pith. sign in

arxiv: 2606.09347 · v2 · pith:U7W4FTLFnew · submitted 2026-06-08 · 💻 cs.CV

IB-HFN: Information Bottleneck-Driven SAR-Optical Fusion Network for High-Fidelity Cloud Removal

Haojun Guo , Fan Feng , Ziquan Wang , Yongsheng Zhang , Ying Yu This is my paper

Pith reviewed 2026-06-27 16:57 UTC · model grok-4.3

classification 💻 cs.CV
keywords cloud removalSAR-optical fusioninformation bottleneckremote sensingimage reconstructionspeckle noise suppressionmultimodal fusionsatellite imagery
0
0 comments X

The pith

An information bottleneck in SAR-optical fusion suppresses speckle noise while routing clear optical details for cloud removal.

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

The paper develops IB-HFN to recover surface details hidden by clouds in optical images by drawing on complementary SAR observations. Prior fusion approaches often pass SAR speckle noise forward or over-smooth textures because they concatenate modalities directly and supervise only at the pixel level. IB-HFN keeps modality streams separate until a dedicated fusion stage, then applies a channel-wise variational information bottleneck to compress SAR features and a local-global gate to select reliable optical regions. A joint loss that mixes feature regularization with image-level terms on accuracy, structure, spectrum, and sharpness, plus a dynamic weighting schedule, stabilizes the process. Experiments on the SEN12MS-CR dataset under difficult spatio-temporal splits report gains in structural preservation and spectral fidelity.

Core claim

IB-HFN achieves superior structural preservation and spectral fidelity by using a dual-stream backbone to maintain modality-specific representations, a Spatial Information Bottleneck Fusion module that compresses SAR features through a channel-wise variational information bottleneck to suppress unstructured speckle noise, and a parallel local-global gating mechanism that predicts clear-sky regions and routes optical details via a Dirac-initialized skip connection, all trained with feature-level bottleneck regularization plus image-level constraints on reconstruction accuracy, structural consistency, spectral fidelity, and contrastive sharpness under a dynamic weighting schedule.

What carries the argument

Spatial Information Bottleneck Fusion module that compresses SAR features via channel-wise variational information bottleneck while a local-global gating mechanism routes optical details through Dirac-initialized skip connections.

If this is right

  • SAR speckle noise is prevented from propagating into the final optical reconstruction.
  • Noise suppression and texture preservation are handled by separate mechanisms rather than a single pixel-wise loss.
  • Structural consistency and spectral fidelity improve on datasets with challenging spatio-temporal distribution shifts.
  • Dynamic weighting of multiple loss terms reduces hazy artifacts during training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same separation of compression and gating could be tested on other noisy multimodal pairs such as LiDAR and optical data.
  • If the gating predictions prove reliable, the architecture might support incremental updates when new clear-sky optical patches become available.
  • The joint optimization schedule may generalize to any remote-sensing fusion task that must trade off fidelity against artifact suppression.

Load-bearing premise

The channel-wise variational information bottleneck will reliably suppress SAR speckle noise while the local-global gating mechanism will accurately identify clear-sky regions and route optical details without introducing new artifacts or losing fine texture.

What would settle it

Compare output images on SEN12MS-CR test cases containing heavy cloud cover and strong SAR speckle against ground truth; if the method produces more visible noise patterns, blurring, or spectral shifts than the best baseline, the central claim does not hold.

Figures

Figures reproduced from arXiv: 2606.09347 by Fan Feng, Haojun Guo, Ying Yu, Yongsheng Zhang, Ziquan Wang.

Figure 1
Figure 1. Figure 1: Visual illustration of the limitations of existing multimodal cloud [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed Information Bottleneck Driven High Fidelity Network. (a) IB-HFN adopts a dual stream architecture, where a Transformer [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the Joint Reconstruction Constraint. It integrates MS [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison under extreme thick-cloud scenarios with temporal semantic shifts. Recent state-of-the-art methods, such as UnCRtainTS (d) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual ablation of the Quadruple High-Fidelity Loss and SIBF [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of cloud removal results on the LuojiaSET [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Intermediate feature visualization of the proposed SIBF module. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
read the original abstract

