DesnowOracle: Quantifying Mask-Oracle Gap Recovery by a Training-Free Snow Prior in Compact Image Desnowing
DOI:
https://doi.org/10.31436/iiumej.v27i3.4662Keywords:
Image Desnowing, Adverse Weather Perception, Spatial Conditioning, Snow Severity Prior, Compact RestorationAbstract
Snow occludes visual information needed by outdoor perception systems. This study evaluates whether a deterministic, training-free spatial snow-response map improves compact image desnowing beyond the effect of its conditioning branch. All conditions use a fixed approximately 2-million-parameter backbone and matched controls. Under the original 2,500-image, 100-epoch protocol, estimated-map SSM improves mean PSNR by 1.004 dB over Base and by 0.490 and 0.517 dB over zero- and shuffled-input SSM. A paired analysis averaging predictions across three seeds gives median gains of 0.387 and 0.420 dB, respectively. Under an extended protocol that also changes training-set size and learning-rate scheduling, this ordering reverses: estimated-map SSM trails zero and shuffled controls by 0.109 and 0.160 dB in mean PSNR. Thus, the experiments demonstrate protocol sensitivity, not an isolated convergence effect. A mask oracle remains an architecture-specific, synthesis-privileged diagnostic reference; the estimated map recovers 12.4% of its short-protocol PSNR headroom. Results on 1,329 unpaired real images are mixed and indirect: no-reference indices favor the estimated map, whereas unlabelled detector counts do not distinguish its content from the controls. These findings show why spatial-prior claims require matched controls, correct repeated-measures analysis, and protocol checks.
ABSTRAK: Salji menghalang maklumat visual yang diperlukan oleh sistem persepsi luar bangunan. Kajian ini menilai sama ada peta tindak balas salji spatial yang deterministik dan tanpa latihan meningkatkan penyahsaljian imej padat melebihi kesan cabang pengkondisian. Semua keadaan menggunakan rangkaian asas tetap dengan kira-kira 2 juta parameter dan kawalan yang sepadan. Di bawah protokol asal 2,500 imej dan 100 epok, SSM dengan peta anggaran meningkatkan purata PSNR sebanyak 1.004 dB berbanding Base dan sebanyak 0.490 serta 0.517 dB berbanding SSM input sifar dan teracak. Purata analisis ramalan berpasangan merentas tiga benih memberikan peningkatan median masing-masing sebanyak 0.387 dan 0.420 dB. Di bawah protokol lanjutan yang turut mengubah saiz set latihan dan penjadualan kadar pembelajaran, susunan ini berbalik kepada SSM dengan peta anggaran ketinggalan berbanding kawalan sifar dan teracak sebanyak 0.109 dan 0.160 dB dalam purata PSNR. Oleh itu, eksperimen menunjukkan kepekaan terhadap protokol, bukan kesan penumpuan terasing. Topeng Oracle kekal sebagai rujukan diagnostik khusus seni bina dan keistimewaan sintesis; peta anggaran memulihkan 12.4% ruang PSNR protokol pendek. Dapatan daripada 1,329 imej sebenar tanpa pasangan adalah bercampur dan tidak langsung: indeks tanpa rujukan memihak kepada peta anggaran, manakala kiraan pengesan tanpa label tidak membezakan kandungannya daripada kawalan. Dapatan ini menunjukkan bahawa tuntutan sebelum spatial memerlukan kawalan sepadan, analisis ukuran berulang yang betul, dan semakan protokol.
Downloads
Metrics
References
J. Wang, M. Xu, H. Xue, Z. Huo, and F. Luo, "Joint image restoration for object detection in snowy weather," IET Computer Vision, vol. 18, no. 6, pp. 759-771, 2024.
Y.-F. Liu, D.-W. Jaw, S.-C. Huang, and J.-N. Hwang, "DesnowNet: Context-aware deep network for snow removal," IEEE Transactions on Image Processing, vol. 27, no. 6, pp. 3064-3073, 2018.
W.-T. Chen, Z.-K. Huang, C.-C. Tsai, H.-H. Yang, J.-J. Ding, and S.-Y. Kuo, "Learning multiple adverse weather removal via two-stage knowledge learning and multi-contrastive regularization: Toward a unified model," in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022: IEEE, pp. 17632-17641.
W.-T. Chen, H.-Y. Fang, J.-J. Ding, C.-C. Tsai, and S.-Y. Kuo, "JSTASR: Joint size and transparency-aware snow removal algorithm based on modified partial convolution and veiling effect removal," in European Conference on Computer Vision, 2020: Springer, pp. 754-770.
W.-T. Chen et al., "All snow removed: Single image desnowing algorithm using hierarchical dual-tree complex wavelet representation and contradict channel loss," in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 4196-4205.
K. Zhang, R. Li, Y. Yu, W. Luo, and C. Li, "Deep dense multi-scale network for snow removal using semantic and depth priors," IEEE Transactions on Image Processing, vol. 30, pp. 7419-7431, 2021.
S. A. Agyemang, H. Shi, X. Nie, and N. Y. Asabere, "An integrated multi-scale context-aware network for efficient desnowing," Engineering Applications of Artificial Intelligence, vol. 151, p. 110769, 2025.
