}
read the original abstract

Significant advancements have been made in semantic image synthesis in remote sensing. However, existing methods still face formidable challenges in balancing semantic controllability and diversity. In this paper, we present a Hybrid Semantic Embedding Guided Generative Adversarial Network (HySEGGAN) for controllable and efficient remote sensing image synthesis. Specifically, HySEGGAN leverages hierarchical information from a single source. Motivated by feature description, we propose a hybrid semantic Embedding method, that coordinates fine-grained local semantic layouts to characterize the geometric structure of remote sensing objects without extra information. Besides, a Semantic Refinement Network (SRN) is introduced, incorporating a novel loss function to ensure fine-grained semantic feedback. The proposed approach mitigates semantic confusion and prevents geometric pattern collapse. Experimental results indicate that the method strikes an excellent balance between semantic controllability and diversity. Furthermore, HySEGGAN significantly improves the quality of synthesized images and achieves state-of-the-art performance as a data augmentation technique across multiple datasets for downstream tasks.

Figures

Figures reproduced from arXiv: 2411.14781 by the authors.

Figure 1
Figure 1. Schematic diagram of the proposed method. (a) Traditional semantic [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The training pipeline of the hybrid semantic embedding guided GAN is illustrated as follows. The pink section represents the computation of the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Detailed structure of the HSFMResBlock used in Fig. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Illustrate the architecture of encoder layers and decoder layers in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative visualization of synthesis results for various methods on the GID-15 task. The left two columns are the semantic mask and the corresponding [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Downstream task enhanced performance with results from GID [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Robustness to the unknown class, the results are from both GID-15 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Failure examples of the proposed method, the results are from ISPRS [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 48 canonical work pages

  1. [1]

    Siedob: Semantic image editing by disentangling object and background,

    W. Luo, S. Yang, X. Zhang, and W. Zhang, “Siedob: Semantic image editing by disentangling object and background,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 1868–1878. 1

    work page 2023

  2. [2]

    Editgan: High-precision semantic image editing,

    H. Ling, K. Kreis, D. Li, S. W. Kim, A. Torralba, and S. Fidler, “Editgan: High-precision semantic image editing,” Advances in Neural Information Processing Systems, vol. 34, pp. 16 331–16 345, 2021. 1

    work page 2021

  3. [3]

    Sesame: Semantic editing of scenes by adding, manipulating or erasing objects,

    E. Ntavelis, A. Romero, I. Kastanis, L. Van Gool, and R. Timofte, “Sesame: Semantic editing of scenes by adding, manipulating or erasing objects,” in Computer Vision–ECCV 2020: 16th European Confer- ence, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXII 16 . Springer, 2020, pp. 394–411. 1

    work page 2020

  4. [4]

    Incremental learning for semantic segmentation of large-scale remote sensing data,

    O. Tasar, Y . Tarabalka, and P. Alliez, “Incremental learning for semantic segmentation of large-scale remote sensing data,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 12, pp. 3524–3537, 2019. 1

    work page 2019

  5. [6]

    A survey on continual semantic segmentation: Theory, challenge, method and application,

    B. Yuan and D. Zhao, “A survey on continual semantic segmentation: Theory, challenge, method and application,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024. 1

    work page 2024

  6. [7]

    Birds of a feather flock together: Category-divergence guidance for domain adaptive seg- mentation,

    B. Yuan, D. Zhao, S. Shao, Z. Yuan, and C. Wang, “Birds of a feather flock together: Category-divergence guidance for domain adaptive seg- mentation,” IEEE Transactions on Image Processing, vol. 31, pp. 2878– 2892, 2022. 1

    work page 2022

  7. [8]

    Inherit with distillation and evolve with contrast: Exploring class incremental semantic segmentation without exemplar memory,

    D. Zhao, B. Yuan, and Z. Shi, “Inherit with distillation and evolve with contrast: Exploring class incremental semantic segmentation without exemplar memory,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 11 932–11 947, 2023. 1

    work page 2023

  8. [9]

    Learning at a glance: Towards in- terpretable data-limited continual semantic segmentation via semantic- invariance modelling,

    B. Yuan, D. Zhao, and Z. Shi, “Learning at a glance: Towards in- terpretable data-limited continual semantic segmentation via semantic- invariance modelling,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 1

    work page 2024

Show all 56 references
  1. [10]

    Low light video enhancement using synthetic data produced with an intermediate domain mapping,

    D. Triantafyllidou, S. Moran, S. McDonagh, S. Parisot, and G. Slabaugh, “Low light video enhancement using synthetic data produced with an intermediate domain mapping,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part X...

