Programming Assignment 1

Yiming Liu (yimingl@princeton.edu)

Oct 9


Multiple Training Pairs

I have implemented 'multiple training pairs' in my program. It is very useful in super-resolution result. With this feature, even if the source image has several parts which look different with each other, as long as we provide the high-resolution images for these parts, the analogy algorithm can hallucinate the super-resolution result for us.

The following images compare the difference between the algorithm without multiple training pairs and with multiple training pairs. It can be observed that when we only use the 3rd training pair, the quality of the result is badly degraded.



Common parameters:

A and A'




B


Result without multiple training pairs (only use the 3rd pair)


Result with multiple training pairs