# unsupervised_vehicle_reidentification_with_progressive_adaptation__dd799fd7.pdf

Unsupervised Vehicle Re-identification with Progressive Adaptation

Jinjia Peng1,2 , Yang Wang3,4 , Huibing Wang1,2 , Zhao Zhang3,4 , Xianping Fu1,2 , Meng Wang3,4

1College of Information and Science Technology, Dalian Maritime University, Liaoning, Dalian 2Pengcheng Laboratory, Shenzhen, Guangdong 3Key Laboratory of Knowledge Engineering with Big Data, Ministry of education, Hefei University of Technology, China 4School of Computer Science and Information Engineering, Hefei University of Technology, China {jinjiapeng, huibing.wang,fxp}@dlmu.edu.cn, yangwang@hfut.edu.cn, {cszzhang,eric.mengwang}@gmail.com

Vehicle re-identification (re ID) aims at identifying vehicles across different non-overlapping cameras views. The existing methods heavily relied on welllabeled datasets for ideal performance, which inevitably causes fateful drop due to the severe domain bias between the training domain and the real-world scenes; worse still, these approaches required full annotations, which is labor-consuming. To tackle these challenges, we propose a novel progressive adaptation learning method for vehicle re ID, named PAL, which infers from the abundant data without annotations. For PAL, a data adaptation module is employed for source domain, which generates the images with similar data distribution to unlabeled target domain as pseudo target samples . These pseudo samples are combined with the unlabeled samples that are selected by a dynamic sampling strategy to make training faster. We further proposed a weighted label smoothing (WLS) loss, which considers the similarity between samples with different clusters to balance the confidence of pseudo labels. Comprehensive experimental results validate the advantages of PAL on both Vehicle ID and Ve Ri-776 dataset.

1 Introduction

Vehicle-related research has attracted a great deal of attention, ranging from vehicle detection, tracking [Tang et al., 2019] to classification [Hu et al., 2017]. Unlike them, vehicle re-identification (re ID) aims to match a specific vehicle across scenes captured from multiple non-overlapping cameras, which is of vital significance to intelligent transport. Most of the existing vehicle re ID methods, in particular for deep learning models, usually adopt the supervised approaches [Zhao et al., 2019; Lou et al., 2019; Bai et al., 2018; Wang et al., 2017; Guo et al., 2019] for an ideal performance. However, they suffer from the following limitations.

Corresponding Author

On one hand, due to the domain bias, well-trained vehicle re ID models under these supervised methods may suffer from a poor performance when directly deployed to the real-world large-scale camera networks. On the other hand, these methods heavily relied on the full annotations, i.e., the identity labels of all the training data from multiple cross-view cameras. However, it is labor expensive to annotate large-scale unlabeled data in the real-world scenes. In particular for the vehicle re ID task, it is always required to annotate the same vehicle under all cameras. Hence, how to incrementally optimize the vehicle re ID algorithms utilizing the combination of the abundant unlabeled data and existing well-labeled data is practical but challenging. To these ends, a few unsupervised strategies have been proposed. Specifically, [Bashir et al., 2019] takes the cluster to be pseudo labels and then select the reliable pseudo-labeled samples for training. However, the incorrect annotations assigned by clustering are inevitable. One may try transferring images from the well-labeled domain to the unlabeled domain via style transfer for the unsupervised vehicle re ID [Isola et al., 2017; Yi et al., 2017]. The generated images are employed to train the re ID model, which preserves the identity information from well-labeled domain, while learn the style of unlabeled domain. However, this solution is limited by the learned style that is different from the unlabeled domain, and may fail to adapt to the real-world scenes without label information. In this paper, we propose a novel unsupervised method, named PAL, together with Weighted Label Smoothing (WLS) loss to better exploit the unlabeled data, while adapt the target domain to vehicle re ID progressively . Unlike the existing unsupervised re ID methods, a novel adaptation module is proposed to generate pseudo target images , which learns the style of unlabeled domain and preserves the identity information of the labeled domain. Besides, dynamic sampling strategy is proposed to select reliable pseudo-labeled data from the clustering result. Furthermore, the fusion data that combines the pseudo target images and reliable pseudo-labeled data is employed to train the re ID model in the subsequent training. To facilitate the presentation, we illustrate the major framework in Fig. 1.

