You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Dear Mehta,
Is the SSIM implementation correct for the GPU or is it still without the GPU implementation?
Plus, does it work if the weight is left as 1 or does it need to be scaled so that the network can be trained? Currently im trying it on 64x64 patches and im getting an SSIM of approximately 64, which should not be the case.
Also, I'm very new to caffe, so please excuse this simple question: Can this layer be used in conjunction with your L1 Loss layer?
Thank you.
The text was updated successfully, but these errors were encountered:
Yes, can be used in conjunction with the L1 loss layer.
You'll have to work out the loss weight yourself though, based on the problem you are using it for, your network, learning rate, solver, other losses .......
It is a DSSIM layer, and is not normalized by the number of pixels, so scales with it. The forward pass will tell you what is exactly happening in the code
Dear Mehta,
Is the SSIM implementation correct for the GPU or is it still without the GPU implementation?
Plus, does it work if the weight is left as 1 or does it need to be scaled so that the network can be trained? Currently im trying it on 64x64 patches and im getting an SSIM of approximately 64, which should not be the case.
Also, I'm very new to caffe, so please excuse this simple question: Can this layer be used in conjunction with your L1 Loss layer?
Thank you.
The text was updated successfully, but these errors were encountered: