pith. sign in

arxiv: 2312.03798 · v2 · pith:2ROPSC5Hnew · submitted 2023-12-06 · 💻 cs.CV

Single Image Reflection Removal with Patch Reflectance Prior

Dongshen Han , Heechan Yoon , Hyukmin Kwon , Hyun-Cheol Kim , Hyon-Gon Choo , Seungkyu Lee , Chaoning Zhang This is my paper

Pith reviewed 2026-05-24 05:06 UTC · model grok-4.3

classification 💻 cs.CV
keywords single image reflection removalreflection intensity priorpatch-based learningRPENPRRNtransformer U-Netimage restoration
0
0 comments X

The pith

Learning reflection intensity prior from image patches achieves state-of-the-art single image reflection removal

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

The paper proposes a general reflection intensity prior to capture the intensity of reflections in images. It introduces the Reflection Prior Extraction Network to learn this prior from regional patches, allowing for non-uniform reflection modeling. The Prior-based Reflection Removal Network then uses this prior in a transformer U-Net architecture to remove reflections. This is intended to handle the diverse degradations on glass surfaces in real-world images more effectively than prior methods relying on specific assumptions.

Core claim

A general reflection intensity prior learned from regional patches by the Reflection Prior Extraction Network can be used by the Prior-based Reflection Removal Network to achieve state-of-the-art accuracy in single image reflection removal.

What carries the argument

The Reflection Prior Extraction Network (RPEN), which segments images into regional patches to learn non-uniform reflection prior for use in the Prior-based Reflection Removal Network (PRRN).

Load-bearing premise

The reflection intensity prior learned from regional patches is sufficiently general to handle diverse image degradations on the glass surface in real-world images.

What would settle it

A counterexample would be a real-world image set with novel glass degradations where the method does not achieve higher accuracy than existing SIRR methods.

Figures

Figures reproduced from arXiv: 2312.03798 by Chaoning Zhang, Dongshen Han, Heechan Yoon, Hyon-Gon Choo, Hyukmin Kwon, Hyun-Cheol Kim, Seungkyu Lee.

Figure 1
Figure 1. Figure 1: The pipeline of our proposed process [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Patch segmentation and reflection intensity prior calculation for each patch. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sample qualitative comparison results on CDR and Real20 datasets. Our proposed method [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training loss and validation patch pixel error with different segmentation patch counts. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison of patch resolution demonstrates an improvement in reflection removal [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual comparison of patch resolution demonstrates the impact of RPEN on the activation [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
read the original abstract

Single Image Reflection Removal (SIRR) in real-world images is a challenging task due to diverse image degradations occurring on the glass surface during light transmission and reflection. Many existing methods rely on specific prior assumptions to resolve the problem. In this paper, we propose a general reflection intensity prior that captures the intensity of the reflection phenomenon and demonstrate its effectiveness. To learn the reflection intensity prior, we introduce the Reflection Prior Extraction Network (RPEN). By segmenting images into regional patches, RPEN learns non-uniform reflection prior in an image. We propose Prior-based Reflection Removal Network (PRRN) using a simple transformer U-Net architecture that adapts reflection prior fed from RPEN. Experimental results on real-world benchmarks demonstrate the effectiveness of our approach achieving state-of-the-art accuracy in SIRR.

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 / 1 minor

Summary. The manuscript proposes a reflection intensity prior learned by the Reflection Prior Extraction Network (RPEN) from regional image patches to capture non-uniform reflection intensities. This prior is adapted by the Prior-based Reflection Removal Network (PRRN), implemented as a transformer U-Net, to perform single-image reflection removal (SIRR). The authors claim this yields state-of-the-art accuracy on real-world benchmarks.

