SJ rests in grass.
I did teh standard implementaion of the methods mentioned in handout. For SSD and SSD + min error path, I randomly sampled 500 patches and sort them based on SSD. I then randomly pick 1 from the lowest 5 ssd error patches. It performs reasonaly well on most textures. However, it did poorly on random3.png and apples.png. Those txtures have relative large space structures which might not be captured by the patch we choose.
This is also just the implementation from handout. However, I reduced the sample size to 200 to speed up. The transfer also works generally well, except for the albert einstein one, which is overexposed and has low variation on face brightness.
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() |
![]() ![]() ![]() |
![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() |