Eye landmarks detection via two-level cascaded CNNs with multi-task learning
作者：
Highlights：
•
• We propose an eye landmarks detection algorithm, that aims to open and close eye images.
• We introduce the pipeline of creating OCE dataset, with eye images in different states.
• We train two-level cascaded CNNs for eye landmarks detection by using OCE dataset.
• CNNs at the first level predict good initial positions of eyes through multi-task learning with eye state estimation.
• Experimental results reveal our algorithms acceptable performance of eye localization.
摘要
•We propose an eye landmarks detection algorithm, that aims to open and close eye images.•We introduce the pipeline of creating OCE dataset, with eye images in different states.•We train two-level cascaded CNNs for eye landmarks detection by using OCE dataset.•CNNs at the first level predict good initial positions of eyes through multi-task learning with eye state estimation.•Experimental results reveal our algorithms acceptable performance of eye localization.
论文关键词：Eye landmarks detection,Two-level cascaded networks,Multi-task learning,OCE dataset,Eye state estimation
论文评审过程：Received 25 September 2017, Revised 22 December 2017, Accepted 27 January 2018, Available online 6 February 2018, Version of Record 16 February 2018.
论文官网地址：https://doi.org/10.1016/j.image.2018.01.008
