車牌識別 LPR 的深度學習方法通常建立在深度卷積神經網絡上。在訓練
集上進行訓練，然後在測試集上進行測試並進行性能評估。訓練集和測試集皆
從真實場景中收集。這樣的收集方式可能會引發隱私權益問題。因此我們開發
了一個生成器來製作人造車牌，這對訓練 LPR 網絡很有幫助。生成器建立在最
先進的樣式傳輸網絡 StyleGAN2 上，可以生成各種各樣的人工車照。為了加
強大角度的車牌辨識效能，我們另外收集了一個擁有大量高解析度且含有豐富
大角度車牌的資料集 AVL-LP。為了在人工車牌上生成所需的字符，我們的架
構參考了條件設置，以允許將所需字符作為標籤輸入，以生成指定字符的人造
車牌。實驗證實，生成的車牌可以在各種方向和光照條件下出現，並具有很好
的字符多樣性。對比研究證明，基於生成車牌的訓練比基於真實車牌的訓練可
以產生更好的 LPR 性能。

The deep learning method of license plate recognition LPR is usually based on
deep convolutional neural networks. Train on the training set, then test and evaluate the performance on the test set. Both training set and test set are collected from real scenes. Such collection methods may raise privacy issues. So we developed a generator to make artificial license plates, which is very helpful for training LPR networks. The generator is built on the most advanced style transmission network StyleGAN2, which can generate a variety of artificial car photos. In order to enhance the performance of large-angle license plate recognition, we have also collected a large amount of high-resolution and rich large-angle license plate data set AVL-LP. In addition, in order to generate the required characters on the artificial license plate, our architecture refers to the condition setting to allow the required characters to be input as the label to generate the artificial license plate with the specified characters. Experiments show that the generated license plates can appear in various directions and lighting conditions, and have good character diversity. Comparative studies have shown that training based on generating license plates can produce better LPR performance than training based on real license plates.

[1] Abbas M. Al-Ghaili, Syamsiah Mashohor, Abdul Rahman Ramli, and Alyani Ismail. Vertical-edge-based car-license plate detection method. IEEE Transactions on Vehicular Technology, 62(1):26–38, 2013.
[2] Martin Arjovsky, Soumith Chintala, and L ́eon Bottou. Wasserstein gan, 2017.
[3] G. J. et al. ultralytics / yolov5: v3.0. Aug. 2020. [Online].https://doi.org/10.5281/zenodo.3983579.
[4] K. He, G. Gkioxari, P. Dollar, and R. Girshick. Mask r-cnn. volume 2017-October, pages 2980–2988, 2017. citedBy 2400.
[5] G.-S. Hsu, J.-C. Chen, and Y.-Z. Chung. Application-oriented license plate recognition, 2013. IEEE transactions on vehicular technology, vol. 62, no. 2, pp. 552–561,.
[6] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adversarial networks with limited data. InProc. NeurIPS, 2020.
[7] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks,2019
[8] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten,Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of StyleGAN. InProc. CVPR, 2020.
[9] Kwang-Ju Kim, Pyong-Kun Kim, Yun-Su Chung, and Doo-Hyun Choi. Multi-scale detector for accurate vehicle de-tection in traffic surveillance data.IEEE Access, 7:78311–78319, 2019.
[10] Rayson Laroca, Evair Severo, Luiz A. Zanlorensi, Luiz S.Oliveira, Gabriel Resende Gonc ̧alves, William Robson Schwartz, and David Menotti.A robust real-time automatic license plate recognition based on the YOLO detector. CoRR, abs/1802.09567, 2018.
[11] Hui Li and Chunhua Shen. Reading car license plates us-ing deep convolutional neural networks and lstms.CoRR,abs/1601.05610, 2016.
[12] Xuebo Liu, Ding Liang, Shi Yan, Dagui Chen, Yu Qiao, andJunjie Yan. FOTS: fast oriented text spotting with a unified network.CoRR, abs/1801.01671, 2018.
