車輛重識別在智能交通系統中是一項非常重要的多媒體技術，目的是在不同地點且非重疊的道路攝影機中根據給定的車輛影像，找出相同的車輛，藉由此技術可得知此車輛在什麼時間點與地點出現過，儘管最近的進展證實了它的進步，此技術仍然存在多項挑戰: 1) 由於視角的不同造成巨大的類內差異性； 2) 車輛會因為同顏色或類型而產生不同車輛間相似度非常高的問題； 3)由於訓練資料不平衡造成長尾分佈的問題，會造成網路提取特徵時也導致特徵空間不平衡，而忽略了數量較稀少的類別。在本篇論文中我們提出了新穎的Balanced Swapping and Channel-wise Diversity學習網路針對上述提到的問題做改進。我們的網路有著以下的特點: 1) Balanced Swapping: 一個平衡的子集合訓練方案，減輕在重識別任務中因為資料不平衡導致的特徵空間展延不平衡，使用此方法，尾部類別在訓練時期並不會被忽略掉； 2) Channel-wise Diversity: 特徵空間中將具有相似外觀但不同車輛間距離推遠，並使相同車輛但不同視點樣本間的距離不是非常接近，提出了基於通道分群損失函數，以將特徵分離到不同的視圖，如此一來更具判別性的特徵可以被網路抽取出來； 3) Orientation based Arc-margin: 將不同方向但相同ID車輛在特徵空間中推開一點距離，藉由此方法讓相同方向車輛在特徵空間中更靠近然而不同方向可以保留一點距離來學到更有效的特徵空間。我們的方法不僅簡單且有效。實驗結果表明了我們的方法在車輛重識別中有巨大的提升同時也超越了目前最先進的方法。

Vehicle re-identification (ReID) is an active research topic for intelligent transportation systems, aiming to find a given vehicle across various non-overlapping camera views. Despite gaining significant progress, vehicle ReID still poses several technical challenges: 1) large intra-class variations due to view-point changes; 2) subtle inter-class appearances especially for vehicles of the same models or colors; 3) long-tailed data annotations. In this paper, we propose a novel approach, featured with balanced swapping and channel-wise Diversity, to address the issues above. Our approach has three distinct features, including 1) Balanced swapping is an balanced subset training scheme to reduce imbalanced feature span caused by long-tailed data distributions in vehicle ReID, preventing tail classes from being neglected in model training; 2) Channel-wise diversity is developed to encourage feature segments of self-embedding to be orthogonal so that diverse characteristics and local regions of vehicles can be captured to highlight subtle inter-class differences; 3) Orientation based arc-margin is designed to cluster intra-class vehicle orientations from which view-specific, discriminative features can be derived. Experiments show that our method significantly improves vehicle ReID performances over the state-of-the-arts.

摘要----------------------------------------------------i
Abstract-----------------------------------------------ii
I. Introduction-------------------------------------1
II. Related works------------------------------------5
III. Proposed method----------------------------------8
A. Re-Identification Baseline-----------------------8
B. Diversity Loss-----------------------------------10
C. Central Arc-margin Loss--------------------------11
D. Long Tailed Balanced Swapping--------------------13
IV. Experiments--------------------------------------15
A. Datasets and metrics-----------------------------15
B. Experiments Step---------------------------------15
C. Comparison with State-of-The-Arts----------------16
D. Experiments on the VeRi-Wild Dataset-------------18
E. Experiments on VeRi Dataset----------------------19
F. Ablation Study-----------------------------------20
1) Effectiveness of Diversity Loss------------------21
2) Effectiveness of Central Arc-margin Loss---------22
3) Effectiveness of Long Tailed Balanced Swappin----22
G. Visualization of the embedding space with t-SNE--22
H. Distance analysis for each embedding group-------23
I. Unsupervised cross-domain results----------------25
V. Conclusion---------------------------------------25
References---------------------------------------------26

