search

carrierlxk / GraphMemVOS

Video Object Segmentation with Episodic Graph Memory Networks (ECCV2020 spotlight)
94 stars 8 forks source link

Prior Masks for Support Set #2

Open khaghanijavad opened 4 years ago

khaghanijavad commented 4 years ago

Hi, thanks for your interesting work. It seems to start the interference on a sequence, you need the first N=3 frames with their masks to form the support set. However, you just have the first frame and its mask at the interference time. I think that you need an auxiliary method to segment the second and third frame before using your method. How did you handle it? Thanks

carrierlxk commented 4 years ago

Hi khaghanijavad, For the second and their frame, we use the matching method directly for prediction. Then, we use these three frames to build the memory and predict the fourth frame.