Significance. If the patch-derived prior generalizes beyond the training distribution to diverse degradations, the approach could offer a practical way to inject adaptive intensity information into SIRR pipelines while retaining a relatively simple transformer backbone. The explicit separation of prior extraction (RPEN) from removal (PRRN) is a clear architectural choice that could be reused.

major comments (2)
  1. [Abstract] Abstract: The central claim that the learned reflection intensity prior is 'general' and sufficient to resolve 'diverse image degradations' is load-bearing for the SOTA assertion, yet the prior is obtained by RPEN from training patches; no cross-degradation hold-out experiment or physics-based validation is described that would confirm generality to unseen blur, noise, or transmission variations.
  2. [Abstract] Abstract: The SOTA claim on real-world benchmarks is stated without reference to specific baselines, test-set sizes, error bars, or data-handling protocol; this omission prevents verification that the reported accuracy improvement is attributable to the prior rather than implementation details.
minor comments (1)
  1. [Title/Abstract] Title uses 'Patch Reflectance Prior' while the abstract consistently refers to 'reflection intensity prior'; aligning terminology would reduce reader confusion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful comments. We address the two major points on the abstract below, proposing targeted revisions to improve clarity without altering the core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the learned reflection intensity prior is 'general' and sufficient to resolve 'diverse image degradations' is load-bearing for the SOTA assertion, yet the prior is obtained by RPEN from training patches; no cross-degradation hold-out experiment or physics-based validation is described that would confirm generality to unseen blur, noise, or transmission variations.

    Authors: The prior is explicitly constructed to capture non-uniform reflection intensities via patch segmentation in RPEN, allowing it to adapt to the varied transmission and reflection effects present in real-world training data. Evaluations on real-world benchmarks already include images with mixed degradations, supporting the claim through empirical results rather than synthetic hold-outs. We agree the abstract would be strengthened by explicit qualification and will revise it to emphasize that generality is with respect to non-uniform intensity patterns observed in real data, while adding a short discussion paragraph on design assumptions. revision: partial

  2. Referee: [Abstract] Abstract: The SOTA claim on real-world benchmarks is stated without reference to specific baselines, test-set sizes, error bars, or data-handling protocol; this omission prevents verification that the reported accuracy improvement is attributable to the prior rather than implementation details.

    Authors: The abstract prioritizes brevity, but the full manuscript (Section 4 and tables) details comparisons to multiple published baselines on standard real-world test sets, with quantitative metrics. We will revise the abstract to name the primary competing methods and note the evaluation protocol, ensuring readers can immediately contextualize the SOTA claim. revision: yes

Circularity Check

0 steps flagged

No circularity: data-driven prior learned from training patches and evaluated on external benchmarks

full rationale

The paper introduces RPEN to learn a non-uniform reflection intensity prior by segmenting training images into regional patches, then feeds this learned prior into PRRN (a transformer U-Net) to perform reflection removal. This is a standard supervised learning pipeline with no mathematical derivation chain that reduces to its own inputs by construction. No equations or self-citations are presented that would make the central claim (effectiveness of the learned prior) equivalent to the training data or to a prior result by the same authors. The method is self-contained against external real-world benchmarks, with performance measured as SOTA accuracy rather than a forced statistical prediction of a fitted quantity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract; no explicit free parameters, axioms, or invented entities beyond standard neural network training are identifiable. The approach relies on learned network weights as implicit fitted parameters.

pith-pipeline@v0.9.0 · 5679 in / 1171 out tokens · 37662 ms · 2026-05-24T05:06:30.355180+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Single image reflection suppression

    Nikolaos Arvanitopoulos, Radhakrishna Achanta, and Sabine Susstrunk. Single image reflection suppression. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4498–4506, 2017

    work page 2017

  2. [2]

    Cold diffusion: Inverting arbitrary image transforms without noise

    Arpit Bansal, Eitan Borgnia, Hong-Min Chu, Jie S Li, Hamid Kazemi, Furong Huang, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Cold diffusion: Inverting arbitrary image transforms without noise. arXiv preprint arXiv:2208.09392, 2022

    work page arXiv 2022

  3. [3]

    Reverse attention for salient object detection

    Shuhan Chen, Xiuli Tan, Ben Wang, and Xuelong Hu. Reverse attention for salient object detection. In Proceedings of the European conference on computer vision (ECCV) , pages 234–250, 2018

    work page 2018

  4. [4]

    Imagenet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large- scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    work page 2009

  5. [5]

    Location-aware single image reflection removal

    Zheng Dong, Ke Xu, Yin Yang, Hujun Bao, Weiwei Xu, and Rynson WH Lau. Location-aware single image reflection removal. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5017–5026, 2021

    work page 2021

  6. [6]