[13] Siwei Lyu, Ming-Ching Chang, Dawei Du, Wenbo Li,Yi Wei, Marco Del Coco, Pierluigi Carcagn`ı, Arne Schu-mann, Bharti Munjal, Dinh-Quoc-Trung Dang, Doo-HyunChoi, Erik Bochinski, Fabio Galasso, Filiz Bunyak, GunaSeetharaman, Jang-Woon Baek, Jong Taek Lee, Kannap-pan Palaniappan, Kil-Taek Lim, Kiyoung Moon, Kwang-JuKim, Lars Sommer, Meltem Brandlmaier, Min-Sung Kang,Moongu Jeon, Noor M. Al-Shakarji, Oliver Acatay, Pyong-Kun Kim, Sikandar Amin, Thomas Sikora, Tien Dinh, To-bias Senst, Vu-Gia-Hy Che, Young-Chul Lim, Young-minSong, and Yun-Su Chung.Ua-detrac 2018: Report ofavss2018 iwt4s challenge on advanced traffic monitoring.In2018 15th IEEE International Conference on AdvancedVideo and Signal Based Surveillance (AVSS), pages 1–6,2018.
[14] Mehdi Mirza and Simon Osindero. Conditional generativeadversarial nets, 2014.
[15] Augustus Odena, Christopher Olah, and Jonathon Shlens.Conditional image synthesis with auxiliary classifier GANs.In Doina Precup and Yee Whye Teh, editors, Proceedings of the 34th International Conference on Machine Learning, volume 70 of Proceedings of Machine Learning Research, pages 2642–2651. PMLR, 06–11 Aug 2017.
[16] Joseph Redmon, Santosh Divvala, Ross Girshick, and AliFarhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016
[17] Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Logeswaran, Bernt Schiele, and Honglak Lee. Generative adversarial text to image synthesis, 2016.
[18] Zied Selmi, Mohamed Ben Halima, and Adel M. Alimi. Deep learning system for automatic license plate detection and recognition. In2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), vol-ume 01, pages 1132–1138, 2017.
[19] Yi Sun, Xiaogang Wang, and Xiaoou Tang. Deep learning face representation by joint identification-verification, 2014.
[20] Longyin Wen, Dawei Du, Zhaowei Cai, Zhen Lei, Ming-Ching Chang, Honggang Qi, Jongwoo Lim, Ming-Hsuan Yang, and Siwei Lyu. DETRAC: A new benchmark and protocol for multi-object tracking.CoRR, abs/1511.04136,2015.
[21] Zhenbo Xu, Wei Yang, Ajin Meng, Nanxue Lu, and Huan Huang. Towards end-to-end license plate detection and recognition: A large dataset and baseline. In Proceedings of the European Conference on Computer Vision (ECCV),pages 255–271, 2018.
[22] Jian Ye, Zhe Chen, Juhua Liu, and Bo Du. Textfusenet : Scene text detection with richer fused features. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pages 516–522. International Joint Conferences on Artificial Intelligence Organization, 2020.
[23] Chengpu Yu, Mei Xie, and Jin Qi. A novel system design of license plate recognition. In2008 International Symposium on Computational Intelligence and Design, volume 2, pages114–117, 2008.
[24] Yule Yuan, Wenbin Zou, Yong Zhao, Xinan Wang, XuefengHu, and Nikos Komodakis. A robust and efficient approachto license plate detection.IEEE Transactions on Image Pro-cessing, 26(3):1102–1114, 2017.
[25] Fukai Zhang, Ce Li, and Feng Yang. Vehicle detection in ur-ban traffic surveillance images based on convolutional neural networks with feature concatenation.Sensors, 19(3), 2019.
[26] Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiao-gang Wang, Xiaolei Huang, and Dimitris Metaxas. Stackgan : Text to photo-realistic image synthesis with stacked genera-tive adversarial networks, 2017.
[27] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei AEfros. Unpaired image-to-image translation using cycle-consistent adversarial networkss. InComputer Vision(ICCV), 2017 IEEE International Conference on, 2017.
[28] Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets, 2014.
[29] Gabriel Resende Gonc¸alves, Sirlene Pio Gomes da Silva, David Menotti, and William Robson Schwartz. Benchmark for license plate character segmentation. Journal of Electronic Imaging, 25(5):053034, 2016.