1. 程式人生 > >2018 10-708 (CMU) Probabilistic Graphical Models {Lecture 22} [Applications in Computer Vision (cont’d) + Gaussian Process]

2018 10-708 (CMU) Probabilistic Graphical Models {Lecture 22} [Applications in Computer Vision (cont’d) + Gaussian Process]

take log to Normal distribution, it's L2 loss, which might be the cause of the blurry results of VAE

interpreting the meaning of latent vars is difficult

 

 

like a stretching: stretch left ones to more left and right ones to more right

 

 

 

 

 

 

 

 

 

 

 

we cant really tell which animal is exactly. seems like they are combinations of different animals.

mode collaspe is a problem that hasn't been resolved yet.

 

three eyes... => counting issues 

 

 

 

 


 

 

 

 

 

 

larger model (fully modelled CRF has much better results, but meanwhile lead to the much more computational cost) 

 

 

 

 

smooth kernel: local similarity

apperaance kernel: location similarity and feature similarity

so ususlly w1 > w2 

 

 

 

 

do the low-pass filtering and simplize it by convolution

 

 

a moving kernel

 

 

0 iteration: from an unary classifier

10 iteration: through CRF