search

foolwood / DCFNet

DCFNet: Discriminant Correlation Filters Network for Visual Tracking
MIT License
214 stars 66 forks source link

cos window multiple feature map (training) #10

Open pabloe4993 opened 6 years ago

pabloe4993 commented 6 years ago

Hi,

I noticed that while training, the CNN output (before DCF) is not mult with cos window (or hann) in contradiction to tracking pipeline that mult the CNN output.

Why?

he010103 commented 6 years ago

Maybe it is hard to backward if multiplying the hanning window when training

foolwood commented 6 years ago

Because the hanning window will 'tell' DCFNet where is the target. (will make the DCFNet think the object with high RGB is target.)

I want to train the network with hard cases.

@he010103 it is easy now. ^.^ I try to use pytorch recently (autograd).