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

Next Training Stage Source Dataset

Generated Images (Pseudo Target Samples)

Training Samples with Weights

Pseudo Target Samples Generated

Feature Learning With Fusion Samples

Domain Adaptive re ID Model

Dynamic Sampling Strategy

DBSCAN Dimensionality

} , , , { ), ( d c b a i d w ik i

Unlabeled Target Dataset

Figure 1: Illustration of the PAL framework. The images are transferred from the source domain to target domain for pseudo target samples generation. During each iteration, we 1) train the re ID model by the proposed WLS-based feature learning network that utilizes the fusion data, combining the pseudo target samples with selected samples, and 2) assign pseudo labels with various weights for unlabeled images and select reliable samples according to a dynamic sampling strategy. A means pseudo-labeled images and B is generated images.

Our major contributions are summarized as follows:

A novel progressive method, named PAL, is proposed for unsupervised vehicle re ID to better adapt the unlabeled domain, which iteratively updates the model by WLS based feature learning network, and adopts dynamic sampling strategy to assign labels for selected reliable unlabeled data.

To make full use of the existing labeled data, PAL employs a data adaptation module based on Generative Adversarial Network (GAN) for generating images as the pseudo target samples , which are combined with the selected samples from unlabeled domain for model training.

The feature learning network with WLS loss is proposed, which considers the similarity between the samples and different clusters to balance the confidence of pseudo labels to improve the performance.

Experimental results on benchmark data sets validate the superiority of our method, which is even better than some supervised vehicle re ID approaches.

2 Progressive Adaptation Learning for Vehicle Re ID

In this section, we formally discuss our proposed technique of progressive adaption learning, named PAL, for vehicle re ID. Specifically, as shown in Fig.1, a data adaptation module based on GAN is trained to transfer the well-labeled images to the unlabeled target domain, which aims at smoothing the domain bias and make full use of the existing source domain images. Then the generated images are employed as the pseudo target samples and combined with selected unlabeled samples to serve as the input to Res Net50 [He et al., 2015] for feature learning, which adapts the target domain progressively. When the model is trained, WLS loss is proposed to balance the confidence of unlabeled samples and different clusters, which exploits the pseudo labels with different weights, according to the model trained by the last

iteration. Then the output features of the re ID model are employed to select reliable samples by dynamic sampling strategy. Finally, the pseudo target samples with accurate labels and selected samples from unlabeled domain with pseudo labels are combined to be the training sets for the next iteration. In this way, a more stable adaptive model could be learned progressively.

2.1 Pseudo Target Samples Generated Network For a target domain, the supervised learning approaches are limited by the unlabeled samples, which can t be utilized to train re ID model. Though there are well-labeled datasets, directly applying them to target domain may suffer from a poor performance because of the domain bias mainly caused by diversified illuminations and complicated environment. To remedy this, Cycle GAN [Zhu et al., 2017; Lin et al., 2019] is employed to make full use of these welllabeled data, which generates pseudo target samples to narrow down the domain bias by transferring the style between source domain and target domain. The generated images share the similar style with the target domain while preserving the identity information of the source domain. Specifically, it comprises of two generator-discriminator pairs, (G, DT ) and (F, DS), which maps a sample from source (target) domain to target (source) domain and generate a sample which is indistinguishable from those in the target(source) domain[Almahairi et al., 2018]. For PAL, besides the traditional adversarial losses and cycle-consistent loss, a content loss [Taigman et al., 2016] is utilized to preserve the label information from the source domain, which is formulated as:

Lid(G, F, X, Y ) = Ey pdata(y)||F(y) y||1 + Ex pdata(x)||G(x) x||1, (1)

where X and Y represent the source domain and target domain, respectively. pdata(x) and pdata(y) represent the sample distributions in the source and target domain. One may wonder why the generated network could make full use of the well-labeled data, we answer this question from the following two aspects:

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

Through Cycle GAN, the generated pseudo target samples have similar distribution for the target domain, which reduces the bias between source and target domain. Furthermore, the identity information of source domain is also preserved by turning the content loss during the transferring phrase, implying that the well labeled annotations could be re-utilized in the subsequence.

