I tried to implement beer bottles detection approach like in this article ([login to view URL]), but faced some problems as a new in CV:
1) Sometimes model detects not bottle but glasses (and threshold doesn't help). So, probably, it's needed to train another custom model for bottle detection. I attach 70 labeled images in archive
2) I didn't implement PCA for cutting to 16 dimensions. Provbably this is the reason why bottle comparison doesnt work well.
3) Instead of annoy I used flann matcher.
Now my accuracy is about 50% and this doesn't sound good. I attched my Python code ([login to view URL]).
The whole idea of program is that it should give a name (now just a photo) of beer bottles from the image.
7 freelancers are bidding on average $24 for this job
My preferred method of freelancing is an interactive approach to project solving. I have an MSEE specializing in Digital Signal/Image/RF Processing. I do my work in MATLAB (expert). I also do Python programming.