Synthetic aperture radar (SAR)-assisted optical cloud removal aims to recover surface information obscured by clouds in optical remote sensing images by exploiting complementary SAR observations. Existing multimodal fusion methods typically rely on direct spatial concatenation and pixel-wise supervision, which can propagate SAR speckle noise into optical reconstruction and lead to over-smoothed results. To address these limitations, we propose an Information Bottleneck-driven High-Fidelity Network (IB-HFN) for SAR-assisted optical cloud removal. IB-HFN employs a dual-stream backbone to preserve modality-specific representations before deep semantic fusion, thereby mitigating premature cross-modal contamination. At the fusion stage, we introduce a Spatial Information Bottleneck Fusion module that compresses SAR features through a channel-wise variational information bottleneck to suppress unstructured speckle noise. In parallel, a local-global gating mechanism predicts clear-sky regions and routes reliable optical details through a Dirac-initialized skip connection, decoupling noise suppression from texture preservation. We further develop a joint optimization strategy that integrates feature-level bottleneck regularization with image-level constraints on reconstruction accuracy, structural consistency, spectral fidelity, and contrastive sharpness. A dynamic weighting schedule balances these objectives to stabilize training and reduce hazy artifacts. Experiments on the SEN12MS-CR dataset under challenging spatio-temporal splits demonstrate that IB-HFN achieves superior structural preservation and spectral fidelity over existing methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript proposes the Information Bottleneck-driven High-Fidelity Network (IB-HFN) for SAR-assisted optical cloud removal. It employs a dual-stream backbone to preserve modality-specific representations, a Spatial Information Bottleneck Fusion module that applies a channel-wise variational information bottleneck to compress SAR features and suppress speckle, a local-global gating mechanism with Dirac-initialized skip connections to route reliable optical texture, and a joint optimization combining feature-level bottleneck regularization with image-level losses on reconstruction accuracy, structural consistency, spectral fidelity, and contrastive sharpness. Experiments on the SEN12MS-CR dataset under spatio-temporal splits are stated to demonstrate superior structural preservation and spectral fidelity relative to existing methods.

Significance. If the empirical superiority holds under detailed quantitative scrutiny, the approach offers a principled way to decouple speckle suppression from texture preservation in multimodal remote-sensing fusion, potentially improving upon direct concatenation methods that propagate noise or cause over-smoothing.

major comments (1)
  1. [Spatial Information Bottleneck Fusion module (abstract and §3)] Spatial Information Bottleneck Fusion module (abstract and §3): the central claim that the channel-wise variational information bottleneck suppresses unstructured speckle noise while enabling superior structural preservation rests on an assumption that may not hold. SAR speckle is multiplicative and spatially correlated; a channel-wise operation on per-channel statistics cannot explicitly model or remove these spatially structured components, raising the possibility that residual noise propagates or forces over-smoothing and directly undermining the reported gains on the spatio-temporal splits.
minor comments (1)
  1. [Abstract] The abstract asserts quantitative superiority but supplies no metrics, baselines, or error bars; including at least the key PSNR/SSIM values and comparison table references would strengthen the summary.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and the substantive comment on the Spatial Information Bottleneck Fusion module. Below we respond directly to the concern raised.

read point-by-point responses

  1. Referee: [Spatial Information Bottleneck Fusion module (abstract and §3)] Spatial Information Bottleneck Fusion module (abstract and §3): the central claim that the channel-wise variational information bottleneck suppresses unstructured speckle noise while enabling superior structural preservation rests on an assumption that may not hold. SAR speckle is multiplicative and spatially correlated; a channel-wise operation on per-channel statistics cannot explicitly model or remove these spatially structured components, raising the possibility that residual noise propagates or forces over-smoothing and directly undermining the reported gains on the spatio-temporal splits.

    Authors: We agree that SAR speckle is multiplicative and spatially correlated, and that a purely channel-wise variational information bottleneck does not contain an explicit spatial noise model. The convolutional feature extractors preceding the bottleneck do, however, operate on spatially local receptive fields, so the per-channel statistics are computed on representations that already encode local spatial context. The information-bottleneck objective then regularizes these features toward task-relevant content, which in practice reduces the transmission of speckle-related variance. This is consistent with the observed improvements in structural (SSIM, edge metrics) and spectral (SAM) fidelity on the spatio-temporal splits of SEN12MS-CR. We do not claim an explicit spatial speckle filter; rather, the module provides a principled compression regularizer whose empirical effect is noise suppression without the over-smoothing seen in direct-concatenation baselines. We will add a short clarifying paragraph in §3.2 acknowledging the multiplicative/spatial nature of speckle and the reliance on convolutional context plus empirical validation. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical network proposal without derivation chain

