在本篇論文中，我們針對顏色恆常性(color constancy)的問題提出一個新的端對端(end-to-end)非監督式(unsupervised)的解決方法。我們以生成對抗網路(generative adversarial network)的架構構築學習模型。和現今主流的模型比較之下，其產生出來的結果是有潛力的。

In this paper, we present a novel end-to-end unsupervised solution for solving color constancy problems. We build our model with a
generative adversarial network and show that it generates promising results compared to the current state-of-art model.

中文摘要i
Abstract ii
Acknowledgements iii
List of Figures vi
List of Tables vii
1 Introduction 1
2 RelatedWork 3
2.1 Color constancy . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Convolutional neural network . . . . . . . . . . . . . . . 4
2.3 Supervised and Unsupervised learning . . . . . . . . . . 5
2.4 Generative model . . . . . . . . . . . . . . . . . . . . . 6
2.4.1 GANs . . . . . . . . . . . . . . . . . . . . . . . 6
2.4.2 Equilibrium of training . . . . . . . . . . . . . . 7
3 Auto White Balance and Unsupervised Learning 9

3.1 Data Preprocessing . . . . . . . . . . . . . . . . . . . . 9
3.2 Model Architecture . . . . . . . . . . . . . . . . . . . . 12
3.3 Self-Attention Layer . . . . . . . . . . . . . . . . . . . 16
3.4 Objective Function . . . . . . . . . . . . . . . . . . . . 16
3.5 Adaptive Threshold . . . . . . . . . . . . . . . . . . . . 17
4 Experiments 18
4.1 Training . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Training stability . . . . . . . . . . . . . . . . . . . . . 28
5 Conclusions 29
Bibliography 30

[1] J. von Kries, “Chromatic adaptation, festschrift der alberchtludwiguniversität,” 1902.
[2] Y. Qian, K. Chen, J. Nikkanen, J.-K. Kämäräinen, and J. Matas,
“Recurrent color constancy,” ICCV, 2017.
[3] O. Sidorov, “Conditional gans for multi-illuminant color constancy: Revolution or yet another approach?” CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019.
[4] S. Bianco and C. Cusano, “Quasi-unsupervised color constancy,”
CVPR, 2019.
[5] J. Qiu, H. Xu, and Z. Ye, “Color constancy by reweighting image feature maps,” IEEE Transactions on Image Processing, 2020.
[6] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-
Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” NIPS, 2014.
[7] M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein gan,” arXiv preprint arXiv:1701.07875, 2017.
[8] I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. Courville, “Improved training of wasserstein gans,” NIPS, 2017.
[9] J. Zhao, M. Mathieu, and Y. LeCun, “Energy-based generative adversarial network,” ICLR, 2017.
[10] D. Berthelot, T. Schumm, and L. Metz, “Began: Boundary equilibrium generative adversarial networks,” arXiv:1703.10717, 2017.
[11] C.-C. Chang, C. H. Lin, C.-R. Lee, D.-C. Juan, W. Wei, and H.-T. Chen, “Escaping from collapsing modes in a constrained space,”
ECCV, 2018.
[12] P. Das, A. S. Baslamisli, Y. Liu, S. Karaoglu, and T. Gevers, “Color constancy by gans: An experimental survey,” arXiv:1812.03085,
2018.
[13] Y. Hu, B. Wang, and S. Lin, “Fc 4: Fully convolutional color constancy with confidence-weighted pooling,” in Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 4085–4094.
[14] I. Tolstikhin, O. Bousquet, S. Gelly, and B. Schoelkopf, “Wasserstein auto-encoders,” arXiv:1711.01558, 2017.
[15] Z. Zhang, R. Zhang, Z. Li, Y. Bengio, and L. Paull, “Perceptual generative autoencoders,” ICML, 2020.
[16] A. Makhzani, J. Shlens, N. Jaitly, I. Goodfellow, and B. Frey, “Adversarial autoencoders,” ICLR, 2016.
[17] Y. Pu, Z. Gan, R. Henao, X. Yuan, A. Stevens, and L. Carin, “Variational autoencoder for deep learning of images, labels and captions,” NIPS, 2016.
[18] Y. Bengio, L. Yao, G. Alain, and P. Vincent, “Generalized denoising auto-encoders as generative models,” NIPS, 2013.
[19] P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol, “Extracting and composing robust features with denoising autoencoders,”
ICML, 2008.
[20] D. Ji, J. Kwon, M. McFarland, and S. Savarese, “Deep view morphing,” IEEE Conference on Computer Vision and Pattern Recognition, 2017.
[21] H. Zhang, I. Goodfellow, D. Metaxas, and A. Odena, Selfattention generative adversarial networks,” arXiv:1805.08318v2, 2019.
[22] T. Miyato, T. Kataoka, M. Koyama, and Y. Yoshida, “Spectral normalization for generative adversarial networks,” ICLR, 2018.
[23] P. V. Gehler, C. Rother, A. Blake, T. Minka, and T. Sharp, “Bayesian color constancy revisited,” CVPR, 2008.
[24] L. Shi and B. Funt., “Re-processed version of the gehler
color constancy dataset of 568 images.” [Online]. Available:
http://www.cs.sfu.ca/ colour/data/
[25] G. Hemrit, G. D. Finlayson, A. Gijsenij, P. Gehler, S. Bianco, B. Funt, M. Drew, and L. Shi, “Rehabilitating the colorchecker
dataset for illuminant estimation,” arXiv:1805.12262, 2018.
[26] N. Bani´c, K. Košˇcevi´c, and S. Lonˇcari´c, “Unsupervised learning for color constancy,” arXiv:1712.00436, 2017.
[27] M. Heusel, H. Ramsauer, T. Unterthiner, and B. Nessler, “Gans
trained by a two time-scale update rule converge to a local nash
equilibrium,” NIPS, 2017.
[28] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv:1412.6980, 2014.
[29] G. Buchsbaum, “A spatial processor model for object colour perception,” Journal of the Franklin Institute, 1980.
[30] D. H. Brainard and B. A. Wandell, “Analysis of the retinex theory of color vision,” JOSA A, 1986.
[31] G. D. Finlayson and E. Trezzi, “Shades of gray and colour constancy,” Color Imaging Conference, 2004.
[32] T. G. J. van de Weijer and A. Gijsenij, “Edge-based color constancy,” TIP, 2007.