pith. sign in

arxiv: 1906.09433 · v1 · pith:LDUOV4H3new · submitted 2019-06-22 · 💻 cs.CV

Deep Single Image Deraining Via Estimating Transmission and Atmospheric Light in rainy Scenes

Yinglong Wang , Qinfeng Shi , Ehsan Abbasnejad , Chao Ma , Xiaoping Ma , Bing Zeng This is my paper

Pith reviewed 2026-05-25 18:22 UTC · model grok-4.3

classification 💻 cs.CV
keywords single image derainingatmospheric scattering modeltransmission estimationatmospheric light estimationdeep neural networkimage restoration
0
0 comments X

The pith

A deep network estimates transmission and atmospheric light to remove rain using the atmospheric scattering model.

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

The paper establishes that modeling rainy image formation with the atmospheric scattering model enables learning of transmission and atmospheric light to achieve superior deraining. Previous deep methods either directly estimate the background or subtract a learned rain residual. This new approach also yields estimates of scene parameters. A sympathetic reader would care if it leads to more accurate rain removal and additional information about the scene.

Core claim

By learning transmission and atmospheric light in rainy scenes with the image degradation model and the proposed networks, the method obtains better de-raining results than selected state-of-the-art works. The approach uses a robust evaluation of global atmospheric light as ground truth, a triangle-shaped network to learn atmospheric light, and ShuffleNet Units in the transmission network.

What carries the argument

The image degradation model from atmospheric scattering principles, which models rainy image formation as a function of transmission and atmospheric light.

If this is right

  • The de-raining image is obtained by applying the image degradation model with the estimated transmission and atmospheric light.
  • A triangle-shaped network structure learns atmospheric light for every rainy image, using the robust estimate as ground truth.
  • ShuffleNet Units enable efficient learning of the transmission map in the transmission network.
  • Subjective and objective comparisons show the method outperforms selected state-of-the-art deraining works.

Where Pith is reading between the lines

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

  • The same scattering model and joint estimation strategy could be tested on other degradations that share similar physical formation processes.
  • The recovered transmission maps might supply auxiliary depth-like information usable in downstream vision tasks under weather conditions.
  • The staged training of atmospheric light followed by fine-tuning with the transmission network points to potential benefits from end-to-end joint optimization in related restoration problems.

Load-bearing premise

The atmospheric scattering model derived from haze physics accurately describes the formation of rainy images.

What would settle it

A controlled test on synthetic rainy images with known ground-truth transmission and atmospheric light values, checking whether the network estimates match those values closely enough to produce a clear recovered image.

Figures

Figures reproduced from arXiv: 1906.09433 by Bing Zeng, Chao Ma, Ehsan Abbasnejad, Qinfeng Shi, Xiaoping Ma, Yinglong Wang.

Figure 1
Figure 1. Figure 1: An example of a real-world rainy image, its de-raining [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Some results of using dark channel prior to a rainy image. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: This figure shows our whole network structure. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: This figure shows our revised ShuffleNet Units. Shuf [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: In the first column, the red point is the pixel whose value [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: This figure shows the rain-removed results on synthetic rainy images. Here, we show two rainy images which have wide rain [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: This figure shows rain-removed results on some real-world rainy images. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The first line is a synthetic rainy image and the second is [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: This figure shows some dehazing results and obtained [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
read the original abstract

Rain removal in images/videos is still an important task in computer vision field and attracting attentions of more and more people. Traditional methods always utilize some incomplete priors or filters (e.g. guided filter) to remove rain effect. Deep learning gives more probabilities to better solve this task. However, they remove rain either by evaluating background from rainy image directly or learning a rain residual first then subtracting the residual to obtain a clear background. No other models are used in deep learning based de-raining methods to remove rain and obtain other information about rainy scenes. In this paper, we utilize an extensively-used image degradation model which is derived from atmospheric scattering principles to model the formation of rainy images and try to learn the transmission, atmospheric light in rainy scenes and remove rain further. To reach this goal, we propose a robust evaluation method of global atmospheric light in a rainy scene. Instead of using the estimated atmospheric light directly to learn a network to calculate transmission, we utilize it as ground truth and design a simple but novel triangle-shaped network structure to learn atmospheric light for every rainy image, then fine-tune the network to obtain a better estimation of atmospheric light during the training of transmission network. Furthermore, more efficient ShuffleNet Units are utilized in transmission network to learn transmission map and the de-raining image is then obtained by the image degradation model. By subjective and objective comparisons, our method outperforms the selected state-of-the-art works.

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 paper claims that modeling rainy image formation with the atmospheric scattering equation allows a deep network to jointly estimate per-image atmospheric light A (via a triangle-shaped architecture) and transmission map t (via ShuffleNet units), after which the clear image is recovered by the degradation model; the resulting method is reported to outperform selected prior deraining works on both subjective and objective metrics.

Significance. If the haze-derived model is shown to be an adequate approximation for rain and the estimated maps are accurate, the approach supplies a physically interpretable alternative to direct residual subtraction and yields auxiliary outputs (t and A) that may be useful for downstream tasks. The adoption of ShuffleNet units for the transmission branch is a concrete efficiency contribution.

major comments (1)
  1. [Abstract and §3] Abstract and §3 (modeling paragraph): the central claim that the atmospheric scattering model I = J·t + A(1-t) accurately describes rainy image formation is load-bearing for the reconstruction step and for interpreting the learned maps as physical quantities, yet the manuscript supplies no quantitative check (e.g., residual error on real paired data or comparison with streak-specific forward models) that would confirm the approximation suffices for the claimed performance gains; rain physics (additive discrete streaks, depth-dependent density) differ from volumetric scattering.
minor comments (1)
  1. [Abstract] The abstract refers to 'objective comparisons' and 'selected state-of-the-art works' without naming the metrics or baselines in the summary paragraph; this should be stated explicitly for clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the single major comment below and will revise the manuscript accordingly to strengthen the modeling justification.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (modeling paragraph): the central claim that the atmospheric scattering model I = J·t + A(1-t) accurately describes rainy image formation is load-bearing for the reconstruction step and for interpreting the learned maps as physical quantities, yet the manuscript supplies no quantitative check (e.g., residual error on real paired data or comparison with streak-specific forward models) that would confirm the approximation suffices for the claimed performance gains; rain physics (additive discrete streaks, depth-dependent density) differ from volumetric scattering.

    Authors: We acknowledge that the atmospheric scattering model is an approximation for rain formation, which involves discrete additive streaks and depth-dependent density rather than pure volumetric scattering. The manuscript relies on empirical performance gains rather than a direct model-fit validation. To address this, the revised manuscript will add a quantitative analysis subsection (likely in §4 or a new §3.4) that reports reconstruction residuals on synthetic data generated from both the scattering model and streak-specific forward models, plus any available real paired examples. This will explicitly test whether the approximation suffices for the observed improvements while noting its limitations. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation applies external model assumption to new task

full rationale

The paper adopts the atmospheric scattering degradation model I = J·t + A(1-t) as a modeling premise for rainy images, then trains separate networks to estimate t and A before reconstructing J via the same equation. This is an assumption about model applicability rather than a self-referential definition or fitted input renamed as prediction. No equations reduce claimed outputs to inputs by construction, no load-bearing uniqueness theorems are imported via self-citation, and performance claims rest on external subjective/objective comparisons to other methods. The derivation chain is therefore self-contained against benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central modeling choice is the applicability of the atmospheric scattering equation to rain; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption The image degradation model derived from atmospheric scattering principles models the formation of rainy images.
    Explicitly invoked in the abstract as the basis for estimating transmission and atmospheric light.

pith-pipeline@v0.9.0 · 5798 in / 1097 out tokens · 22536 ms · 2026-05-25T18:22:43.210830+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

39 extracted references · 39 canonical work pages · 4 internal anchors

  1. [1]

    C. H. Bahnsen and T. B. Moeslund. Rain removal in traffic surveillance: does it matter? IEEE Transac- tions on Intelligent Transportation Systems , pages 1– 18, Oct. 2018. 1

    work page 2018

  2. [2]

    Barnum, T

    P. Barnum, T. Kanade, and S. Narasimhan. Spatio- temporal frequency analysis for removing rain and snow from videos. In International Workshop on Pho- tometric Analysis For Computer Vision (PACV 2007), pages 8–p, Rio de Janeiro, Brazil,, Oct. 2007. INRIA. 2

    work page 2007

  3. [3]

    P. C. Barnum, S. Narasimhan, and T. Kanade. Anal- ysis of rain and snow in frequency space. Interna- tional Journal of Computer Vision , 86(2):256–274, Jan. 2010. 2

    work page 2010

  4. [4]