full rationale

The paper describes an empirical deep learning architecture (IB-HFN) for SAR-optical cloud removal, relying on a dual-stream backbone, Spatial Information Bottleneck Fusion module with channel-wise variational information bottleneck, local-global gating, and joint optimization with dynamic weighting. No mathematical derivations, first-principles predictions, or equations are presented that could reduce to inputs by construction. Claims rest on experimental results on SEN12MS-CR under spatio-temporal splits, with no self-citation load-bearing steps or fitted parameters renamed as predictions. The method is self-contained as a proposed architecture validated empirically.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract contains no mathematical derivations, free parameters, or explicit axioms; all claims rest on the unverified performance of the described modules.

pith-pipeline@v0.9.1-grok · 5775 in / 1098 out tokens · 27841 ms · 2026-06-27T16:57:03.454842+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 38 canonical work pages · 4 internal anchors

  1. [1]

    Spatial and temporal distribution of clouds observed by MODIS onboard the Terra and Aqua satellites,

    M. D. King, S. Platnick, W. P. Menzel, S. A. Ackerman, and P. A. Hubanks, “Spatial and temporal distribution of clouds observed by MODIS onboard the Terra and Aqua satellites,”IEEE Transactions on Geoscience and Remote Sensing, vol. 51, no. 7, pp. 3826–3852, 2013. [Online]. Available: https://doi.org/10.1109/TGRS.2012.2227333

    work page doi:10.1109/tgrs.2012.2227333 2013

  2. [2]

    Cloud-guided fusion with sar-to-optical translation for thick cloud removal,

    X. Xiang, Y . Tan, and L. Yan, “Cloud-guided fusion with sar-to-optical translation for thick cloud removal,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 0196–2892, 2024. [Online]. Available: https://doi.org/10.1109/TGRS.2024.3431556

    work page doi:10.1109/tgrs.2024.3431556 2024

  3. [3]

    Transfusion-cr: Two-phase sar-to-optical translation and deep feature fusion for cloud removal,

    R. Liu, S. Meng, Y . Peng, X. Tianet al., “Transfusion-cr: Two-phase sar-to-optical translation and deep feature fusion for cloud removal,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 0196–2892, 2024. [Online]. Available: https: //doi.org/10.1109/TGRS.2024.3439854

    work page doi:10.1109/tgrs.2024.3439854 2024

  4. [4]

    An unsupervised sar and optical image fusion network based on structure-texture decomposition,

    Y . Ye, W. Liu, L. Zhou, T. Peng, Q. Xuet al., “An unsupervised sar and optical image fusion network based on structure-texture decomposition,” IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1545–598X,

  5. [5]

    Available: https://doi.org/10.1109/LGRS.2022.3219341

    [Online]. Available: https://doi.org/10.1109/LGRS.2022.3219341

    work page doi:10.1109/lgrs.2022.3219341 2022

  6. [6]

    Clear: A fast cloud removal method for optical remote sensing images combining class-based linear regression and iterative residual compensation,

    H. Guo, Y . Zheng, H. Xu, and L. Bruzzone, “Clear: A fast cloud removal method for optical remote sensing images combining class-based linear regression and iterative residual compensation,” IEEE Transactions on Geoscience and Remote Sensing, vol. 64, 2026. [Online]. Available: https://doi.org/10.1109/TGRS.2026.3651961

    work page doi:10.1109/tgrs.2026.3651961 2026

  7. [7]

    From adversarial competition to iterative refinement: A critical review of generative ai for remote sensing scene classification,

    L. Xu, Q. Luo, L. Chen, T. Ni, Q. Lin, and S. Zhao, “From adversarial competition to iterative refinement: A critical review of generative ai for remote sensing scene classification,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, pp. 10 573–10 634, 2026. [Online]. Available: https://doi.org/10.1109/JSTARS.2026.3670966

    work page doi:10.1109/jstars.2026.3670966 2026

  8. [8]

    Thick cloud removal using cloud mask-guided sar-optical fusion network,

    S. Meng, S. Li, X. Wang, J. Yang, G. Song, and Y . Ding, “Thick cloud removal using cloud mask-guided sar-optical fusion network,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026. [Online]. Available: https: //doi.org/10.1109/JSTARS.2026.3651496

    work page doi:10.1109/jstars.2026.3651496 2026

  9. [9]