2.2 Feature Learning Network with WLS Loss Feature learning plays a vital role of the PAL, which trains the model by combining the generated pseudo target images with the selected pseudo labeled samples. For the pseudo target images , it s easy to obtain the label information. However, how to assign labels for the pseudo labeled samples reasonably is a big challenge, due to the following facts: If the clustering centroids serve as the pseudo labels, it may cause ambiguous prediction during the training phase due to the inaccurate clustering results. Moreover, it is not reasonable to assign the same labels to all the samples regardless of the distance to the clustering centroids. Hence, WLS loss is proposed to set the pseudo label distribution as a weighted distribution over all clusters, which effectively regularizes the feature learning network to the target training data distribution. Specifically, we model the virtual label distribution as a weighted distribution over all clusters for unlabeled data according to the distance between the features and each centroid of clusters. Thus, the weights over all clusters are different in WLS loss. In this way, a dictionary α is constructed to record the weights. For an image g, the weights of the label can be calculated as:

K αg k, k [1, K], (2)

where αg k represents the weight of the image g over the kth cluster. To obtain αg k, unlabeled samples are clustered to obtain centroids set C = {c1, c2, ..., ck}, k [1, K], which is introduced in section 2.3. K is the number of clusters, while the similarity between g and ck can be calculated as dg k = ||fg fck||2, where f represents the feature of images or centroids. The set of distance of image g over K centroids could be described as: dg = {dg 1, dg 2, ..., dg k}, k [1, K]. Inspired with [Huang et al., 2019], then all elements in dg are sorted with descending order, and saved to dsg. αk is obtained by taking the corresponding index of dsg k in the set of dsg:

αg k = (1 dg k max(dg)) ψdsg(dg k), (3)

where ψdsg( ) is the index of dg k in dsg. Thus, the corresponding relationship between images and cluster centroids is constructed with different weights. In order to filter noise, top-m in w are selected as reliable weights, with others set to 0. To this end, we have:

wg k = 0, if wg k < θ wg k, if wg k >= θ, (4)

where θ is a threshold. Thus, the WLS loss of unlabeled data ℓwls can be formulated as:

k=1 wklog(p(k)) (5)

Besides the real unlabeled samples, there are some generated images by Cycle GAN that are combined to train the re ID model. The training loss is defined as follows:

ℓ= (1 σ) log(p(y)) σ λ

k=1 wklog(p(k)) (6)

For a generated image λ = 0, the loss is equivalent to crossentropy loss and y is the label of the generated image. When λ = 1 means the image is from the unlabeled data and y is the cluster it belongs to. Beyond that, for the unlabeled data, σ is a smoothing factor between cross-entropy loss and WLS loss.

2.3 Dynamic Sampling Strategy It is crucial to obtain appropriately selected candidates to exploit the unlabeled data. When the model is weak, small reliable measure is set, which is nearby to the cluster centroids in the feature space. As the model becomes stronger in subsequent iterations, various instances should be adaptively selected as the training examples. Hence, a dynamic sampling strategy is proposed to ensure the reliability of selected pseudo-labeled samples. As shown in Fig.1, images in the target domain is processed by the well-trained re ID model to output the features with high dimensions. Most methods select the K-Means to generate clusters, which are required to be initialized by the cluster centroids. However, it is uncertain on how many categories are required in the target domain. Hence, DBSCAN is selected as the clustering method. Specifically, instead of employing the fixed clustering radius, the paper employs a dynamic cluster radius rad that is calculated by K-Nearest Neighbors (KNN). After DBSCAN, in order to filter noise, some of top reliable samples are selected to be assigned with soft labels, according the distance between features of the samples and cluster centroids. For our method, samples with ||fg, cfg||2 < γ are satisfied for the next iteration for training model, where fg is the feature of g-th image and cfg is the feature of the cluster centroid where the fg belongs to. Our method is summarized in Algorithm 1.

