pith. sign in

arxiv: 2605.16967 · v1 · pith:VIEPR2YOnew · submitted 2026-05-16 · 💻 cs.CV

Expandable, Compressible, Mineable: Open-World Thermal Image Restoration

Pu Li , Huafeng Li , Yafei Zhang , Wen Wang , Neng Dong , Jie Wen This is my paper

Pith reviewed 2026-05-19 20:25 UTC · model grok-4.3

classification 💻 cs.CV
keywords thermal infrared image restorationcontinual learningopen-world adaptationmodel expansionmodel compressionknowledge miningimage degradations
0
0 comments X

The pith

ECMRNet adapts to new thermal degradations by expanding isolated subspaces, pruning redundancies, and mining historical knowledge.

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

The paper proposes ECMRNet for thermal infrared image restoration in open-world settings where degradations keep emerging and evolving. Most existing methods assume a fixed closed set of degradations and cannot adapt continually without full retraining. ECMRNet treats adaptation as an expand-compress-mine loop: it expands new group-isolated subspaces for novel degradations while freezing old ones, prunes redundant channels through structural entropy minimization to limit growth, and mines transferable components from past representations to handle compound degradations. Experiments show the resulting model delivers stronger restoration across single and mixed degradations while using fewer parameters and less computation.

Core claim

ECMRNet unifies continual degradation learning as an expand-compress-mine closed-loop process. It decomposes intermediate representations into group-isolated subspaces and achieves strict parameter isolation and fast adaptation to new degradations by freezing historical groups and isomorphically expanding new ones. Structural Entropy Pruning identifies and removes redundant channel groups via two-dimensional structural entropy minimization. A Sub-degradation Knowledge Mining Module dynamically retrieves and recombines transferable components from historical representations to improve restoration under compound degradations.

What carries the argument

Group-isolated subspaces that isolate parameters for different degradations, allowing historical groups to be frozen while new isomorphic groups expand, together with structural entropy pruning for compression and a knowledge mining module for recombining past components.

If this is right

  • Restoration performance on previously seen degradations remains intact after new degradations are added.
  • Model size stays bounded as the number of degradation tasks grows because pruning removes redundant groups.
  • Compound degradations are handled more effectively by recombining knowledge mined from earlier tasks.
  • The overall approach yields higher performance than prior methods at lower parameter count and computational cost.

Where Pith is reading between the lines

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

  • The isolation and mining mechanisms could be tested on continual learning tasks outside thermal imaging, such as visible-light restoration or video enhancement.
  • If the subspaces remain cleanly separated, the method might support selective unlearning of specific degradations without affecting others.
  • Combining the expand-compress-mine loop with sensor-specific data streams could support long-term deployment on edge devices that encounter changing environmental conditions.

Load-bearing premise

Decomposing intermediate representations into group-isolated subspaces permits strict parameter isolation and fast adaptation to new degradations without interference or loss of previously learned restoration capability.

What would settle it

Sequentially training the model on a series of new degradations and then measuring whether restoration accuracy on the original set of degradations drops substantially.

Figures

Figures reproduced from arXiv: 2605.16967 by Huafeng Li, Jie Wen, Neng Dong, Pu Li, Wen Wang, Yafei Zhang.

Figure 1
Figure 1. Figure 1: Conceptual illustration of the proposed ECMRNet, em￾bodying the “expand–compress–mine” closed-loop mechanism for continual TIR restoration. 1 arXiv:2605.16967v1 [cs.CV] 16 May 2026 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the proposed ECMRNet. ECMRNet unifies continual TIR degradation learning into an expand–compress– mine closed loop: SCGE expands stage-wise channel groups for interference-free adaptation, SEP prunes redundant groups via 2D-SE minimization to bound model growth, and SKMM dynamically mines transferable components from historical sub-degradation representations to enhance compound-degradation… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on the HM-TIR dataset. More qualitative results are provided in the Appendix C. Importantly, SKMM does not assume all historical knowl￾edge should be reused; instead, it selectively weights chan￾nels via sample-adaptive low- rank mining matrix A and injects only transferable components, without requiring extra sub-degradation supervision. This enables the com￾pound branch to continua… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on the EN dataset. More qualita￾tive results are provided in the Appendix C. 4.3. Results on Real-world TIR To evaluate the real-world generalization of ECMRNet, we conduct experiments on three TIR datasets: EN (low con￾trast + blur), TIR100 (blur + noise), and AWMM (low con￾trast + blur + noise). As reported in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Model Convergence in fine-tuning stage Effectiveness of SEP. From [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Additional qualitative comparison on the M3 FD dataset. D. Additional Ablation Studies We further analyze two key hyperparameters in SEP: the sample pool size N and the frequency threshold ρ. We report the post-pruning inference complexity (Params/FLOPs) and restoration performance (PSNR/SSIM) on three single-degradation tasks (low contrast, blur, and noise), with results shown in [PITH_FULL_IMAGE:figures… view at source ↗
Figure 8
Figure 8. Figure 8: Additional qualitative comparison on real-world datasets (EN, TIR100, and AWMM). The impact of ρ [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

