Weakly Supervised Object Detection Based on Domain Transfer and Continuation Multiple Instance Learning

Xinpei Tian

《World Scientific Research Journal》 7卷6期 (2021/06) Pp. 443-452

https://doi.org/10.6911/WSRJ.202106_7(6).0056

摘要

This paper focuses on the problems of high object labeling cost during the detection process of strongly supervised object detection and lower accuracy of the detection results of weakly supervised object detection. This paper proposes an object detection model based on weakly supervised learning, which combines domain transfer technology and the Continuation Multiple Instance Learning algorithm. First, the CycleGAN domain transfer technology is used to convert the source domain image dataset into comic, watercolor, and clipart three types of dataset images similar to the object domain image. Then, the feature extraction of object domain image is accomplished by C‐MIL algorithm, and obtains the instance label estimation through the two modules of instance selection and instance partition; finally, using the obtained domain transfer image and pseudo‐label annotation image to fine‐tune the pre‐trained object detector in turn. The experimental results show that, compared with the detection results of the weakly supervised object detection of domain transfer, the improved model has a higher average accuracy of the detection results of the three datasets.

關鍵字

Weakly-supervised learning ； C‐MIL ； Domain transfer ； Pseudo-label annotation

參考文獻

Bilen H, Vedaldi A. Weakly Supervised Deep Detection Networks[C]// Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016:2846-2854.

W. Ge, W. Huang, S. Guo and M. Scott, "Label-PEnet: Sequential Label Propagation and Enhancement Networks for Weakly Supervised Instance Segmentation," 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), 2019, pp. 3344-3353, doi: 10.1109/ICCV.2019.00344.

P. Tang et al., "PCL: Proposal Cluster Learning for Weakly Supervised Object Detection," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 1, pp. 176-191, 1 Jan. 2020, doi: 10.1109/TPAMI.2018.2876304.

Y. Chen and W. H. Hsu, "Saliency Aware: Weakly Supervised Object Localization," ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, United Kingdom, 2019, pp. 1907-1911, doi: 10.1109/ICASSP.2019.8682756.

N. Inoue, R. Furuta, T. Yamasaki and K. Aizawa, "Cross-Domain Weakly-Supervised Object Detection Through Progressive Domain Adaptation," 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 2018, pp. 5001-5009.