Advertisement

COPS: A Real-Time Cross-Domain Object Part Segmentation System

  • Xueqing He
Conference paper
Part of the IFIP Advances in Information and Communication Technology book series (IFIPAICT, volume 546)

Abstract

Although the object part segmentation is widely applied to surveillance video analysis and smart recommendation and so on, however, it does not show a good performance in cross-domain testing. This means the segmentation model has to label various data in different scenarios and it is costly due to the time and labor cost. Accordingly, in the paper, we would like to propose a real-time cross-domain object part segmentation system (COPS) based on the work of Cross-domain Human Parsing via Adversarial Feature and Label Adaptation [2]. Several vital techniques are applied in this real-time cross-domain object part segmentation system, including object detection, object tracking, and cross-domain adaptation object part segmentation. Taking an unconstrained benchmark dataset with rich pixel-wise labeling as the source domain, the real-time cross-domain object part segmentation system aims to segment frames of target domain videos without any additional manual labeling in real-time. Compared with the traditional approaches, this system is demonstrated to be a highly efficient and useful one among most practical applications, and the exploration on the challenging issue will contribute our real-time cross-domain object part segmentation system and push human parsing into next step. Therefore, we would like to present the details of our real-time cross-domain object part segmentation system in the following parts.

Keywords

Cross-domain Object part segmentation Real-time system 

Notes

Acknowledgments

This work was performed and supported by the China Agricultural University. And we also would like to thank Defa Zhu for his technical and theoretical help.

References

  1. 1.
    Chen, L.C., Papandreou, G., Kokkinos, I., Murphy, K., Yuille, A.L.: DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. arXiv preprint arXiv:1606.00915 (2016)
  2. 2.
    Liu, S., et al.: Cross-domain human parsing via adversarial feature and label adaptation. In: AAAI (2018)Google Scholar
  3. 3.
    Liu, S., Zhang, T.Z., Cao, X.C., Xu, C.S.: Structural correlation filter for robust visual tracking. In: CVPR (2016)Google Scholar
  4. 4.
    Liang, X., Shen, X., Feng, J., Lin, L., Yan, S.: Semantic object parsing with graph LSTM. In: Leibe, B., Matas, J., Sebe, N., Welling, M. (eds.) ECCV 2016. LNCS, vol. 9905, pp. 125–143. Springer, Cham (2016).  https://doi.org/10.1007/978-3-319-46448-0_8CrossRefGoogle Scholar
  5. 5.
    McClosky, D., Charniak, E., Johnson, M.: Automatic domain adaptation for parsing. In: ACL (2010)Google Scholar
  6. 6.
    Mao, X.D., Li, Q., Xie, H.R., Lau, R.Y., Wang, Z., Smolley, S.P.: The grid: least squares generative adversarial networks. arXiv:1611.04076 (2016)
  7. 7.
    Ren, S.Q., He, K.M., Girshick, R., Sun, J.: Faster R-CNN: towards real-time object detection with region proposal networks. In: NIPS (2015)Google Scholar
  8. 8.
  9. 9.
    Zhang, Y.: Fully convolutional adaptation networks for semantic segmentation. In: CVPR (2018)Google Scholar

Copyright information

© IFIP International Federation for Information Processing 2019

Authors and Affiliations

  1. 1.China Agricultural UniversityBeijingPeople’s Republic of China

Personalised recommendations