In open-world settings, thermal infrared (TIR) image degradations continuously emerge and evolve, while most existing all-in-one restoration methods are built on a closed-set assumption and struggle to continually adapt to novel degradations. To address this, we propose ECMRNet, an Expandable, Compressible, and Mineable Restoration Network for open-world TIR restoration from a continual learning perspective. Conceptually, ECMRNet unifies continual degradation learning as an "expand-compress-mine" closed-loop process, enabling sustained adaptation to new degradations with controllable evolution. Structurally, ECMRNet decomposes intermediate representations into group-isolated subspaces, and achieves strict parameter isolation and fast adaptation to new degradations by freezing historical groups and isomorphically expanding new ones. To curb model growth as tasks accumulate, we present Structural Entropy Pruning, which identifies and removes redundant channel groups via two-dimensional structural entropy minimization, achieving information contribution-driven adaptive compression. Moreover, we design a Sub-degradation Knowledge Mining Module that dynamically retrieves and recombines transferable components from historical representations to improve restoration under compound degradations. Experimental results demonstrate that ECMRNet achieves superior overall performance across diverse single and compound degradations while using fewer parameters and lower computational cost. The source code is available at https://github.com/Kust-lp/ECMRNet.

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

2 major / 2 minor

Summary. The manuscript proposes ECMRNet, an Expandable, Compressible, and Mineable Restoration Network for open-world thermal infrared (TIR) image restoration from a continual learning perspective. It frames adaptation as an expand-compress-mine closed-loop process: intermediate representations are decomposed into group-isolated subspaces with historical groups frozen for isolation and new groups expanded isomorphically; Structural Entropy Pruning removes redundant channel groups via two-dimensional entropy minimization; and a Sub-degradation Knowledge Mining Module dynamically retrieves and recombines historical components for compound degradations. The central claim is superior overall performance across single and compound degradations with fewer parameters and lower computational cost, supported by publicly released source code.

Significance. If the empirical claims hold, the work is significant for addressing open-world continual adaptation in TIR restoration, where degradations evolve and closed-set all-in-one methods fail. The parameter-efficient design via controlled expansion and entropy-driven compression, together with the mining module for compound cases, offers a practical advance. Explicit credit is due for releasing source code, which supports reproducibility and allows independent verification of the isolation and compression mechanisms.

major comments (2)
  1. [Structural description of ECMRNet] Structural description of ECMRNet (abstract and method section): The claim that group-isolated subspaces plus freezing historical groups yields 'strict parameter isolation' and 'zero loss of prior capability' is load-bearing for the no-forgetting guarantee, yet no orthogonality constraint, leakage bound, or ablation measuring cross-group activation is provided. The Sub-degradation Knowledge Mining Module could re-introduce interference for compound degradations sharing latent factors, and Structural Entropy Pruning risks discarding low-entropy but useful historical components without quantified impact on prior-task performance.
  2. [Experimental results section] Experimental results section: The headline claim of superior performance and parameter savings is stated without reference to specific tables, ablation details, error bars, or statistical tests. To substantiate robustness against post-hoc dataset splits or hyper-parameter choices, the full experiments must include quantitative comparisons to baselines, per-degradation breakdowns, and controls for interference/forgetting.
minor comments (2)
  1. [Method] Clarify the precise definition and implementation of 'isomorphically expanding' new groups, ideally with a short equation or pseudocode in the method section.
  2. [Method] Ensure consistent notation for channel groups and entropy terms across figures and text to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment point by point below, providing clarifications on the design choices and outlining specific revisions that will be incorporated into the manuscript.