3 Experiments 3.1 Datasets and Evaluation Metrics The experiments are conducted over the following two typical data sets for Vehicle re ID: Ve Ri-776 and Vehicle ID. Ve Ri776 [Liu et al., 2018] is a large-scale urban surveillance vehicle dataset for re ID, which contains over 50,000 images of 776 vehicles, where 37,781 images of 576 vehicles are employed as training set, while 11,579 images of 200 vehicles are employed as a test set. A subset of 1,678 images in the test set generates the query set. Vehicle ID [Liu et al., 2016a] is a surveillance dataset from the real-world scenario, which

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

Algorithm 1 PAL for Unsupervised Vehicle re ID

Require: Numbers of images on the target domain N, labeled Source domain S, unlabeled target domain T, iteration number M, cluster number K, reliability threshold γ, new Training Set D Ensure: An encoder E for target domain 1: Transfer style from S to T by GAN to generate pseudo target images ST 2: Initialization E(0) with ST, D: D = ST 3: for i=1 to M do 4: Train E(i) with D utilizing WLS-based feature learning network, compute ft = E(i)(D) 5: Reduce dimension by manifold learning f = mad(ft), calculate number and centroids of clusters: (K, C) = DBSCAN(f) 6: Select features of centroids: {ck}K k=1 {fck}K k=1 7: D= ST 8: for k=1 to K do 9: for g = 1 to N do 10: if ||fg, fck||2 > γ then 11: D = D Tg 12: Calculate weights wg k by Eq.(2)(3)(4) 13: end if 14: end for 15: end for 16: end for

contains 221,763 images corresponding to 26,267 vehicles in total. From the original testing data, four subsets, which contain 800, 1,600, 2,400 and 3,200 vehicles, are extracted for vehicle search for multi-scales. CMC curve and m AP are employed to evaluate the overall performance for all test images. For each query, its average precision (AP) is computed from the precision-recall curve.

3.2 Implementation Details For Cycle GAN, the model is trained in the tensorflow [Abadi et al., 2016]. It is worth mentioning that any label notation aren t utilized during the learning procedure. In the stage of the feature learning, the Res Net50 [He et al., 2015] is employed as the backbone network. For PAL, the images are transferred by Cycle GAN from source domain to target domain, which are as Pseudo target samples for training the feature learning model. Considering the limit of device, when training re ID model on the Ve Ri-776, 10,000 transferred images from Vehicle ID are utilized as the pseudo target images . The same implementations are conducted when the re ID model is trained on the Vehicle ID. Besides that, when training the unsupervised model on Vehicle ID, only 35,000 images from the Vehicle ID are selected as the training set. Moreover, any annotations of target domain aren t employed in our framework.

3.3 Comparison with the State-of-the-art Methods In this section, the results of the comparison between PAL and other state-of-the-art methods are reported in Tables 1, 2 and Figures 2, 3, which includes: (1) FACT [Liu et al., 2016b]; (2) FACT+Plate-SNN+STR [Liu et al., 2016b]; (3)

0 10 20 30 40 50 Rank

Matching Rate

Cycle GAN Direct Transfer Baseline System PAL(iter=1) PAL

Figure 2: CMC results of several typical methods on Ve Ri-776. The proposed PAL outperforms other compared methods, especially the Cycle GAN and the Direct Transfer . Besides that, compared with the PAL (iter=1) , PAL has a high improvement in m AP, which demonstrates the progressive learning could increase the adaptive ability for the re ID model in the unlabeled target domain.

Method m AP(%) Rank1(%) Rank5(%) FACT 18.75 52.21 72.88 FACT+Plate-SNN+STR 27.77 61.44 78.78 VR-PROUD 22.75 55.78 70.02 PUL 17.06 55.24 66.27 Cycle GAN 21.82 55.42 67.34 Direct Transfer 19.39 56.14 68.00 Baseline System 31.94 58.58 73.24 PAL 42.04 68.17 79.91

Table 1: Performance of different methods on Ve Ri-776. The best results are shown in bold face. PAL can achieve best performance.

