notebook.community
Edit and run
Exercie 4 result is included into imagesearch.py Used cross-correlation instead of L2 distance. This made the otuput result far better, but not good enough It looks like comparison of sift features seems to take more than naive distance