search

kmaninis / OSVOS-caffe

One-Shot Video Object Segmentation
http://vision.ee.ethz.ch/~cvlsegmentation/osvos/
GNU General Public License v3.0
171 stars 67 forks source link

use boundary snap only #9

Open berylyellow opened 5 years ago

berylyellow commented 5 years ago

Hi，thx for your pro. I‘m doing post processing after segmentation，but I do not have good ideas. You mention boundary snap in the paper，but I do not know how to use it to test the effect. Could you help me？

kmaninis commented 5 years ago

Hi, there is code for snapping here: https://github.com/kmaninis/OSVOS-caffe/blob/master/src/ucms/snap.m

I hope this helps.

berylyellow commented 5 years ago

THX~ Is it that you train a CNN model to detect all counters in the image，and then you pick the counter which most fit the mask you get in foreground branch ？ The counter seems not that accurate in Qualitative results on DAVIS 2016 in your paper. Is it because of video detection or counter-detection model? I'm doing segmentation on pictures. If I use your counter-detection model and snap method, will I receive a better result? Do you have any good ideas how to refine the mask when we get segmentation mask by network? Or whether there are good ways to improve the final result? Thank you for your reply. Best wishes~