Person Search

• Person Search

Person search has drawn considerable attention recently with the increasing demand for person re-identification from real world scene images. The conventional person re-identification takes a set of cropped person images and selects the images among them matching to a given query person. On the other hand, the person search methods directly find the query person from a set of scene images where multiple people may appear in a single image. Since the person search works on unconstrained images, it has a strong potential to be applied to numerous practical applications. For example, it can be used to deploy visual surveillance systems to monitor and trace the suspects from CCTV video sequences. Also, it can be applied to augmented reality systems with mobile camera devices to provide visual information useful for social entertainment.

On the other hand, collecting and labeling datasets for person search require significant cost of time and labor. To reduce the burden of labeling, we proposed weakly supervised person search for the first time, which does not require the identity labels of each person. Recently, to completely mitigate the burden of labeling and collecting dataset and also privacy concern, we utilized unreal dataset for training the person search network, introducing domain generalizable framework for the first time to alleviate domain gap between unreal and real datasets.

• Publications

[1]	Byeong-Ju Han, Kuhyeun Ko, and Jae-Young Sim. “End-to-end trainable trident person search network using adaptive gradient propagation.” Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021.
[2]	Byeong-Ju Han, Kuhyeun Ko, and Jae-Young Sim. “Context-aware unsupervised clustering for person search.”, 32nd British Machine Vision Conference 2021
[3]	Minyoung Oh, Duhyun Kim, and Jae-Young Sim. “Domain Generalizable Person Search Using Unreal Dataset.”, AAAI2024