Paper#1 [MVOS-OL] Online Meta Adaptation for Fast Video Object Segmentation

There are two adaptations that update a segmentation network on video in inference time. The first part, meta-learner, updates the base network with the first input frame. Second, weights are fine-tuned with masks came from the following frames.

Overall architecture of MVOS-OL

• Title: Online Meta Adaptation for Fast Video Object Segmentation
• Paper: https://www.ncbi.nlm.nih.gov/pubmed/30640597
• Code(in PyTorch): https://github.com/huaxinxiao/MVOS-OL