read point-by-point responses

  1. Referee: [Structural description of ECMRNet] Structural description of ECMRNet (abstract and method section): The claim that group-isolated subspaces plus freezing historical groups yields 'strict parameter isolation' and 'zero loss of prior capability' is load-bearing for the no-forgetting guarantee, yet no orthogonality constraint, leakage bound, or ablation measuring cross-group activation is provided. The Sub-degradation Knowledge Mining Module could re-introduce interference for compound degradations sharing latent factors, and Structural Entropy Pruning risks discarding low-entropy but useful historical components without quantified impact on prior-task performance.

    Authors: We thank the referee for this observation. The strict parameter isolation follows directly from the architecture: intermediate features are decomposed into disjoint group subspaces, historical groups are explicitly frozen (no gradients or updates), and new groups are expanded isomorphically with separate parameters. This design precludes parameter overlap or modification of prior weights. Nevertheless, we acknowledge that explicit supporting analyses are absent. In the revision we will add (i) a quantitative ablation measuring cross-group activation and leakage, (ii) an orthogonality analysis of the learned group subspaces, and (iii) before/after pruning performance on prior tasks to quantify any impact of Structural Entropy Pruning. We will also include targeted experiments on the Sub-degradation Knowledge Mining Module to measure interference under shared latent factors for compound degradations. revision: yes

  2. Referee: [Experimental results section] Experimental results section: The headline claim of superior performance and parameter savings is stated without reference to specific tables, ablation details, error bars, or statistical tests. To substantiate robustness against post-hoc dataset splits or hyper-parameter choices, the full experiments must include quantitative comparisons to baselines, per-degradation breakdowns, and controls for interference/forgetting.

    Authors: We agree that the experimental presentation can be strengthened for transparency and statistical rigor. The current manuscript already contains quantitative comparisons to baselines and reports overall parameter and compute savings; however, we will revise the experimental section to (i) explicitly reference every performance claim to the corresponding table or figure, (ii) add error bars computed over multiple random seeds, (iii) report statistical significance tests, (iv) provide per-degradation performance breakdowns, and (v) include additional ablation controls that isolate interference and forgetting effects. These expansions will directly address concerns about robustness to dataset splits and hyper-parameter choices. revision: yes

Circularity Check

0 steps flagged

No significant circularity: claims rest on architectural definition and experiments

full rationale

The paper defines ECMRNet via explicit architectural mechanisms (group-isolated subspaces, freezing/expansion, entropy pruning, mining module) and presents performance as an experimental outcome. No equations appear in the provided text that reduce any 'prediction' or superiority claim to a fitted parameter or input by construction. No self-citations are invoked as load-bearing uniqueness theorems. The derivation chain is self-contained: the isolation property follows directly from the stated freezing/expansion rule, and overall results are benchmarked externally rather than derived tautologically.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on the premise that feature subspaces can be made strictly isolated by architectural freezing and that structural entropy is a reliable proxy for information contribution; no numerical free parameters are named in the abstract.

axioms (2)
  • domain assumption Freezing historical parameter groups prevents catastrophic forgetting in continual degradation learning.
    Invoked when the paper states that historical groups are frozen to achieve strict parameter isolation.
  • domain assumption Two-dimensional structural entropy minimization identifies redundant channel groups without harming restoration quality.
    Used to justify the Structural Entropy Pruning step.
invented entities (1)
  • Sub-degradation Knowledge Mining Module no independent evidence
    purpose: Dynamically retrieve and recombine transferable components from historical representations for compound degradations.
    New module introduced to handle compound cases; no independent evidence outside the paper is provided.

pith-pipeline@v0.9.0 · 5780 in / 1452 out tokens · 43991 ms · 2026-05-19T20:25:28.982056+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

76 extracted references · 76 canonical work pages · 2 internal anchors

  1. [1]

    arXiv preprint arXiv:2411.10708 , year=

    AllRestorer: All-in-One Transformer for Image Restoration under Composite Degradations , author=. arXiv preprint arXiv:2411.10708 , year=

    work page arXiv

  2. [2]