Mixed Diff+CCL [Liu et al., 2016a]; (4)VR-PROUD [Bashir et al., 2019]; (5) Cycle GAN [Zhu et al., 2017]. This is method of style transfer, which is employed for the domain adaptation; (6) Direct Transfer: It directly employed the welltrained re ID model by the [Zheng et al., 2018] on source domain to the target domain; 7)Baseline System. Compared with the framework of PAL, it utilizes the original samples from source domain instead of generated data and the re ID model is only trained with cross-entropy (CE) loss; 8)PUL [Fan et al., 2018]. The methods of (1), (2) and (3) are supervised vehicle re ID approaches. And others are unsupervised methods. Specially, the PUL is an unsupervised adaptation method of person re ID. Since only a few works focused on the unsupervised vehicle re ID, PUL is compared with the proposed PAL in this paper. There are some other methods that similar with PUL. However, most of them require special annotations, such as labels for segmenting or detecting keypoints, which are not annotated in the existing vehicle re ID datasets. From the Tables 1, 2, we note that the proposed method achieves the best performance among the compared with methods with Rank-1 = 68.17%, m AP = 42.04% on Ve Ri-776, Rank-1 = 50.25%, 44.25%, 41.08%, 38.19%, m AP = 53.50%, 48.05%, 45.14%, 42.13% on Vehicle ID with the test size of 800, 1600, 2400, 3200, respectively. Compared with PUL [Fan et al., 2018] and VR-PROUD [Bashir et al., 2019], PAL has 24.98% and 19.29% gains on Ve Ri-776, respectively. Our model also outperforms the PUL and VR-PROUD in Rank-1, Rank-5 and m AP on Vehicle I-

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

Method Test size = 800 (%) Test size = 1600 (%) Test size = 2400 (%) Test size = 3200 (%) m AP Rank1 Rank5 m AP Rank1 Rank5 m AP Rank1 Rank5 m AP Rank1 Rank5 FACT 49.53 67.96 44.63 64.19 39.91 60.49 Mixed Diff+CCL 49.00 73.50 42.80 66.80 38.20 61.60 PUL 43.90 40.03 56.03 37.68 33.83 49.72 34.71 30.90 47.18 32.44 28.86 43.41 Cycle GAN 42.32 37.29 58.56 34.92 30.00 49.96 31.89 27.15 46.52 29.17 24.83 42.17 Direct Transfer 40.58 35.48 57.26 33.59 28.86 48.34 30.50 26.08 44.02 27.90 23.85 39.76 Baseline System 42.96 39.11 55.24 38.03 34.04 50.91 34.04 30.10 48.41 31.98 28.24 43.77 PAL 53.50 50.25 64.91 48.05 44.25 60.95 45.14 41.08 59.12 42.13 38.19 55.32

Table 2: Performance of various methods over different re ID methods on Vehicle ID. It is notable that the best results are shown in bold face. PAL can achieve best performance in most situations. Mixed Diff+CCL can also achieve good performance.

Method Generated Images

Original Images

Table 3: The settings for different ablation models.

D. For these methods, the K-Means is employed to assign pseudo-labels for unlabeled samples. Due to the uncertainty on how many categories, the K-Means is not appropriate to be utilized in the re ID task. In addition, compared with Direct Transfer , it is obvious that our proposed PAL achieves 22.65% and 12.03% gains in m AP and Rank-1 on Ve Ri-776. It also has similar improvements on Vehicle ID. Furthermore, compared with the supervised approaches, such as FACT [Liu et al., 2016b], Mixed Diff+CCL [Liu et al., 2016a] and FACT+Plate-SNN+STR [Liu et al., 2016b], PAL achieves improvements on Ve Ri-776 and Vehicle ID, validating that PAL is more adaptive to different domains. Compared with the Cycle GAN [Zhu et al., 2017] that adapts the domain bias by style transfer, our method has large improvements on both Ve Ri-776 and Vehicle ID. The proposed PAL achieves 20.22% and 12.75% improvements in m AP and Rank-1 on Ve Ri-776, respectively. Similarly, our method has 12.96%, 14.25%, 13.93% and 13.36% gains in Rank-1 on Vehicle ID with the test sets of 800, 1600, 2400 and 3200. The significant improvements are mainly due to the fact that PAL exploits the similarity among unlabeled samples through iteration for unsupervised vehicle re ID. Though the generated images have the style of target domain, they are just served as the pseudo samples. The real samples in the target domain could be more reliable to generate the discriminative features during the stage of training. These results suggest that reliable samples in target domain is an important component for the unsupervised re ID task, which indicate that PAL could make full use of the unlabeled samples in target domain. Compared with Baseline System , PAL has large improvements both on Ve Ri-776 and Vehicle ID. The PAL achieves 10.1% increase in m AP on Ve Ri-776, and 10.54%, 10.02%, 11.1%, 10.15% improvements in m AP on Vehicle ID with different test sets, respectively. These indicate that the pseudo target images and weighted label smoothing are two core components in PAL which lead the re ID model