[1] Y. Zhou and L. Shao, “Aware Attentive Multi-view Inference for Vehicle Re-identification,” in CVPR, 2018.
[2] R. Chu, Y. Sun, Y. Li, Z. Liu, C. Zhang, and Y. Wei, “Vehicle Re-identification with Viewpoint-aware Metric Learning,” in ICCV, 2019.
[3] B. He, J. Li, Y. Zhao, and Y. Tian, “Part-regularized Near-duplicate Vehicle Re-identification,” in CVPR, 2019.
[4] Z. Tang, M. Naphade, S. Birchfield, J. Tremblay, W. Hodge, R. Kumar, S. Wang, and X. Yang, “Pamtri: Pose-aware Multi-task Learning for Vehicle Re-identification using Highly Randomized Synthetic Data,” in ICCV, 2019.
[5] Y.Lou, Y.Bai, J.Liu, S.Wang, and L.Duan,“Veri-wild:Alargedataset and a New Method for Vehicle Re-identification in the Wild,” in CVPR, 2019.
[6] J.Liu, Y.Sun, C.Han, Z.Dou, and W.Li,“Deep Representation Learning on Long-tailed Data: A Learnable Embedding Augmentation Perspective,” in CVPR, 2020.
[7] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in CVPR, 2016.
[8] Y. Sun, L. Zheng, Y. Yang, Q. Tian, and S. Wang, “Beyond Part Models: Person Retrieval with Refined Part Pooling (and a Strong Convolutional Baseline),” in ECCV 2018.
[9] H. Luo, Y. Gu, X. Liao, S. Lai, and W. Jiang, “Bag of Tricks and a Strong Baseline for Deep Person Re-identification,” in CVPR Workshops, 2019.
[10] F. Schroff, D. Kalenichenko, and J. Philbin, “Facenet: A Unified Embedding for Face Recognition and Clustering,” CVPR, 2015.
[11] A. Hermans, L. Beyer, and B. Leibe, “In Defense of the Triplet Loss for Person Re-identification,” in ICCV 2017.
[12] Y.Wen, K.Zhang, Z.Li, and Y.Qiao,“A Discriminative Feature Learning Approach for Deep Face Recognition,” in ECCV 2016.
[13] J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “Arcface: Additive Angular Margin Loss for Deep Face Recognition,” in CVPR, 2019.
[14] P. Khorramshahi, A. Kumar, N. Peri, S. S. Rambhatla, J.-C. Chen, and R. Chellappa, “A Dual-path Model with Adaptive Attention for Vehicle Re-identification,” ICCV, 2019.
[15] Y. Yao, L. Zheng, X. Yang, M. Naphade, and T. Gedeon, “Simulating Content Consistent Vehicle Datasets with Attribute Descent,” in ECCV 2020.
[16] Y. Bai, Y. Lou, F. Gao, S. Wang, Y. Wu, and L.-Y. Duan, “Group-sensitive Triplet Embedding for Vehicle Reidentification,” IEEE Transactions on Multimedia, 2018.
[17] D. Cao, X. Zhu, X. Huang, J. Guo, and Z. Lei, “Domain Balancing: Face Recognition on Long-tailed Domains,” in CVPR 2020.
[18] R. Kumar, E. Weill, F. Aghdasi, and P. Sriram, “Vehicle Re-identification: An Efficient Baseline Using Triplet Embedding,” in IJCNN 2019.
[19] P. Xie, “Learning Compact and Effective Distance Metrics with Diversity Regularization,” in Joint European Conference on Machine Learning and Knowledge Discovery in Databases. 2015.
[20] Z. Gong, P. Zhong, Y. Yu, and W. Hu, “Diversity-promoting Deep Structural Metric Learning for Remote Sensing Scene Classification,” IEEE Transactions on Geoscience and Remote Sensing, 2017.
[21] X. Liu, W. Liu, T. Mei, and H. Ma, “Provid: Progressive and Multi-modal Vehicle Reidentification for Large-scale Urban Surveillance,” IEEE Transactions on Multimedia, 2017.
[22] Z. Zhong, L. Zheng, D. Cao, and S. Li, “Re-ranking Person Re-identification with k-reciprocal Encoding,” in CVPR, 2017.
[23] Y. Zhou and L. Shao, “Aware Attentive Multi-view Inference for Vehicle Re-identification,” in CVPR, 2018.
[24] L.Yang, P.Luo, C.ChangeLoy, and X.Tang,“A Large-scale Car Dataset for Fine-grained Categorization and Verification,” in CVPR, 2015.
[25] F. Schroff, D. Kalenichenko, and J. Philbin, “Facenet: A Unified Embedding for Face Recognition and Clustering,” in CVPR, 2015.
[26] X. Liu, W. Liu, T. Mei, and H. Ma, “A Deep Learning-based Approach to Progressive Vehicle Re-identification for Urban Surveillance,” in ECCV, 2016.
[27] H.Liu, Y.Tian, Y.Yang, L.Pang, and T.Huang, “Deep Relative Distance Learning: Tell the Difference Between Similar Vehicles,” in CVPR, 2016.
[28] Y. Yuan, K. Yang, and C. Zhang, “Hard-aware Deeply Cascaded Embedding,” in ICCV, 2017.
[29] J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired Image-to-image Translation Using Cycle-consistent Adversarial Networks,” in ICCV, 2017.
[30] X. Liu, S. Zhang, Q. Huang, and W. Gao, “Ram: A Region-aware Deep Model for Vehicle Re-identification,” in ICME, 2018.
[31] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-cam: Visual Explanations from Deep Networks via Gradient-based Localization,” in ICCV, 2017.
[32] L. v. d. Maaten and G. Hinton, “Visualizing Data Using t-SNE,” Journal of machine learning research, 2008.
[33] Chen, T.-S., Liu, C.-T., Wu, C.-W. and Chien, S.-Y., “Orientation-aware Vehicle Re-identification with Semantics-guided Part Attention Network,” ECCV 2020.
[34] Meng, D., Li, L., Liu, X., Li, Y., Yang, S., Zha, Z.-J., Gao, X., Wang, S. and Huang, Q., ”Parsing-based Viewaware Embedding Network for Vehicle Re-Identification,” CVPR2020.
[35] Deng, J., Guo, J., Xue, N. and Zafeiriou, S., ArcFace: Additive Angular Margin Loss for Deep Face Recognition. CVPR 2019.