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

    Complexity experts are task-discriminative learners for any image restoration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  3. [3]

    Gustafsson and Zheng Zhao and Jens Sjölund and Thomas B

    Ziwei Luo and Fredrik K. Gustafsson and Zheng Zhao and Jens Sjölund and Thomas B. Schön , title=. 2024 , pages=

    work page 2024

  4. [4]

    Proceedings of the International Conference on Learning Representations , year=

    Universal Image Restoration Pre-training via Degradation Classification , author=. Proceedings of the International Conference on Learning Representations , year=

    work page

  5. [5]

    arXiv preprint arXiv:2504.08219 , year=

    VL-UR: Vision-Language-guided Universal Restoration of Images Degraded by Adverse Weather Conditions , author=. arXiv preprint arXiv:2504.08219 , year=

    work page arXiv

  6. [6]

    arXiv preprint arXiv:2505.18679 , year=

    Manifold-aware Representation Learning for Degradation-agnostic Image Restoration , author=. arXiv preprint arXiv:2505.18679 , year=

    work page arXiv

  7. [7]

    Prores: Explor- ing degradation-aware visual prompt for universal image restoration.arXiv preprint arXiv:2306.13653, 2023

    Prores: Exploring degradation-aware visual prompt for universal image restoration , author=. arXiv preprint arXiv:2306.13653 , year=

    work page arXiv

  8. [8]

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

    Multimodal prompt perceiver: Empower adaptiveness generalizability and fidelity for all-in-one image restoration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  9. [9]

    Proceedings of the European Conference on Computer Vision , pages=

    Instructir: High-quality image restoration following human instructions , author=. Proceedings of the European Conference on Computer Vision , pages=. 2024 , organization=

    work page 2024

  10. [10]

    Breaking Degradation Coupling: A Structural Entropy Guided Decoupled Framework and Benchmark for Infrared Enhancement

    Breaking Degradation Coupling: A Structural Entropy Guided Decoupled Framework and Benchmark for Infrared Enhancement , author=. arXiv preprint arXiv:2604.22886 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    Proceedings of the European Conference on Computer Vision , pages=

    Onerestore: A universal restoration framework for composite degradation , author=. Proceedings of the European Conference on Computer Vision , pages=. 2024 , organization=

    work page 2024

  12. [12]

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

    Selective hourglass mapping for universal image restoration based on diffusion model , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  13. [13]

    Prompt-in-prompt learning for universal image restoration.arXiv preprint arXiv:2312.05038, 2023

    Prompt-in-prompt learning for universal image restoration , author=. arXiv preprint arXiv:2312.05038 , year=

    work page arXiv

  14. [14]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Simultaneous intensity bias estimation and stripe noise removal in infrared images using the global and local sparsity constraints , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2019 , publisher=

    work page 2019

  15. [15]

    IEEE Transactions on Image Processing , volume=

    Exploring video denoising in thermal infrared imaging: Physics-inspired noise generator, dataset, and model , author=. IEEE Transactions on Image Processing , volume=. 2024 , publisher=

    work page 2024

  16. [16]

    Infrared Physics & Technology , volume=

    HCTIRdeblur: A hybrid convolution-transformer network for single infrared image deblurring , author=. Infrared Physics & Technology , volume=. 2023 , publisher=

    work page 2023

  17. [17]

    Infrared Physics & Technology , volume=

    Thermal wave image deblurring based on depth residual network , author=. Infrared Physics & Technology , volume=. 2021 , publisher=

    work page 2021

  18. [18]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    A combined stripe noise removal and deblurring recovering method for thermal infrared remote sensing images , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2022 , publisher=

    work page 2022

  19. [19]

    Infrared Physics & Technology , volume=

    Infrared thermal image denoising with symmetric multi-scale sampling network , author=. Infrared Physics & Technology , volume=. 2023 , publisher=

    work page 2023

  20. [20]

    Infrared Physics & Technology , volume=

    A real-time contrast enhancement algorithm for infrared images based on plateau histogram , author=. Infrared Physics & Technology , volume=. 2006 , publisher=

    work page 2006

  21. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Thermal-Aware Low-Light Image Enhancement: A Real-World Benchmark and a New Light-Weight Model , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  22. [22]