Iteration CEL (%) BS (%) m AP Rank1 Rank5 m AP Rank1 Rank5 iter1 27.71 59.89 72.10 25.04 57.33 71.33 iter2 33.76 64.12 78.06 30.19 58.40 73.53 iter3 35.73 65.55 78.18 32.49 59.41 73.06 iter4 36.01 63.28 77.47 32.63 59.77 74.07 iter5 33.86 60.90 77.11 32.86 60.96 74.91 iter6 34.03 62.09 75.38 31.94 58.58 73.24

Table 4: Performance of comparison between CEL and BS on Ve Ri776.

Iteration CEL (%) BS (%) m AP Rank1 Rank5 m AP Rank1 Rank5 iter1 35.69 31.43 49.54 34.93 30.71 48.41 iter2 35.89 31.44 49.65 34.00 29.90 46.98 iter3 36.49 32.39 50.84 34.33 30.29 47.14 iter4 36.62 32.73 51.01 33.62 29.67 46.38 iter5 37.33 33.69 51.25 34.08 30.15 46.94 iter6 37.12 33.45 51.36 34.04 30.10 46.94

Table 5: Performance of comparison between CEL and BS on Vehicle ID(2400).

trained by our method to be more robust to different domains. We discuss more details in the next section.

3.4 Ablation Studies We conduct ablation studies on two major components of PAL, i.e., the data adaptation module and WLS, which are shown in Fig.4. The settings are depicted in Table 3. All of them share the similar structure with PAL. Generated Images means employing the transferred images from source domain and image of target domain to train the models, while Original Images means to utilize the original images of source domain and samples from target domain for unsupervised vehicle re ID. WLS, CE represent that employ the WLS and cross-entropy loss to train re ID models, respectively. Fig.4 shows that PAL achieves the best performance on two datasets, demonstrating that the data adaptation module and WLS are effective to adapt to unlabeled domain.

The Effectiveness of Generated Samples. To demonstrate the effectiveness of the generated samples, BS and CEL are compared, the results are reported in Tables 4 and 5. For CEL, we utilize Cycle GAN to translate labeled images from the source domain to the target domain, and regard generated images as the pseudo target samples . Then the pseudo tar-

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

0 5 10 15 20 25 Rank

Matching Rate

Cycle GAN Direct Transfer Baseline System PAL(iter=1) PAL

(a) Test size=800

0 5 10 15 20 25 Rank

Matching Rate

Cycle GAN Direct Transfer Baseline System PAL(iter=1) PAL

(b) Test size=1600

0 5 10 15 20 25 Rank

Matching Rate

Cycle GAN Direct Transfer Baseline System PAL(iter=1) PAL

(c) Test size=2400

0 5 10 15 20 25 Rank

Matching Rate

Cycle GAN Direct Transfer Baseline System PAL(iter=1) PAL

(d) Test size=3200

Figure 3: CMC curves of several typical methods on Vehicle ID. From the curves, it is obvious that better results could be achieved when the model is tested on the different datasets, which demonstrates PAL is effective for different test sets.

