some questions
masonwang513 opened this issue · 3 comments
masonwang513 commented
why is it correct to accumulate losses of several frames and compute gradient just based on last frame?
lyxok1 commented
According to the computation graph, output mask prediction at each frame will be included into gradient computation, not just the last frame.
shoutOutYangJie commented
Hi，Do you add youtube-vos train dataset for training?
lyxok1 commented
@shoutOutYangJie Yes, Youtube and DAVIS are hybrid to train the segmentation network, but no synthetic sequence from static image is used.