TIGP (SNHCC) -- Underwater Salient Object Detection
- LecturerProf. Yan-Tsung Peng (Department of Computer Science, National Chengchi University)
Host: TIGP (SNHCC) - Time2021-11-01 (Mon.) 14:00 ~ 16:00
- LocationVirtual only
Live Stream
Join the talk on 【Webex】, or open Webex and enter Meeting number: 2510 739 2296 and Password: iisnov
Abstract
There is little work done for underwater saliency objection detection (SOD), but it is vital to artificial intelligence-driven underwater analysis. Recent research has shown that depth information would increase SOD accuracy, but it may not be accessible to most RGB datasets. Since image blurriness could be an estimate of underwater scene depth, in this talk, we present to use a self-derived blurriness cue and fuse it into the RGB stream to boost SOD accuracy. Besides, we design a specific data augmentation method for underwater SOD tasks to further boost the performance. Our work also contributes a public underwater SOD dataset to the field of underwater SOD research.