Iteration OIMG (%) BS (%) m AP Rank1 Rank5 m AP Rank1 Rank5 iter1 28.61 60.90 74.19 25.04 57.33 71.33 iter2 30.11 62.09 75.32 30.19 58.40 73.53 iter3 30.52 61.02 74.43 32.49 59.41 73.06 iter4 32.51 63.70 76.16 32.63 59.77 74.07 iter5 33.90 65.19 76.34 32.86 60.96 74.91 iter6 37.33 67.69 79.02 31.94 58.58 73.24

Table 6: Performance of comparison between OIMG and BS on Ve Ri-776.

get samples are combined with the images in the target domain to train the re ID model. Both CEL and BS are trained by cross-entropy loss. According to the last iteration, compared with BS, the m AP of CEL increases by 2.09% on Ve Ri-776. Besides that, it also rises to 37.12% and 33.45% in m AP and Rank-1 on Vehicle ID, demonstrating that the generated images learned the important style information from the target domain, which narrow down the domain gap.

The Effectiveness of WLS. We compare BS with OIMG to validate the effectiveness of the WLS. Tables 6 and 7 show the comparisons on Ve Ri-776 and Vehicle ID, where the proposed WLS achieves better performance than cross-entropy loss. According to the last iteration, compared with the BS, the m AP and Rank-1 accuracy increased by 5.39% and 9.11% on Ve Ri-776 for OIMG, respectively. The similar conclusions hold on Vehicle ID, which indicates that the WLS loss has better generation ability to achieve discriminative representation during the stage of training.

1 2 3 4 5 6 Iteration

1 2 3 4 5 6

1 2 3 4 5 6

1 2 3 4 5 6

1 2 3 4 5 6

1 2 3 4 5 6

Figure 4: The Comparison Results. (a), (b), (c) are the m AP, Rank1 and Rank-5 of four comparison methods on Ve Ri-776. (d), (e), (f) are the results of four comparison methods in every iteration on Vehicle ID, respectively.

Iteration OIMG (%) BS (%) m AP Rank1 Rank5 m AP Rank1 Rank5 iter1 35.45 31.19 48.95 34.93 30.71 48.41 iter2 34.48 30.26 48.25 34.00 29.90 46.98 iter3 34.94 30.74 48.93 34.33 30.29 47.14 iter4 35.22 31.20 48.66 33.62 29.67 46.38 iter5 34.48 30.35 48.21 34.08 30.15 46.94 iter6 38.95 34.90 52.69 34.04 30.10 46.94

Table 7: Performance of comparison between OIMG and BS on Vehicle ID (2400).

4 Conclusion

In this paper, we propose an unsupervised vehicle re ID framework, named PAL, which iteratively updates the feature learning model and estimates pseudo labels for unlabeled data for target domain adaptation. The extensive experiments of the developed algorithm has been carried out over benchmark datasets for Vehicle Re-id. It can be observed from the results that compared with other existing unsupervised methods, PAL could achieve superior performance, and even achieve better performance than some typical supervised models.

Acknowledgments

This work was supported in part by the National Key Research and Development Program of China under grant 2018YFB0804205, by the National Natural Science Foundation of China Grant 61806035, U1936217, 61370142, 61272368, 61672365, 61732008 and 61725203, China Postdoctoral Science Foundation 3620080307, by the Dalian Science and Technology Innovation Fund 2019J11CY001, by the Fundamental Research Funds for the Central Universities Grant 3132016352, by the Liaoning Revitalization Talents Program, XLYC1908007.

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)