    Idf-cr: Iterative diffusion process for divide-and- conquer cloud removal in remote-sensing images,

    M. Wanget al., “Idf-cr: Iterative diffusion process for divide-and- conquer cloud removal in remote-sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 0196–2892, 2024. [Online]. Available: https://doi.org/10.1109/TGRS.2024.3378720

    work page doi:10.1109/tgrs.2024.3378720 2024

  10. [10]

    Adaptive fusion-based two-flow spectral–spatial transformer network for sar-based optical cloud removal,

    M. Wang, J. Yao, B. Zhang, and J. Zhou, “Adaptive fusion-based two-flow spectral–spatial transformer network for sar-based optical cloud removal,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 62, 2025. [Online]. Available: https://doi.org/10.1109/JSTARS.2025.3629856

    work page doi:10.1109/jstars.2025.3629856 2025

  11. [11]

    Task-Driven Prompt Learning: A Joint Framework for Multi-modal Cloud Removal and Segmentation

    Z. Zhang, J. Li, S. Shi, and Q. Yuan, “Task-driven prompt learning: A joint framework for multi-modal cloud removal and segmentation,”arXiv preprint arXiv:2601.12052, 2026. [Online]. Available: https://doi.org/10.48550/arXiv.2601.12052

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2601.12052 2026

  12. [12]

    Diffcr: A fast conditional diffusion framework for cloud removal from optical satellite images,

    X. Zou, K. Li, J. Xing, Y . Zhang, S. Wang, and L. Jin, “Diffcr: A fast conditional diffusion framework for cloud removal from optical satellite images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 0196–2892, 2024. [Online]. Available: https://doi.org/10.1109/TGRS.2024.3365806

    work page doi:10.1109/tgrs.2024.3365806 2024

  13. [13]

    Sader: Structure-aware diffusion framework with deterministic resampling for multi-temporal remote sensing cloud removal,

    Y . Zhang, Q. Chen, Y . Liu, W. Li, and J. Guan, “Sader: Structure-aware diffusion framework with deterministic resampling for multi-temporal remote sensing cloud removal,”arXiv preprint arXiv:2602.00536, 2026. [Online]. Available: https://doi.org/10.48550/arXiv.2602.00536

    work page doi:10.48550/arxiv.2602.00536 2026

  14. [14]

    Uncrtaints: Uncertainty quantification for cloud removal in optical satellite time series,

    P. Ebel, V . S. F. Garnot, M. Schmitt, J. D. Wegner, and X. X. Zhu, “Uncrtaints: Uncertainty quantification for cloud removal in optical satellite time series,” inCVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, 2023, pp. 2160–7508. [Online]. Available: https://doi.org/10.1109/CVPRW59228.2023.00202

    work page doi:10.1109/cvprw59228.2023.00202 2023

  15. [15]

    Enhancing cloud-removed regions in multispectral optical images using sar edge features,

    M. Wang, X. Zhao, and L. Ren, “Enhancing cloud-removed regions in multispectral optical images using sar edge features,”IEEE Geoscience and Remote Sensing Letters, vol. 22, 2025. [Online]. Available: https://doi.org/10.1109/LGRS.2025.3574323 12

    work page doi:10.1109/lgrs.2025.3574323 2025

  16. [16]

    Glf-cr: Sar-enhanced cloud removal with global-local fusion,

    F. Xu, Y . Shi, P. Ebel, L. Yu, G.-S. Xia, W. Yang, and X. X. Zhu, “Glf-cr: Sar-enhanced cloud removal with global-local fusion,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 192, pp. 268–278,

  17. [17]

    Available: https://doi.org/10.48550/arXiv.2206.02850

    [Online]. Available: https://doi.org/10.48550/arXiv.2206.02850

    work page doi:10.48550/arxiv.2206.02850

  18. [18]

    Thiefcloud: A thickness fused thin cloud removal network for optical remote sensing image with self-supervised learnable cloud prior,

    A. Zhao, R. Feng, and X. Li, “Thiefcloud: A thickness fused thin cloud removal network for optical remote sensing image with self-supervised learnable cloud prior,”IEEE Transactions on Circuits and Systems for Video Technology, 2025. [Online]. Available: https://doi.org/10.1109/TCSVT.2025.3585720

    work page doi:10.1109/tcsvt.2025.3585720 2025

  19. [19]