    IEEE Transactions on Geoscience and Remote Sensing , year=

    Infrared image dynamic range compression based on adaptive contrast adjustment and structure preservation , author=. IEEE Transactions on Geoscience and Remote Sensing , year=

    work page

  23. [23]

    IEEE Signal Processing Letters , volume=

    Fast motion-deblurring of IR images , author=. IEEE Signal Processing Letters , volume=. 2022 , publisher=

    work page 2022

  24. [24]

    Optics & Laser Technology , volume=

    An infrared thermal image denoising method focusing on noise feature learning , author=. Optics & Laser Technology , volume=. 2025 , publisher=

    work page 2025

  25. [25]

    Infrared Physics & Technology , volume=

    Thermal infrared spectrometer on-orbit defocus assessment based on blind image blur kernel estimation , author=. Infrared Physics & Technology , volume=. 2023 , publisher=

    work page 2023

  26. [26]

    Infrared Physics & Technology , volume=

    An infrared image enhancement method via content and detail two-stream deep convolutional neural network , author=. Infrared Physics & Technology , volume=. 2023 , publisher=

    work page 2023

  27. [27]

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

    DEAL: Data-Efficient Adversarial Learning for High-Quality Infrared Imaging , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  28. [28]

    Advances in Neural Information Processing Systems , pages =

    Enhancing Infrared Vision: Progressive Prompt Fusion Network and Benchmark , author=. Advances in Neural Information Processing Systems , pages =

    work page

  29. [29]

    Proceedings of the International Conference on Machine Learning , pages=

    Learning transferable visual models from natural language supervision , author=. Proceedings of the International Conference on Machine Learning , pages=. 2021 , organization=

    work page 2021

  30. [30]

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

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  31. [31]

    arXiv preprint arXiv:2402.02090 , year=

    All-weather multi-modality image fusion: Unified framework and 100k benchmark , author=. arXiv preprint arXiv:2402.02090 , year=

    work page arXiv

  32. [32]

    IEEE Transactions on Image Processing , volume=

    Image quality assessment: from error visibility to structural similarity , author=. IEEE Transactions on Image Processing , volume=. 2004 , publisher=

    work page 2004

  33. [33]

    IEEE Transactions on Image Processing , volume=

    NIMA: Neural image assessment , author=. IEEE Transactions on Image Processing , volume=. 2018 , publisher=

    work page 2018

  34. [34]

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

    Musiq: Multi-scale image quality transformer , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  35. [35]

    completely blind

    Making a “completely blind” image quality analyzer , author=. IEEE Signal Processing Letters , volume=. 2012 , publisher=

    work page 2012

  36. [36]

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

    Restormer: Efficient transformer for high-resolution image restoration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  37. [37]

    Optics Express , volume=

    Why thermal images are blurry , author=. Optics Express , volume=. 2024 , publisher=

    work page 2024

  38. [38]

    IEEE Transactions on Image Processing , volume=

    Nonuniformity correction of infrared image sequences using the constant-statistics constraint , author=. IEEE Transactions on Image Processing , volume=. 1999 , publisher=

    work page 1999

  39. [39]

    Applied optics , volume=

    Single-image-based nonuniformity correction of uncooled long-wave infrared detectors: A deep-learning approach , author=. Applied optics , volume=. 2018 , publisher=

    work page 2018

  40. [40]

    Advances in Neural Information Processing Systems , volume=

    Evidential deep learning to quantify classification uncertainty , author=. Advances in Neural Information Processing Systems , volume=

    work page

  41. [41]

    IEEE Transactions on Information Theory , volume=

    Structural information and dynamical complexity of networks , author=. IEEE Transactions on Information Theory , volume=. 2016 , publisher=

    work page 2016

  42. [42]

    Proceedings of the 2025 ACM Web Conference (WWW) , pages=

    Community detection in large-scale complex networks via structural entropy game , author=. Proceedings of the 2025 ACM Web Conference (WWW) , pages=. 2025 , organization=

    work page 2025

  43. [43]

    Infrared Physics & Technology , volume=

    Target attention deep neural network for infrared image enhancement , author=. Infrared Physics & Technology , volume=. 2021 , publisher=

    work page 2021

  44. [44]

