search

AliaksandrSiarohin / motion-cosegmentation

Reference code for "Motion-supervised Co-Part Segmentation" paper
Other
653 stars 142 forks source link

some problems about model training #37

Open newExplore-hash opened 3 years ago

newExplore-hash commented 3 years ago

hi AliaksandrSiarohin, so awesome project, about model training, i find there is no discriminator (include updating discriminator parameters) in the script of train.py, just have generator and segmentation, is there any problem when i retrain motion-cosegmentation?

AliaksandrSiarohin commented 3 years ago

Discriminator is not used for motion segmentation. Moreover it also not strictly needed for Fomm.

newExplore-hash commented 3 years ago

Discriminator is not used for motion segmentation. Moreover it also not strictly needed for Fomm.

yeah i understand this is a reconstruction process, and there is no adversarial loss, in other words, no adversarial learning in this process. so discriminator is not strictly needed for motion segmentation. Is that so? And if i retrain this model and add discriminator with adv learning for training process, will it improve the effect?

AliaksandrSiarohin commented 3 years ago

I Don't think it will change something.