DarkLight Networks for Action Recognition in the Dark

Rui Chen, Jiajun Chen, Zixi Liang, Huaien Gao, Shan Lin; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 846-852

Abstract


Human action recognition in the dark is a significanttask with various applications, e.g., night surveillance andself-driving at night. However, the lack of video datasetsfor human actions in the dark hinders its development. Recently, a public dataset ARID has been introduced tostimulate progress for the task of human action recogni-tion in dark videos. Currently, there are multiple mod-els that perform well for action recognition in videos shotunder normal illumination. However, research shows thatthese methods may not be effective in recognizing actionsin dark videos. In this paper, we construct a novel neuralnetwork architecture: DarkLight Networks, which involves (i) a dual-pathway structure where both dark videos andits brightened counterpart are utilized for effective videorepresentation; and (ii) a self-attention mechanism, whichfuses and extracts corresponding and complementary fea-tures from the two pathways. Our approach achieves state-of-the-art results on ARID. Code is available at:https://github.com/Ticuby/Darklight-Pytorch

Related Material


[pdf] [supp]
[bibtex]
@InProceedings{Chen_2021_CVPR, author = {Chen, Rui and Chen, Jiajun and Liang, Zixi and Gao, Huaien and Lin, Shan}, title = {DarkLight Networks for Action Recognition in the Dark}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2021}, pages = {846-852} }