    Optics express , volume=

    Stripe and ring artifact removal with combined wavelet—Fourier filtering , author=. Optics express , volume=. 2009 , publisher=

    work page 2009

  45. [45]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Remote sensing image stripe noise removal: From image decomposition perspective , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2016 , publisher=

    work page 2016

  46. [46]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Ultra-high-definition low-light image enhancement: A benchmark and transformer-based method , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  47. [47]

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

    Efficient Diffusion as Low Light Enhancer , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  48. [48]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Learning with self-calibrator for fast and robust low-light image enhancement , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    work page

  49. [49]

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

    Darkir: Robust low-light image restoration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  50. [50]

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

    Rethinking coarse-to-fine approach in single image deblurring , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  51. [51]

    Proceedings of the European Conference on Computer Vision , pages=

    Stripformer: Strip transformer for fast image deblurring , author=. Proceedings of the European Conference on Computer Vision , pages=. 2022 , organization=

    work page 2022

  52. [52]

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

    Efficient frequency domain-based transformers for high-quality image deblurring , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  53. [53]

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

    Efficient visual state space model for image deblurring , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  54. [54]

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

    Cycleisp: Real image restoration via improved data synthesis , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  55. [55]

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

    Ap-bsn: Self-supervised denoising for real-world images via asymmetric pd and blind-spot network , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  56. [56]

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

    Iterative denoiser and noise estimator for self-supervised image denoising , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  57. [57]

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

    Score priors guided deep variational inference for unsupervised real-world single image denoising , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  58. [58]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Learning without forgetting , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2017 , publisher=

    work page 2017

  59. [59]

    Proceedings of the national academy of sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the national academy of sciences , volume=. 2017 , publisher=

    work page 2017

  60. [60]

    International conference on machine learning , pages=

    Continual learning through synaptic intelligence , author=. International conference on machine learning , pages=. 2017 , organization=

    work page 2017

  61. [61]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Memory aware synapses: Learning what (not) to forget , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

    work page

  62. [62]

    Progressive Neural Networks

    Progressive neural networks , author=. arXiv preprint arXiv:1606.04671 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  63. [63]

    European conference on computer vision , pages=

    Lira: Lifelong image restoration from unknown blended distortions , author=. European conference on computer vision , pages=. 2020 , organization=

    work page 2020

  64. [64]

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

    Packnet: Adding multiple tasks to a single network by iterative pruning , author=. Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

    work page

  65. [65]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Piggyback: Adapting a single network to multiple tasks by learning to mask weights , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

    work page

  66. [66]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Target-aware dual adversarial learning and a multi-scenario multi-modality benchmark to fuse infrared and visible for object detection , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  67. [67]

    Neurocomputing , volume=

    Single infrared image enhancement using a deep convolutional neural network , author=. Neurocomputing , volume=. 2019 , publisher=

    work page 2019

  68. [68]

    Information Fusion , pages=

    All-weather multi-modality image fusion: Unified framework and 100k benchmark , author=. Information Fusion , pages=. 2026 , publisher=

    work page 2026

  69. [69]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Exploring clip for assessing the look and feel of images , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  70. [70]

    Advances in Neural Information Processing Systems , volume=

    Promptir: Prompting for all-in-one image restoration , author=. Advances in Neural Information Processing Systems , volume=

    work page

  71. [71]

    European conference on computer vision , pages=

    Simple baselines for image restoration , author=. European conference on computer vision , pages=. 2022 , organization=

    work page 2022

  72. [72]

    13th international conference on learning representations, ICLR 2025 , pages=

    Adair: Adaptive all-in-one image restoration via frequency mining and modulation , author=. 13th international conference on learning representations, ICLR 2025 , pages=. 2025 , organization=

    work page 2025

  73. [73]

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

    Continuous adverse weather removal via degradation-aware distillation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  74. [74]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    Test-Time Degradation Adaptation for Open-Set Image Restoration , author=. Proceedings of the 41st International Conference on Machine Learning , pages=. 2024 , organization=

    work page 2024

  75. [75]

    Advances in Neural Information Processing Systems , pages=

    Learning both weights and connections for efficient neural networks , author=. Advances in Neural Information Processing Systems , pages=

    work page

  76. [76]

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

    Learning efficient convolutional networks through network slimming , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page