This is regarding the DispNet network. I am using the DispNetCorr1D network to register images. I understand the network is only trained to register Right image to the left image. Thus the DispNetCorr1D always gives negative values. In my case, I want to get both positive and negative values of shift. How can I use the weights of the current network to register both Left to Right images and Right to Left images?
For example if I just do
Left to Right registration
python run-flownet-dispnet.py ../models/DispNetCorr1D/model/DispNetCorr1D_CVPR2016.caffemodel ../models/DispNetCorr1D/model/deploy.prototxt ../data/dispnet/imgL.png ../data/dispnet/imgR.png ../data/dispnet/floLR.flo
nikolausmayer
commented
5 years ago
Hi,
This is regarding the DispNet network. I am using the DispNetCorr1D network to register images. I understand the network is only trained to register Right image to the left image. Thus the DispNetCorr1D always gives negative values. In my case, I want to get both positive and negative values of shift. How can I use the weights of the current network to register both Left to Right images and Right to Left images? For example if I just do Left to Right registration python run-flownet-dispnet.py ../models/DispNetCorr1D/model/DispNetCorr1D_CVPR2016.caffemodel ../models/DispNetCorr1D/model/deploy.prototxt ../data/dispnet/imgL.png ../data/dispnet/imgR.png ../data/dispnet/floLR.flo
If I do Right to Left registration
python run-flownet-dispnet.py ../models/DispNetCorr1D/model/DispNetCorr1D_CVPR2016.caffemodel ../models/DispNetCorr1D/model/deploy.prototxt ../data/dispnet/imgR.png ../data/dispnet/imgL.png ../data/dispnet/floRL.flo
In the second case, the disparity map should have been positive.