Z. Chen, Y. Sun, X. Bi, and J. Yue, "Lightweight image de-snowing: A better trade-off between network capacity and performance," Neural Networks, vol. 165, pp. 896-908, 2023.
J. Lin, N. Jiang, Z. Zhang, W. Chen, and T. Zhao, "LMQFormer: A laplace-prior-guided mask query transformer for lightweight snow removal," IEEE Transactions on Circuits and Systems for Video Technology, vol. 33, no. 11, pp. 6225-6235, 2023.
X. Guo, X. Wang, X. Fu, and Z.-J. Zha, "Deep unfolding network for image desnowing with snow shape prior," IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 5, pp. 4740-4752, 2025.
S. Chen, T. Ye, Y. Liu, E. Chen, J. Shi, and J. Zhou, "SnowFormer: Scale-aware transformer via context interaction for single image desnowing," arXiv preprint arXiv:2208.09703, 2022.
Y. Zhang, "BAT-Net: Bidirectional Attention Transformer Network for Joint Single-Image Desnowing and Snow Mask Prediction," Information, vol. 16, no. 11, p. 966, 2025.
G. Wang, Y. Zhou, F. Shi, and Z. Jia, "SnowMamba: Achieving more precise snow removal with Mamba," Applied Sciences, vol. 15, no. 10, p. 5404, 2025.
J. Lai et al., "Snowmaster: Comprehensive real-world image desnowing via mllm with multi-model feedback optimization," in 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025: IEEE, pp. 4302-4312.
Q. Guan, Q. Yang, X. Chen, T. Song, G. Jin, and J. Jin, "Weatherbench: A real-world benchmark dataset for all-in-one adverse weather image restoration," in Proceedings of the 33rd ACM international conference on multimedia, 2025, pp. 12607-12613.
J. Jiang, Z. Zuo, G. Wu, K. Jiang, and X. Liu, "A survey on all-in-one image restoration: Taxonomy, evaluation and future trends," IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025.
B. Li, X. Liu, P. Hu, Z. Wu, J. Lv, and X. Peng, "All-in-one image restoration for unknown corruption," in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022: IEEE, pp. 17431-17441.
J. M. J. Valanarasu, R. Yasarla, and V. M. Patel, "Transweather: Transformer-based restoration of images degraded by adverse weather conditions," in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022: IEEE, pp. 2343-2353.
C. Li, F. Sun, H. Zhou, Y. Xie, Z. Li, and L. Zhu, "Multi-weather restoration: An efficient prompt-guided convolution architecture," IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 2, pp. 1436-1450, 2024.
Y. Wen, T. Gao, Z. Li, J. Zhang, K. Zhang, and T. Chen, "All-in-one weather-degraded image restoration via adaptive degradation-aware self-prompting model," IEEE Transactions on Multimedia, vol. 27, pp. 3343-3355, 2025.
D. Xie, X. Hu, Y. Zhou, and S. Duan, "All-in-one adverse weather removal via dual state space-based diffusion model with degradation-aware guidance," Pattern Recognition, vol. 171, p. 112081, 2026.
Y. Cui, S. W. Zamir, S. Khan, A. Knoll, M. Shah, and F. Khan, "Adair: Adaptive all-in-one image restoration via frequency mining and modulation," in Proceedings of the 13th International Conference on Learning Representations (ICLR), 2025.
E. Perez, F. Strub, H. De Vries, V. Dumoulin, and A. Courville, "FiLM: Visual reasoning with a general conditioning layer," in Proceedings of the AAAI Conference on Artificial Intelligence, 2018, vol. 32, no. 1.
X. Wang, K. Yu, C. Dong, and C. C. Loy, "Recovering realistic texture in image super-resolution by deep spatial feature transform," in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: IEEE, pp. 606-615.
O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional networks for biomedical image segmentation," in International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015: Springer, pp. 234-241.
T. S. Gunawan and M. Kartiwi, "Stability-Aware Physics-Prior Conditioning for Lightweight Image Dehazing," presented at the 2026 IEEE 12th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA 2026), Kuching, Sarawak, Malaysia, 28-30 July 2026, 2026.
I. Loshchilov and F. Hutter, "Decoupled weight decay regularization," presented at the Proceedings of the 13th International Conference on Learning Representations (ICLR), 2019.
Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, "Image quality assessment: from error visibility to structural similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, 2004.
R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, "The unreasonable effectiveness of deep features as a perceptual metric," in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: IEEE, pp. 586-595.
A. Mittal, R. Soundararajan, and A. C. Bovik, "Making a “completely blind” image quality analyzer," IEEE Signal Processing Letters, vol. 20, no. 3, pp. 209-212, 2013.
A. Mittal, A. K. Moorthy, and A. C. Bovik, "No-reference image quality assessment in the spatial domain," IEEE Transactions on Image Processing, vol. 21, no. 12, pp. 4695-4708, 2012.
N. Venkatanath, D. Praneeth, S. C. Sumohana, and S. M. Swarup, "Blind image quality evaluation using perception based features," in 2015 Twenty First National Conference on Communications (NCC), 2015: IEEE, pp. 1-6.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 IIUM Press

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.