    URL https://openaccess.thecvf

    H. Wu, Y . Qu, S. Lin, J. Zhou, R. Qiao, Z. Zhang, Y . Xie, and L. Ma, “Contrastive learning for compact single image dehazing,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 10 551–10 560. [Online]. Available: https://doi.org/10.1109/CVPR46437.2021.00033

    work page doi:10.1109/cvpr46437.2021.00033 2021

  20. [20]

    Contrastive learning via variational information bottleneck,

    J. Li, Y . Wang, X. Zhang, D. Jiang, W. Dai, and C. Li, “Contrastive learning via variational information bottleneck,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. [Online]. Available: https://doi.org/10.1109/TPAMI.2025.3571990

    work page doi:10.1109/tpami.2025.3571990 2025

  21. [21]

    Deep variational information bottleneck,

    A. A. Alemi, I. Fischer, J. V . Dillon, and K. Murphy, “Deep variational information bottleneck,”arXiv preprint arXiv:1612.00410,

    Pith/arXiv arXiv

  22. [22]

    Deep Variational Information Bottleneck

    [Online]. Available: https://doi.org/10.48550/arXiv.1612.00410

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1612.00410

  23. [24]

    Mmcanet: A multimodal and cross-attention network for cloud removal and exploration of progressive remote sensing images restoration algorithm,

    Y . Zhou, J. Suo, Y . Wang, J. Su, W. Xiao, and Z. Hong, “Mmcanet: A multimodal and cross-attention network for cloud removal and exploration of progressive remote sensing images restoration algorithm,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, 2025. [Online]. Available: https://doi.org/10.1109/TGRS.2025.3556560

    work page doi:10.1109/tgrs.2025.3556560 2025

  24. [25]

    Transformer- based sar-optical fusion network for cloud removal in satellite imagery,

    T.-T. Le, T.-A. Bui, H. Tran, Y .-L. Chang, and P.-J. Lee, “Transformer- based sar-optical fusion network for cloud removal in satellite imagery,” in2025 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan), 2025. [Online]. Available: https://doi.org/10.1109/ICCE-Taiwan66881.2025.11207770

    work page doi:10.1109/icce-taiwan66881.2025.11207770 2025

  25. [26]

    Graph-based robust cloud removal via optical-sar image fusion,

    R. Isono, S. Ono, and A. Ortega, “Graph-based robust cloud removal via optical-sar image fusion,” inIGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium, 2025. [Online]. Available: https://doi.org/10.1109/IGARSS.2025.11243640

    work page doi:10.1109/igarss.2025.11243640 2025

  26. [27]

    Cloud-aware sar fusion for enhanced optical sensing in space missions,

    T.-A. Bui and T.-T. Le, “Cloud-aware sar fusion for enhanced optical sensing in space missions,”2025 IEEE Radar Conference (RadarConf25), 2025. [Online]. Available: https://doi.org/10.1109/ RadarConf2559087.2025.11205145

    arXiv 2025

  27. [28]

    Att-cr: Adaptive triangular transformer for cloud removal,

    Y . Wu, Y . Deng, P. Li, W. Huang, K. Wu, and X. Xin, “Att-cr: Adaptive triangular transformer for cloud removal,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025. [Online]. Available: https://doi.org/10.1109/JSTARS.2025.3596856

    work page doi:10.1109/jstars.2025.3596856 2025

  28. [29]

    Dual-view prompting for cloud removal,

    Y . Deng, W. Huang, Z. Tang, and J. Duan, “Dual-view prompting for cloud removal,”IEEE Transactions on Geoscience and Remote Sensing,

  29. [30]

    Available: https://doi.org/10.1109/TGRS.2025.3614865

    [Online]. Available: https://doi.org/10.1109/TGRS.2025.3614865

    work page doi:10.1109/tgrs.2025.3614865 2025

  30. [31]

    Advancing pose-guided image synthesis with progressive conditional diffusion models,

    F. Shen, H. Ye, J. Zhang, C. Wang, X. Han, and Y . Wei, “Advancing pose-guided image synthesis with progressive conditional diffusion models,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://openreview.net/ forum?id=rHzapPnCgT

    2024

  31. [32]

