search

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

What is the Copy last frame? #57

Closed kimna4 closed 4 years ago

kimna4 commented 5 years ago

Hello, First of all, thank you for publishing great study.

The question is that What is the “Copy-Last-Frame”? I don’t know exactly how I can use this approach as a baseline. Could you tell me about the “Copy-Last-Frame”?

severin1992 commented 5 years ago

You compare the MSE of the following: 1.) Your t+1 prediction vs. the ground truth of t+1 2.) Ground truth of t+1 vs. ground truth of t

Here, 2.) is your baseline (copy last frame). The assumption is that my simplest prediction is just not changing anything. t+1 will exactly be like t. So, when your MSE of 1.) is lower than 2.) your model is actually better than your baseline. @kimna4