臺灣博碩士論文加值系統

因為類神經網路的興起使得很多電腦視覺方面的問題有了很大的突破和進展。然而在電腦視覺的領域中的許多問題，現實應用和學術研究的成果中間還是存在一段不小的落差。監視器下的車輛細項分類的問題就是其中之一。以往此類問題常受限於資料收集不易、車輛種類數量嚴重不均、照片多為低解析度和人工標記過於困難等問題。導致此類問題發展較為緩慢且成果不佳。近年來，因為相關的應用被重視程度明顯增加且相關的資料集相繼出現使得此問題發展越來越快。我們在此篇論文將著重在如何使用網路上的高解析度圖片來改善低解析度圖片之車輛細項分類的表現。我們針對這個問題提出了兩個解法。1.利用前處理網路，加強和復原低解析度的監視器車輛圖片的細部資訊2.利用部分分享權重的方式連接卷積網路我們將我們的方法實驗再BoxCars21K資料集上。實驗顯示我們的方法能在不利用到立體邊框標誌的資訊下，達到差不多甚至超過目前最好且有使用到立體邊框資訊的成果。

AbstractDeep learning has became the most popular topics in computer vision field. Con-volution neural network has achieved impressive performance in most of computervision problems. However, there is still a large gap between academic researchesand real-world applications in many computer vision problems. Fine-grained clas-sification for surveillance images is one of them. There are a few reasons for theslow development of this problem. Data insufficiency, imbalance and low-resolutionimages or video make collecting a fine-grained vehicle data-set for surveillance im-ages cost astonishing labor effort and poor performance of related research works.In recent years, there are several large-scale vehicle data-sets and great researchesshow up. In this paper, we address how to use images collected from internet assupporting data to improve fine-grained classification for surveillance images. Wepropose two novel approaches to connect two different domains (web images andsurveillance images). i.e. 1) A hallucination network to enhance the edge and de-tails of low-resolution images. 2) Partially weight sharing between two convolutionnetworks for efficient connections. We implement our experiment on BoxCars21kdata-set. The experiment results show that our methods can achieve quite close oreven better performance than state-of-the-art result which needs 3D bounding boxlabel.

Acknowledgments i
Abstract iii
List of Figures vii
List of Tables viii
Chapter 1 Introduction 1
Chapter 2 Related Work 3
Chapter 3 Methodology 5
Chapter 4 Experiments 11
Chapter 5 Discussion 15
Chapter 6 Conclusion and Future Work 17
Bibliography 18

[1] J. Bromley, I. Guyon, Y. LeCun, E. S¨ackinger, and R. Shah. Signature verification using a” siamese” time delay neural network. In Advances in Neural Information Processing Systems, pages 737–744, 1994.
[2] D. Cai, K. Chen, Y. Qian, and J.-K. K¨am¨ar¨ainen. Convolutional low-resolution fine-grained classification. arXiv preprint arXiv:1703.05393, 2017.
[3] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pages 248–255. IEEE, 2009.
[4] T. Gebru, J. Krause, Y. Wang, D. Chen, J. Deng, and L. Fei-Fei. Fine-grained car detection for visual census estimation. In AAAI, pages 4502–4508, 2017.
[5] Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell. Caffe: Convolutional architecture for fast feature embedding. In Proceedings of the 22nd ACM international conference on Multimedia, pages 675– 678. ACM, 2014.
[6] J. Johnson, A. Alahi, and L. Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In European Conference on Computer Vision, pages 694–711. Springer, 2016.
[7] J. Krause, T. Gebru, J. Deng, L.-J. Li, and L. Fei-Fei. Learning features and parts forfine-grainedrecognition. InPattern Recognition (ICPR), 2014 22nd International Conference on, pages 26–33. IEEE, 2014.
[8] V. Lajish and S. K. Kopparapu. Mobile phone based vehicle license plate recognition for road policing. arXiv preprint arXiv:1504.01476, 2015.
[9] J. Lezama, Q. Qiu, and G. Sapiro. Not afraid of the dark: Nir-vis face recognition via cross-spectral hallucination and low-rank embedding. arXiv preprint arXiv:1611.06638, 2016.
[10] Y.-L. Lin, V. I. Morariu, W. Hsu, and L. S. Davis. Jointly optimizing 3d model fitting and fine-grained classification. In European Conference on Computer Vision, pages 466–480. Springer, 2014.
[11] H. Liu, Y. Tian, Y. Yang, L. Pang, and T. Huang. Deep relative distance learning: Tell the difference between similar vehicles. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2167–2175, 2016.
[12] J. Sochor, A. Herout, and J. Havel. Boxcars: 3d boxes as cnn input for improved fine-grainedvehiclerecognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3006–3015, 2016.
[13] Z. Wang, S. Chang, Y. Yang, D. Liu, and T. S. Huang. Studying very low resolution recognitionusingdeepnetworks. InProceedingsoftheIEEEConferenceonComputer Vision and Pattern Recognition, pages 4792–4800, 2016.
[14] J. Wu, S. Ding, W. Xu, and H. Chao. Deep joint face hallucination and recognition. arXiv preprint arXiv:1611.08091, 2016.
[15] L. Yang, P. Luo, C. Change Loy, and X. Tang. A large-scale car dataset for finegrained categorization and verification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3973–3981, 2015.