    Imagpose: A unified conditional framework for pose-guided person generation,

    F. Shen and J. Tang, “Imagpose: A unified conditional framework for pose-guided person generation,”Advances in neural information processing systems, vol. 37, pp. 6246–6266, 2024

    2024

  32. [33]

    Long-term talkingface generation via motion-prior conditional diffusion model,

    F. Shen, C. Wang, J. Gao, Q. Guo, J. Dang, J. Tang, and T.-S. Chua, “Long-term talkingface generation via motion-prior conditional diffusion model,” inForty-second International Conference on Machine Learning, 2025

    2025

  33. [34]

    Boost- ing consistency in story visualization with rich-contextual conditional diffusion models,

    F. Shen, H. Ye, S. Liu, J. Zhang, C. Wang, X. Han, and Y . Wei, “Boost- ing consistency in story visualization with rich-contextual conditional diffusion models,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 7, 2025, pp. 6785–6794

    2025

  34. [35]

    Imagedit: Let any subject transform,

    F. Shen, W. Xu, R. Yan, D. Zhang, X. Shu, and J. Tang, “Imagedit: Let any subject transform,”arXiv preprint arXiv:2510.01186, 2025

    Pith/arXiv arXiv 2025

  35. [36]

    Imaghar- mony: Controllable image editing with consistent object quantity and layout,

    F. Shen, X. Du, Y . Gao, J. Yu, Y . Cao, X. Lei, and J. Tang, “Imaghar- mony: Controllable image editing with consistent object quantity and layout,”arXiv preprint arXiv:2506.01949, 2025

    arXiv 2025

  36. [37]

    Imagdressing-v1: Customizable virtual dressing,

    F. Shen, X. Jiang, X. He, H. Ye, C. Wang, X. Du, Z. Li, and J. Tang, “Imagdressing-v1: Customizable virtual dressing,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 7, 2025, pp. 6795–6804

    2025

  37. [38]

    Imaggarment-1: Fine-grained garment generation for controllable fashion design,

    F. Shen, J. Yu, C. Wang, X. Jiang, X. Du, and J. Tang, “Imaggarment-1: Fine-grained garment generation for controllable fashion design,”arXiv preprint arXiv:2504.13176, 2025

    arXiv 2025

  38. [39]

    Multimodal diffusion bridge with attention-based sar fusion for satellite image cloud removal,

    Y . Hu, S. Lohit, U. S. Kamilov, and T. K. Marks, “Multimodal diffusion bridge with attention-based sar fusion for satellite image cloud removal,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, 2025. [Online]. Available: https://doi.org/10.1109/TGRS.2025.3604654

    work page doi:10.1109/tgrs.2025.3604654 2025

  39. [40]

    A local anomaly detection method for occluded targets with multispectral point clouds,

    J. Xu, X. Yu, J. Panet al., “Efficient diffusion model for sar-assisted optical remote sensing image cloud removal,”IEEE International Geoscience and Remote Sensing Symposium, 2025. [Online]. Available: https://doi.org/10.1109/IGARSS55030.2025.11243696

    work page doi:10.1109/igarss55030.2025.11243696 2025

  40. [41]

    Cloud removal method for remote sensing images based on sar-guided conditional diffusion model and style adaptation,

    J. Chen, X. Tong, Y . Sun, and Z. Xiao, “Cloud removal method for remote sensing images based on sar-guided conditional diffusion model and style adaptation,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, pp. 22 770–22 783, 2025. [Online]. Available: https://doi.org/10.1109/JSTARS.2025.3599162

    work page doi:10.1109/jstars.2025.3599162 2025

  41. [42]

    CBEN -- A Multimodal Machine Learning Dataset for Cloud Robust Remote Sensing Image Understanding

    M. Stricker, M. Iwamura, and K. Kise, “Cben: A multimodal machine learning dataset for cloud robust remote sensing image understanding,”arXiv preprint arXiv:2602.12652, 2026. [Online]. Available: https://doi.org/10.48550/arXiv.2602.12652

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2602.12652 2026

  42. [43]

    Mdfanet: Multidimensional feature alignment network for self- supervised pretraining in remote sensing change detection,

    L. Wang, C. Wang, R. Li, J. Yu, H. Sun, and W. Wang, “Mdfanet: Multidimensional feature alignment network for self- supervised pretraining in remote sensing change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025. [Online]. Available: https://doi.org/10.1109/JSTARS. 2025.3644606

    work page doi:10.1109/jstars 2025

  43. [44]