References [Abadi et al., 2016] Mart ın Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, et al. Tensorflow: A system for large-scale machine learning. In 12th {USENIX} Symposium on Operating Systems Design and Implementation ({OSDI} 16), pages 265 283, 2016. [Almahairi et al., 2018] Amjad Almahairi, Sai Rajeswar, Alessandro Sordoni, Philip Bachman, and Aaron Courville. Augmented cyclegan: Learning many-to-many mappings from unpaired data. ar Xiv preprint ar Xiv:1802.10151, 2018. [Bai et al., 2018] Yan Bai, Yihang Lou, Feng Gao, Shiqi Wang, Yuwei Wu, and Ling-Yu Duan. Group-sensitive triplet embedding for vehicle reidentification. IEEE Transactions on Multimedia, 20(9):2385 2399, 2018. [Bashir et al., 2019] Raja Muhammad Saad Bashir, Muhammad Shahzad, and MM Fraz. Vr-proud: Vehicle reidentification using progressive unsupervised deep architecture. Pattern Recognition, 90:52 65, 2019. [Fan et al., 2018] Hehe Fan, Liang Zheng, Chenggang Yan, and Yi Yang. Unsupervised person re-identification: Clustering and fine-tuning. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 14(4):83, 2018. [Guo et al., 2019] Haiyun Guo, Kuan Zhu, Ming Tang, and Jinqiao Wang. Two-level attention network with multigrain ranking loss for vehicle re-identification. IEEE Transactions on Image Processing, 2019. [He et al., 2015] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. ar Xiv preprint ar Xiv:1512.03385, 2015. [Hu et al., 2017] Qichang Hu, Huibing Wang, Teng Li, and Chunhua Shen. Deep cnns with spatially weighted pooling for fine-grained car recognition. IEEE Transactions on Intelligent Transportation Systems, 18(11):3147 3156, 2017. [Huang et al., 2019] Yan Huang, Jingsong Xu, Qiang Wu, Zhedong Zheng, Zhaoxiang Zhang, and Jian Zhang. Multi-pseudo regularized label for generated data in person re-identification. IEEE Transactions on Image Processing, 28(3):1391 1403, 2019. [Isola et al., 2017] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1125 1134, 2017. [Lin et al., 2019] Wu Lin, Wang Yang, and Shao Ling. Cycle-consistent deep generative hashing for cross-modal retrieval. IEEE transactions on image processing: a publication of the IEEE Signal Processing Society, 28(4):1602, 2019. [Liu et al., 2016a] Hongye Liu, Yonghong Tian, Yaowei Yang, Lu Pang, and Tiejun 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. [Liu et al., 2016b] Xinchen Liu, Wu Liu, Tao Mei, and Huadong Ma. A deep learning-based approach to progressive vehicle re-identification for urban surveillance. In European Conference on Computer Vision, pages 869 884. Springer, 2016. [Liu et al., 2018] Xinchen Liu, Wu Liu, Tao Mei, and Huadong Ma. Provid: Progressive and multimodal vehicle reidentification for large-scale urban surveillance. IEEE Transactions on Multimedia, 20(3):645 658, 2018. [Lou et al., 2019] Yihang Lou, Yan Bai, Jun Liu, Shiqi Wang, and Ling-Yu Duan. Embedding adversarial learning for vehicle re-identification. IEEE Transactions on Image Processing, 2019. [Taigman et al., 2016] Yaniv Taigman, Adam Polyak, and Lior Wolf. Unsupervised cross-domain image generation. ar Xiv preprint ar Xiv:1611.02200, 2016. [Tang et al., 2019] Zheng Tang, Milind Naphade, Ming-Yu Liu, Xiaodong Yang, Stan Birchfield, Shuo Wang, Ratnesh Kumar, David Anastasiu, and Jenq-Neng Hwang. Cityflow: A city-scale benchmark for multi-target multicamera vehicle tracking and re-identification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 8797 8806, 2019. [Wang et al., 2017] Zhongdao Wang, Luming Tang, Xihui Liu, Zhuliang Yao, Shuai Yi, Jing Shao, Junjie Yan, Shengjin Wang, Hongsheng Li, and Xiaogang Wang. Orientation invariant feature embedding and spatial temporal regularization for vehicle re-identification. In Proceedings of the IEEE International Conference on Computer Vision, pages 379 387, 2017. [Yi et al., 2017] Zili Yi, Hao Zhang, Ping Tan, and Minglun Gong. Dualgan: Unsupervised dual learning for imageto-image translation. In Proceedings of the IEEE international conference on computer vision, pages 2849 2857, 2017. [Zhao et al., 2019] Yanzhu Zhao, Chunhua Shen, Huibing Wang, and Shengyong Chen. Structural analysis of attributes for vehicle re-identification and retrieval. IEEE Transactions on Intelligent Transportation Systems, 2019. [Zheng et al., 2018] Zhedong Zheng, Liang Zheng, and Yi Yang. A discriminatively learned cnn embedding for person reidentification. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 14(1):13, 2018. [Zhu et al., 2017] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2223 2232, 2017.

Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20)