We study new semi-supervised video object segmentation (VOS) techniques for objects with rapid appearance changes, such as water (flood, wave, etc.).
Accurate semantic segmentation of water and surrounding object from videos captured in the fields (e.g. survalence cameras, traffic cameras) has direct applications in esimating water level and constructing flood hydrographs in urban areas in real-time during flash floods, hurricanes, and other extreme weather events, which remains as a difficult task
Semantic segmentation of water segmentation is technically challenging because water often has rapidly changing appearance caused by free-form self-deformation, environment illumination, reflections, wave, ripples, turbulence, sediment concentration, etc. Such rapidly changing appearance often leads to poor water segmentation in videos
We have maintained a public data benchmark of video water segmentation and a series of models for effective water segmentation.
Video Object Segmentation with Adaptive Feature Bank and Uncertain-Region Refinement
Yongqing Liang, Xin Li, Navid Jafari, Qin Chen
Neural Information Processing Systems (NIPS), 2020