International Core Journal of Engineering 2020-26 | Page 87

澳 method is how to optimize the synthesis algorithm to minimize the distribution difference between the synthetic image and the real image. Besides, based on the training strategy of using synthetic and real mixed data first, and then using real images to perform fine tuning (fusion + RealImg), the model will achieve better training results (mAP 58.9%). Fig. 8 shows partial results of the proposed synthesis algorithm on the FlickrLogos-32 test set. It can be found that the algorithm is robust to multi-scale, multi-view, rotational deformation and partial occlusion of some object. Fig. 8. Qualitative evaluation on FlickrLogo-32. [9] IV. C ONCLUSION In this work, aim at the problem of lack of annotation data in the logo recognition task under the deep learning framework. From the perspective of automatic synthesis of large-scale training data, a context-based logo image synthesis algorithm is proposed based on the existing synthetic ideas. Through detailed experiments on the FlickrLogos-32, it is shown that when only a small amount of annotation data is available, the proposed synthesis algorithm can greatly improve (mAP 8.5%) the performance of the logo recognition algorithm without relying on additional manual annotation, fully validated the effectiveness and superiority of the proposed algorithm. [10] [11] [12] [13] [14] [15] [16] R EFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [17] Y. Fu, Beijing Jiaotong University, 2016. Y. Gao, F. Wang, H. Luan, T.S. Chua, International Conference on Multimedia Retrieval. ACM, vol. 169, 2014. C. Pan, Z. Yan, X. Xu, M. Sun, J. Shao, D. Wu, IET International Conference on Smart and Sustainable City, 123, 2013. K. He, G. Gkioxari, P. Dollar, R. Girshick, International Conference on Computer Vision. IEEE, vol. 2980, 2017. X. Wang, A. Shrivastava, A. Gupta, Conference on Computer Vision and Pattern Recognition. IEEE, 3039, 2017. W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C. Fu, A.C. Berg, Springer International Publishing, 21, 2016. A. Joly, O. Buisson, International Conference on Multimedia, 581, 2009. Y. Kalantidis, L.G. Pueyo, M. Trevisiol, R.V. Zwol, Y. Avrithis, ACM International Conference on Multimedia Retrieval, vol. 1, 2011. [18] [19] [20] [21] [22] [23] 65 S. Romberg, L.G. Pueyo, R. Lienhart, R.V. Zwol, ACM International Conference on Multimedia Retrieval, vol. 1, 2011. S.C.H. Hoi, X. Wu, H. Liu, Y. Wu, H. Wang, H. Xue, Q. Wang, IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 5, no. 2403, 2015. S. Bianco, M. Buzzelli, D. Mazzini, R. Schettini, Neurocomputing, C, vol. 23, 2017. H. Su, X. Zhu, S. Gong, IEEE Winter Conference on Applications of Computer Vision, no. 530, 2017. A. Gupta, A. Vedaldi, A. Zisserman, IEEE Computer Vision and Pattern Recognition, no. 2315, 2016. M. Jaderberg, K. Simonyan, A. Vedaldi, A. Zisserman, International Journal of Computer Vision, vol. 1, no. 1, 2016. G. Georgakis, A. Mousavian, A.C. Berg, J. Kosecka, arXiv preprint arXiv, vol. 1702, no. 07836, 2017. C. Eggert, A. Winschel, R. Lienhart, ACM International Conference on Multimedia, vol. 1283, 2015. Y. Bengio, R. Collobert, J. Weston, ACM International Conference on Machine Learning, no. 41, 2009. A. Oliva, A. Torralba, Trends in Cognitive Sciences, vol. 12, no. 520, 2007. R. Mottaghi, X. Chen, X. Liu, N.G. Cho, S.W. Lee, S. Fidler, R. Urtasun, A. Yuille, IEEE Computer Vision and Pattern Recognition, no. 891, 2014. H. Katti, M.V. Peelen, S.P. Arun, Attention Perception & Psychophysics, vol. 2, no. 1, 2017. B. Zhou, A. Lapedriza, A. Khosla, A. Oliva, A. Torralba, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 6, no. 1452, 2018. J. Gu, S. Gould, arXiv preprint arXiv, vol. 1506, no. 07224, 2015. G. Oliveira, X. Frazão, A. Pimentel, B. Ribeiro, IEEE International Joint Conference on Neural Networks, vol. 985, 2016.