    2020

  2. [11]

    Deep learning for image enhancement and correction in magnetic resonance imaging—state-of-the-art and challenges,

    Z. Chen, K. Pawar, M. Ekanayake, C. Pain, S. Zhong, and G. F. Egan, “Deep learning for image enhancement and correction in magnetic resonance imaging—state-of-the-art and challenges,” Journal of Digital Imaging, vol. 36, no. 1, pp. 204–230, 2023. 1

    2023

  3. [12]

    Fice: Text-conditioned fashion-image editing with guided gan inversion,

    M. Pernu ˇs, C. Fookes, V .ˇStruc, and S. Dobriˇsek, “Fice: Text-conditioned fashion-image editing with guided gan inversion,” Pattern Recognition, p. 111022, 2024. 1

    2024

  4. [13]

    Semantic image synthesis with spatially-adaptive normalization,

    T. Park, M.-Y . Liu, T.-C. Wang, and J.-Y . Zhu, “Semantic image synthesis with spatially-adaptive normalization,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 2337–2346. 1, 2, 3, 7, 8, 9

    2019

  5. [14]

    Semantic probability distribution modeling for diverse semantic image synthesis,

    Z. Tan, Q. Chu, M. Chai, D. Chen, J. Liao, Q. Liu, B. Liu, G. Hua, and N. Yu, “Semantic probability distribution modeling for diverse semantic image synthesis,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 5, pp. 6247–6264, 2023. 1, 2, 3, 7, 8, 9

    2023

  6. [15]

    Semantic image synthesis via adversarial learning,

    H. Dong, S. Yu, C. Wu, and Y . Guo, “Semantic image synthesis via adversarial learning,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 5706–5714. 1, 2

    2017

  7. [16]

    High-resolution image synthesis and semantic manipulation with condi- tional gans,

    T.-C. Wang, M.-Y . Liu, J.-Y . Zhu, A. Tao, J. Kautz, and B. Catanzaro, “High-resolution image synthesis and semantic manipulation with condi- tional gans,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 8798–8807. 1, 2, 6, 7, 8, 9

    2018

  8. [17]

    Sean: Image synthesis with semantic region-adaptive normalization,

    P. Zhu, R. Abdal, Y . Qin, and P. Wonka, “Sean: Image synthesis with semantic region-adaptive normalization,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 5104–5113. 1, 2, 3

    2020

  9. [18]

    Auto-encoding variational bayes,

    D. P. Kingma, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114, 2013. 1

    2013 arXiv

  10. [19]

    Image-to-image translation with conditional adversarial networks,

    P. Isola, J.-Y . Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1125–

    2017

  11. [20]

    Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,

    H. Tang, D. Xu, Y . Yan, P. H. Torr, and N. Sebe, “Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 7870–7879. 1, 2

    2020

  12. [21]

    Efficient semantic image synthesis via class-adaptive normalization,

    Z. Tan, D. Chen, Q. Chu, M. Chai, J. Liao, M. He, L. Yuan, G. Hua, and N. Yu, “Efficient semantic image synthesis via class-adaptive normalization,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 9, pp. 4852–4866, 2022. 1, 2, 3, 7, 8, 9, 11

    2022

  13. [22]

    Oasis: only adversarial supervision for semantic image synthesis,

    V . Sushko, E. Sch ¨onfeld, D. Zhang, J. Gall, B. Schiele, and A. Khoreva, “Oasis: only adversarial supervision for semantic image synthesis,” International Journal of Computer Vision , vol. 130, no. 12, pp. 2903– 2923, 2022. 1, 2, 3

    2022

  14. [23]

    Edge guided gans with multi-scale contrastive learning for semantic image synthesis,

    H. Tang, G. Sun, N. Sebe, and L. Van Gool, “Edge guided gans with multi-scale contrastive learning for semantic image synthesis,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023. 1, 3

    2023

  15. [24]

    Retrieval-based spatially adaptive normalization for semantic image synthesis,

    Y . Shi, X. Liu, Y . Wei, Z. Wu, and W. Zuo, “Retrieval-based spatially adaptive normalization for semantic image synthesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 11 224–11 233. 1, 2, 3 JOURNAL OF LATEX CLASS FILES, VO...

