coxlab / prednet

Code and models accompanying "Deep Predictive Coding Networks for Video Prediction and Unsupervised Learning"
https://arxiv.org/abs/1605.08104
MIT License
759 stars 259 forks source link

Custom Output Shape #51

Closed mitkina closed 5 years ago

mitkina commented 5 years ago

Hello,

I was wondering if there's an easy way to change prednet output shape to be different than the input in the model itself? I am trying to input 128x128x3 images and output 128x128x1 images. I changed the appropriate data_utils functions, but not sure what the best way to do it within prednet.py is?

Thanks! Masha

matinhosseiny commented 5 years ago

The input and output are the same sizes in formula 3 (Relu (A-Ahat)). You can change the design by adding a one by one convolution for input to reduce the input dimension.

mitkina commented 5 years ago

This would be only in layer zero, correct?

matinhosseiny commented 5 years ago

yes

mitkina commented 5 years ago

Thank you!