    Bossu, N

    J. Bossu, N. Hautiere, and J. P. Tarel. Rain or snow detection in image sequences through use of a his- togram of orientation of streaks.International Journal of Computer Vision, 93(3):348–367, July 2011. 2

    work page 2011

  5. [5]

    Brewer and N

    N. Brewer and N. Liu. Using the shape characteris- tics of rain to identify and remove rain from video. In Joint IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition, volume 5342, pages 451–458, Olando, USA, Dec. 2008. 2

    work page 2008

  6. [6]

    D. Y . Chen, C. C. Chen, and L. W. Kang. Visual depth guided color image rain streaks removal using sparse coding. IEEE Transactions on Circuits and Systems for Video Technology, 24(8):1430–1455, Aug. 2014. 2

    work page 2014

  7. [7]

    Y . L. Chen and C. T. Hsu. A generalized low-rank ap- pearance model for spatio-temporally correlated rain streaks. In IEEE International Conference on Com- puter Vision (ICCV 2013), pages 1968–1975, Sydney, Australia, Dec. 2013. IEEE. 3

    work page 2013

  8. [8]

    F. Chollet. Xception: deep learning with depthwise separable convolutions. arXiv:1610.02357, 2016. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  9. [9]

    X. H. Ding, L. Q. Chen, X. H. Zheng, Y . Huang, and D. L. Zeng. Single image rain and snow removal via guided l0 smoothing filter. Multimedia Tools and Ap- plications, 24(8):1–16, Jun. 2015. 3

    work page 2015

  10. [10]

    R. Fattal. Single image dehazing. ACM Transactions on Graphics, (3), Aug. 2008. 2, 3, 4

    work page 2008

  11. [11]

    X. Fu, J. Huang, X. Ding, Y . Liao, and J. Paisley. Clearing the skies: a deep network architecture for single-image rain removal. IEEE Transactions on Im- age Processing, 26(6):2944–2956, July 2017. 2, 3

    work page 2017

  12. [12]

    X. Fu, J. Huang, D. Zeng, Y . Huang, X. Ding, and J. Paisley. Removing rain from single images via a deep detail network. In IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR-2017) , pages 1715–1723, Honolulu, HI, USA, July 2017. IEEE. 2, 3, 6, 7, 8

    work page 2017

  13. [13]

    Y . H. Fu, L. W. Kang, C. W. Lin, and C. T. Hsu. Single-frame-based rain removal via image decompo- sition. In IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP 2011) , pages 1453–1456, Prague, Czech Republic, May

    work page 2011

  14. [14]

    K. He, J. Sun, and X. Tang. Single image haze removal using dark channel prior. IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 33(12):2341– 2353, Dec. 2011. 3, 4

    work page 2011

  15. [15]

    K. He, J. Sun, and X. Tang. Guided image filtering. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(6):1397–1409, June 2013. 3

    work page 2013

  16. [16]

    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- 2017). IEEE, July 2015. 3

    work page 2017

  17. [17]

    D. A. Huang, L. W. Kang, Y . C. F. Wang, and C. W. Lin. Self-learning based image decomposition with applications to single image denoising. IEEE Trans- actions on Multimedia, 16(1):83–93, Jan. 2014. 2

    work page 2014

  18. [18]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    S. Ioffe and C. Szegedy. Batch normalization: accler- ating deep network training by reducing internal co- variate shift. arXiv:1502.03167, 2015. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  19. [19]

    L. W. Kang, C. W. Lin, and Y . H. Fu. Automatic single-image-based rain streaks removal via image de- composition. IEEE Transactions on Image Process- ing, 21(4):1742–1755, Apr. 2012. 2

    work page 2012

  20. [20]

    Koschmeeder

    H. Koschmeeder. Theorie der horizontalen sichtweite. Beitr. Phys. Freien Atm, 12:171–181, 1924. 3, 4, 5

    work page 1924

  21. [21]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton. Im- agenet classification with deep convolutional neural networks. In Conference on Neural Information Pro- cessing Systems (NIPS-2012), 2012. 5

    work page 2012

  22. [22]

    G. Li, X. He, W. Zhang, H. Chang, L. Dong, and L. Lin. Non-locally enhanced encoder-decoder net- work for single image de-raining. In ACM Multime- dia (MM-2018), Seoul, Republic of Korea, Oct. 2018. ACM. 2, 3, 6 9

    work page 2018

  23. [23]

    R. Li, L. F. Cheong, and R. T. Tan. Single image de- raing using scale-aware multi-stage recurrent network. arXiv:1712.06830, 2017. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  24. [24]

    S. Li, W. Ren, J. Zhang, J. Yu, and X. Guo. Fast single image rain removal via a deep decomposition- composition network. arXiv:1804.02688, 2018. 7

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  25. [25]

    X. Li, J. Wu, Z. Lin, H. Liu, and H. Zha. Recur- rent squeeze-and-excitation context aggregation net for single image deraining. In European Conference on Computer Vision (ECCV-2018), Munich, Germany, Sep. 2018. 6, 7

    work page 2018

  26. [26]

    Y . Li, R. T. Tan, X. Guo, J. Lu, and M. S. Brown. Rain streak removal using layer priors. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR 2016), pages 2736–2744, Las Vegas, Nevada, USA, June 2016. 2, 3

    work page 2016

  27. [27]

    Y . Luo, X. Yong, and J. Hui. Removing rain from a single image via discriminative sparse coding. In IEEE International Conference on Computer Vision (ICCV 2015), pages 3397–3405, Boston, MA, USA, Dec. 2015. 2

    work page 2015

  28. [28]

    Mairal, F

    J. Mairal, F. Bach, J. Ponce, and G. Sapiro. Online learning for matrix factorization and sparse coding. Journal of Machine Learning Research, pages 19–60, Mar. 2010. 2

    work page 2010

  29. [29]

    Mallat and Z

    S. Mallat and Z. Zhang. Matching pursuit with time- frequency dictionaries. IEEE Transactions on Signal Processing, (12), Dec. 1993. 2

    work page 1993

  30. [30]

    S. G. Narasimhan and S. K. Nayar. Vision and the atmosphere. International journal of computer vision, (3):233–254, Dec. 2002. 3, 5

    work page 2002

  31. [31]

    S. K. Nayar and S. G. Narasimhan. Vision in bad weather. In IEEE International Conference on Com- puter Vision (ICCV 1999) , Kerkyra, Greece, Greece, Sep. 1999. IEEE. 3, 5

    work page 1999

  32. [32]

    R. T. Tan. Visibility in bad weather from a single image. In IEEE International Conference on Com- puter Vision and Pattern Recognition (CVPR 2008) , Anchorage, AK, USA, July 2008. IEEE. 3, 4

    work page 2008

  33. [33]

    Y . Wang, S. Liu, C. Chen, and B. Zeng. A hierar- chical approach for rain or snow removing in a single color image. IEEE Transactions on Image Processing, 26(8):3936–3950, August 2017. 2

    work page 2017

  34. [34]

    Y . L. Wang, C. Chen, S. Y . Zhu, and B. Zeng. A framework of single-image deraining method based on analysis of rain characteristics. In IEEE International Conference on Image Processing (ICIP 2016) , pages 4087 – 4091, Phoenix, USA, Sep. 2016. IEEE. 2

    work page 2016

  35. [35]

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simon- celli. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 13(4):600–612, April 2004. 6

    work page 2004

  36. [36]

    J. Xu, W. Zhao, P. Liu, and X. Tang. An improved guidance image based method to remove rain and snow in a single image. Computer and Information Science, 5(3):49–55, May 2012. 3

    work page 2012

  37. [37]

    W. Yang, R. Tan, J. Feng, J. Liu, Z. Guo, and S. Yan. Deep joint rain detection and removal from a single image. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR-2017), pages 1685–1694, Honolulu, HI, USA, July 2017. IEEE. 2, 3, 4, 6, 7

    work page 2017

  38. [38]

    Zhang and V

    H. Zhang and V . Patel. Density-aware single image de- raining using a multi-stream dense network. In IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR-2018), pages 1685–1694, Salt Lake City, UT, July 2018. IEEE. 2, 3, 6, 7, 8

    work page 2018

  39. [39]

    Zhang, X

    X. Zhang, X. Zhou, M. Lin, and J. Sun. Shufflenet: a extremely efficient convolutional neural network for moble devices. In IEEE Conference on Computer Vi- sion and Pattern Recognition (CVPR 2018), Salt Lake City, USA, Jun. 2018. IEEE. 5, 6 10

    work page 2018