    2022

  16. [25]

    Place: Adaptive layout- semantic fusion for semantic image synthesis,

    Z. Lv, Y . Wei, W. Zuo, and K.-Y . K. Wong, “Place: Adaptive layout- semantic fusion for semantic image synthesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 9264–9274. 1, 2

    2024

  17. [26]

    Freestyle layout- to-image synthesis,

    H. Xue, Z. Huang, Q. Sun, L. Song, and W. Zhang, “Freestyle layout- to-image synthesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 14 256–14 266. 1, 2

    2023

  18. [27]

    Remote sensing image synthesis via semantic embedding generative adversarial networks,

    C. Wang, B. Chen, Z. Zou, and Z. Shi, “Remote sensing image synthesis via semantic embedding generative adversarial networks,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–11, 2023. 2, 3, 8

    2023

  19. [28]

    Semantic-shape adaptive feature modulation for semantic image synthesis,

    Z. Lv, X. Li, Z. Niu, B. Cao, and W. Zuo, “Semantic-shape adaptive feature modulation for semantic image synthesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 11 214–11 223. 2, 3, 5, 7, 8, 9

    2022

  20. [29]

    Semantically multi-modal image synthesis,

    Z. Zhu, Z. Xu, A. You, and X. Bai, “Semantically multi-modal image synthesis,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 5467–5476. 2, 8

    2020

  21. [30]

    Spatially-adaptive pixelwise networks for fast image translation,

    T. R. Shaham, M. Gharbi, R. Zhang, E. Shechtman, and T. Michaeli, “Spatially-adaptive pixelwise networks for fast image translation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 14 882–14 891. 2

    2021

  22. [31]

    Spatially constrained gan for face and fashion synthesis,

    S. Jiang, H. Liu, Y . Wu, and Y . Fu, “Spatially constrained gan for face and fashion synthesis,” in 2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021) . IEEE, 2021, pp. 01–08. 2, 3

    2021

  23. [32]

    Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,

    H. Tang, D. Xu, Y . Yan, P. H. Torr, and N. Sebe, “Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 7870–7879. 2, 3

    2020

  24. [33]

    Step: Style-based encoder pre-training for multi-modal image synthesis,

    M. Meshry, Y . Ren, L. S. Davis, and A. Shrivastava, “Step: Style-based encoder pre-training for multi-modal image synthesis,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2021, pp. 3712–3721. 3

    2021

  25. [34]

    Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,

    H. Tang, D. Xu, Y . Yan, P. H. Torr, and N. Sebe, “Local class-specific and global image-level generative adversarial networks for semantic- guided scene generation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 7870–7879. 3

    2020

  26. [35]

    Image synthesis via semantic composition,

    Y . Wang, L. Qi, Y .-C. Chen, X. Zhang, and J. Jia, “Image synthesis via semantic composition,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2021, pp. 13 749–13 758. 3

    2021

  27. [36]

    Learning local image descriptors,

    S. A. Winder and M. Brown, “Learning local image descriptors,” in 2007 IEEE Conference on Computer Vision and Pattern Recognition . IEEE, 2007, pp. 1–8. 3

    2007

  28. [37]

    Local feature descriptor for image matching: A survey,

    C. Leng, H. Zhang, B. Li, G. Cai, Z. Pei, and L. He, “Local feature descriptor for image matching: A survey,”IEEE Access, vol. 7, pp. 6424– 6434, 2018. 3

    2018

  29. [38]

    An efficient image descriptor for image classification and cbir,

    A. Shakarami and H. Tarrah, “An efficient image descriptor for image classification and cbir,” Optik, vol. 214, p. 164833, 2020. 3

    2020

  30. [39]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778. 5

    2016

  31. [40]

    Xception: Deep learning with depthwise separable convolu- tions,

    F. Chollet, “Xception: Deep learning with depthwise separable convolu- tions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1251–1258. 5

    2017

  32. [41]

    Rotate to attend: Convolutional triplet attention module,

    D. Misra, T. Nalamada, A. U. Arasanipalai, and Q. Hou, “Rotate to attend: Convolutional triplet attention module,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision , 2021, pp. 3139–3148. 5

    2021

  33. [42]

    Scconv: Spatial and channel reconstruction convolution for feature redundancy,

    J. Li, Y . Wen, and L. He, “Scconv: Spatial and channel reconstruction convolution for feature redundancy,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 6153–6162. 5

    2023

  34. [43]

    You only need adversarial supervision for semantic image synthesis,

    V . Sushko, E. Sch ¨onfeld, D. Zhang, J. Gall, B. Schiele, and A. Khoreva, “You only need adversarial supervision for semantic image synthesis,” arXiv preprint arXiv:2012.04781 , 2020. 6

    2012 arXiv

  35. [44]

    Large scale gan training for high fidelity natural image synthesis,

    A. Brock, “Large scale gan training for high fidelity natural image synthesis,” arXiv preprint arXiv:1809.11096 , 2018. 6

    2018 arXiv

  36. [45]

    Conditional generative adversarial nets,

    M. Mirza, “Conditional generative adversarial nets,” arXiv preprint arXiv:1411.1784, 2014. 6

    2014 arXiv

  37. [46]

    Self-attention generative adversarial networks,

    H. Zhang, I. Goodfellow, D. Metaxas, and A. Odena, “Self-attention generative adversarial networks,” inInternational conference on machine learning. PMLR, 2019, pp. 7354–7363. 6

    2019

  38. [47]

    Spectral normalization for generative adversarial networks,

    T. Miyato, T. Kataoka, M. Koyama, and Y . Yoshida, “Spectral normalization for generative adversarial networks,” arXiv preprint arXiv:1802.05957, 2018. 6

    2018 arXiv

  39. [48]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014. 6

    2014 arXiv

  40. [49]

    Unpaired image-to-image translation with shortest path regularization,

    S. Xie, Y . Xu, M. Gong, and K. Zhang, “Unpaired image-to-image translation with shortest path regularization,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 10 177–10 187. 7, 8, 9

    2023

  41. [50]

    Land-cover classification with high-resolution remote sensing images using transferable deep models,

    X.-Y . Tong, G.-S. Xia, Q. Lu, H. Shen, S. Li, S. You, and L. Zhang, “Land-cover classification with high-resolution remote sensing images using transferable deep models,” Remote Sensing of Environment , vol. 237, p. 111322, 2020. 7, 8

    2020

  42. [51]

    Isprs semantic labeling contest,

    F. Rottensteiner, G. Sohn, M. Gerke, and J. D. Wegner, “Isprs semantic labeling contest,” ISPRS: Leopoldsh ¨ohe, Germany , vol. 1, no. 4, p. 4,

  43. [52]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014. 7

    2014 arXiv

  44. [53]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” Advances in neural information processing systems , vol. 30, 2017. 7

    2017

  45. [54]

    Inception-v3 for flower classification,

    X. Xia, C. Xu, and B. Nan, “Inception-v3 for flower classification,” in 2017 2nd international conference on image, vision and computing (ICIVC). IEEE, 2017, pp. 783–787. 7

    2017

  46. [55]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    N. B. Bynagari, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” Asian Journal of Applied Science and Engineering, vol. 8, no. 25-34, p. 6, 2019. 7, 8

    2019

  47. [56]

    Encoder- decoder with atrous separable convolution for semantic image segmen- tation

    L.-C. Chen, Y . Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder- decoder with atrous separable convolution for semantic image segmen- tation.” in ECCV, 2018, pp. 833–851. 7

    2018

  48. [57]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595. 8

    2018

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.