    The pascal visual object classes (voc) challenge

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. International journal of computer vision , 88:303–338, 2010

    work page 2010

  7. [7]

    A generic deep architecture for single image reflection removal and image smoothing

    Qingnan Fan, Jiaolong Yang, Gang Hua, Baoquan Chen, and David Wipf. A generic deep architecture for single image reflection removal and image smoothing. In Proceedings of the IEEE International Conference on Computer Vision , pages 3238–3247, 2017

    work page 2017

  8. [8]

    Learning semantic associations for mirror detection

    Huankang Guan, Jiaying Lin, and Rynson WH Lau. Learning semantic associations for mirror detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5941–5950, 2022

    work page 2022

  9. [9]

    Long short-term memory

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural computation, 9(8):1735–1780, 1997

    work page 1997

  10. [10]

    Blurring diffusion models

    Emiel Hoogeboom and Tim Salimans. Blurring diffusion models. arXiv preprint arXiv:2209.05557, 2022

    work page arXiv 2022

  11. [11]

    Trash or treasure? an interactive dual-stream strategy for single image reflection separation

    Qiming Hu and Xiaojie Guo. Trash or treasure? an interactive dual-stream strategy for single image reflection separation. Advances in Neural Information Processing Systems , 34:24683– 24694, 2021

    work page 2021

  12. [12]

    Layercam: Exploring hierarchical class activation maps for localization

    Peng-Tao Jiang, Chang-Bin Zhang, Qibin Hou, Ming-Ming Cheng, and Yunchao Wei. Layercam: Exploring hierarchical class activation maps for localization. IEEE Transactions on Image Processing, 30:5875–5888, 2021

    work page 2021

  13. [13]

    Single image reflection removal with physically- based training images

    Soomin Kim, Yuchi Huo, and Sung-Eui Yoon. Single image reflection removal with physically- based training images. In CVPR, 2020

    work page 2020

  14. [14]

    A categorized reflection removal dataset with diverse real-world scenes

    Chenyang Lei, Xuhua Huang, Chenyang Qi, Yankun Zhao, Wenxiu Sun, Qiong Yan, and Qifeng Chen. A categorized reflection removal dataset with diverse real-world scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3040–3048, 2022

    work page 2022

  15. [15]

    Polarized reflection removal with perfect alignment in the wild

    Chenyang Lei, Xuhua Huang, Mengdi Zhang, Qiong Yan, Wenxiu Sun, and Qifeng Chen. Polarized reflection removal with perfect alignment in the wild. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1750–1758, 2020

    work page 2020

  16. [16]

    User assisted separation of reflections from a single image using a sparsity prior

    Anat Levin and Yair Weiss. User assisted separation of reflections from a single image using a sparsity prior. IEEE Transactions on Pattern Analysis and Machine Intelligence , 29(9):1647– 1654, 2007

    work page 2007

  17. [17]

    Learning to perceive transparency from the statistics of natural scenes

    Anat Levin, Assaf Zomet, and Yair Weiss. Learning to perceive transparency from the statistics of natural scenes. Advances in Neural Information Processing Systems , 15, 2002

    work page 2002

  18. [18]

    Single image reflection re- moval through cascaded refinement

    Chao Li, Yixiao Yang, Kun He, Stephen Lin, and John E Hopcroft. Single image reflection re- moval through cascaded refinement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3565–3574, 2020. 10

    work page 2020

  19. [19]

    Single image layer separation using relative smoothness

    Yu Li and Michael S Brown. Single image layer separation using relative smoothness. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 2752–2759, 2014

    work page 2014

  20. [20]

    Exploiting semantic relations for glass surface detection

    Jiaying Lin, Yuen-Hei Yeung, and Rynson Lau. Exploiting semantic relations for glass surface detection. Advances in Neural Information Processing Systems , 35:22490–22504, 2022

    work page 2022

  21. [21]

    Don’t hit me! glass detection in real-world scenes

    Haiyang Mei, Xin Yang, Yang Wang, Yuanyuan Liu, Shengfeng He, Qiang Zhang, Xiaopeng Wei, and Rynson WH Lau. Don’t hit me! glass detection in real-world scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3687–3696, 2020

    work page 2020

  22. [22]

    Image super-resolution via iterative refinement.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022

    Chitwan Saharia, Jonathan Ho, William Chan, Tim Salimans, David J Fleet, and Mohammad Norouzi. Image super-resolution via iterative refinement.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022

    work page 2022

  23. [23]

    Reflection removal using ghosting cues

    YiChang Shih, Dilip Krishnan, Fredo Durand, and William T Freeman. Reflection removal using ghosting cues. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3193–3201, 2015

    work page 2015

  24. [24]

    Region- aware reflection removal with unified content and gradient priors

    Renjie Wan, Boxin Shi, Ling-Yu Duan, Ah-Hwee Tan, Wen Gao, and Alex C Kot. Region- aware reflection removal with unified content and gradient priors. IEEE Transactions on Image Processing, 27(6):2927–2941, 2018

    work page 2018

  25. [25]

    Benchmarking single- image reflection removal algorithms

    Renjie Wan, Boxin Shi, Ling-Yu Duan, Ah-Hwee Tan, and Alex C Kot. Benchmarking single- image reflection removal algorithms. In Proceedings of the IEEE International Conference on Computer Vision, pages 3922–3930, 2017

    work page 2017

  26. [26]

    Depth of field guided reflection removal

    Renjie Wan, Boxin Shi, Tan Ah Hwee, and Alex C Kot. Depth of field guided reflection removal. In 2016 IEEE International Conference on Image Processing (ICIP) , pages 21–25. IEEE, 2016

    work page 2016

  27. [27]

    Corrn: Cooperative reflection removal network

    Renjie Wan, Boxin Shi, Haoliang Li, Ling-Yu Duan, Ah-Hwee Tan, and Alex C Kot. Corrn: Cooperative reflection removal network. IEEE transactions on pattern analysis and machine intelligence, 42(12):2969–2982, 2019

    work page 2019

  28. [28]

    Single image reflection removal exploiting misaligned training data and network enhancements

    Kaixuan Wei, Jiaolong Yang, Ying Fu, David Wipf, and Hua Huang. Single image reflection removal exploiting misaligned training data and network enhancements. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8178–8187, 2019

    work page 2019

  29. [29]

    Single image reflection removal exploiting misaligned training data and network enhancements

    Kaixuan Wei, Jiaolong Yang, Ying Fu, David Wipf, and Hua Huang. Single image reflection removal exploiting misaligned training data and network enhancements. In CVPR, 2019

    work page 2019

  30. [30]

    Single image reflection removal beyond linearity

    Qiang Wen, Yinjie Tan, Jing Qin, Wenxi Liu, Guoqiang Han, and Shengfeng He. Single image reflection removal beyond linearity. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3771–3779, 2019

    work page 2019

  31. [31]

    Seeing deeply and bidirectionally: a deep learning approach for single image reflection removal

    Jie Yang, Dong Gong, Lingqiao Liu, and Qinfeng Shi. Seeing deeply and bidirectionally: a deep learning approach for single image reflection removal. In ECCV, 2018

    work page 2018

  32. [32]

    Unsupervised deraining: Where contrastive learning meets self-similarity

    Yuntong Ye, Changfeng Yu, Yi Chang, Lin Zhu, Xi-le Zhao, Luxin Yan, and Yonghong Tian. Unsupervised deraining: Where contrastive learning meets self-similarity. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5821–5830, 2022

    work page 2022

  33. [33]

    Depth-guided camouflaged object detection

    Jing Zhang, Yunqiu Lv, Mochu Xiang, Aixuan Li, Yuchao Dai, and Yiran Zhong. Depth-guided camouflaged object detection. arXiv preprint arXiv:2106.13217, 2021

    work page arXiv 2021

  34. [34]

    Single image reflection separation with perceptual losses

    Xuaner Zhang, Ren Ng, and Qifeng Chen. Single image reflection separation with perceptual losses. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4786–4794, 2018

    work page 2018

  35. [35]

    Single image reflection removal with absorption effect

    Qian Zheng, Boxin Shi, Jinnan Chen, Xudong Jiang, Ling-Yu Duan, and Alex C Kot. Single image reflection removal with absorption effect. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13395–13404, 2021. 11

    work page 2021