    Deep learning and the information bottleneck principle,

    N. Tishby and N. Zaslavsky, “Deep learning and the information bottleneck principle,” inIEEE Information Theory Workshop (ITW). IEEE, 2015, pp. 1–5

    2015

  44. [45]

    Projection head is secretly an information bottleneck,

    Z. Ouyang, K. Hu, Q. Zhang, Y . Wang, Y . Wanget al., “Projection head is secretly an information bottleneck,” inInternational Conference on Learning Representations, 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2503.00507

    work page doi:10.48550/arxiv.2503.00507 2025

  45. [46]

    Video-bench: Human-aligned video generation benchmark

    J.-H. Hong, H.-J. Kim, K.-S. Jeon, and S.-W. Lee, “Comprehensive information bottleneck for unveiling universal attribution to interpret vision transformers,” in2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. [Online]. Available: https://doi.org/10.1109/CVPR52734.2025.02343

    work page doi:10.1109/cvpr52734.2025.02343 2025

  46. [47]

    Ibcs: Learning information bottleneck-constrained denoised causal subgraph for graph classification,

    R. Yuan, Y . Tang, Y . Xiao, and W. Zhang, “Ibcs: Learning information bottleneck-constrained denoised causal subgraph for graph classification,”IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2024. [Online]. Available: https: //doi.org/10.1109/TPAMI.2024.3508766

    work page doi:10.1109/tpami.2024.3508766 2024

  47. [48]

    Graph information bottleneck for remote sensing segmentation,

    Y . Shou, W. Ai, T. Meng, and N. Yin, “Graph information bottleneck for remote sensing segmentation,”Neurocomputing, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2312.02545

    work page doi:10.48550/arxiv.2312.02545 2024

  48. [49]

    Information-bottleneck driven binary neural network for change detection,

    K. Yin, Z. Zhang, S. Kong, T. Gao, C. Xu, H. Konget al., “Information-bottleneck driven binary neural network for change detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025. [Online]. Available: https: //doi.org/10.48550/arXiv.2507.03504

    work page doi:10.48550/arxiv.2507.03504 2025

  49. [50]

    Refining the information bottleneck via adversarial information separation,

    S. Ning, Z. Wang, L. Wang, B. Chen, S. Liu, X. Wu, J. Zhou, and B. Yang, “Refining the information bottleneck via adversarial information separation,”arXiv preprint arXiv:2602.06549, 2026. [Online]. Available: https://doi.org/10.48550/arXiv.2602.06549

    work page doi:10.48550/arxiv.2602.06549 2026

  50. [51]

    VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck

    F. Zhang, Y . Wu, Z. Wang, X. Wang, C. Lv, X. Huang, and X. Zheng, “Vib-probe: Detecting and mitigating hallucinations in vision-language models via variational information bottleneck,” arXiv preprint arXiv:2601.05547, 2026. [Online]. Available: https: //doi.org/10.48550/arXiv.2601.05547

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2601.05547 2026

  51. [52]

    Hpn-cr: Heterogeneous parallel network for sar-optical data fusion cloud removal,

    P. Gu, W. Liu, S. Feng, T. Wei, J. Wang, H. Chenet al., “Hpn-cr: Heterogeneous parallel network for sar-optical data fusion cloud removal,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 0196–2892, 2025. [Online]. Available: https: //doi.org/10.1109/TGRS.2025.3546489

    work page doi:10.1109/tgrs.2025.3546489 2025

  52. [53]

    Sen12ms- cr-ts: A remote-sensing data set for multimodal multitemporal cloud removal,

    P. Ebel, Y . Xu, M. Schmitt, X. X. Zhuet al., “Sen12ms- cr-ts: A remote-sensing data set for multimodal multitemporal cloud removal,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 0196–2892, 2022. [Online]. Available: https: //doi.org/10.1109/TGRS.2022.3146246

    work page doi:10.1109/tgrs.2022.3146246 2022

  53. [54]

    Hdrsa-net: Hybrid dynamic residual self-attention network for sar-assisted optical image cloud and shadow removal,

    J. Pan, J. Xu, X. Yu, G. Ye, M. Wang, Y . Chen, and J. Ma, “Hdrsa-net: Hybrid dynamic residual self-attention network for sar-assisted optical image cloud and shadow removal,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 218, pp. 258–275, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0924271624004039

    2024