search

sconlyshootery / FeatDepth

This is the offical codes for the methods described in the "Feature-metric Loss for Self-supervised Learning of Depth and Egomotion".
MIT License
247 stars 28 forks source link

Train only Image Reconstruction Model #81

Closed ayushmankumar7 closed 2 years ago

ayushmankumar7 commented 2 years ago

Hello, Thank you for such a wonderful paper. I was working on Image reconstruction. Really liked your regularization using smooth_loss1 and smooth_loss2.

So after running a single epoch, we get a loss_dict from the Autoencoder which has the smooth loss and reconstruction loss(L1 + SSIM) at 4 different scales. My question is how is backpropagation working while we train the Image Reconstruction network?

Thank you!

sconlyshootery commented 2 years ago

The total loss is the sum of all the losses.

ayushmankumar7 commented 2 years ago

Thanks for replying! Yes, I saw that in your code.