search

kukuruza / City-Project

Analyze traffic given a set of optical cameras in urban areas
0 stars 0 forks source link

train features via distorting single patches (for matching) #36

Closed kukuruza closed 9 years ago

kukuruza commented 9 years ago

Implement this paper: http://arxiv.org/pdf/1312.5242v3.pdf @Lotuslisa likes this paper

1) Extract patches from images, labelled by matches (@kukuruza , estimate 3 hours) 2) Train a CNN classifier, and extract features (@kukuruza, estimate 4 hours) 3) Evaluate features for matching (@Lotuslisa )

Lotuslisa commented 9 years ago

Yeah, this is the paper you gave me two weeks ago, and we talked last week. So we begin to train features via distorting single patches now? What could I do?

On Thu, Aug 27, 2015 at 4:35 PM, kukuruza notifications@github.com wrote:

Implement this paper: implement http://arxiv.org/pdf/1312.5242v3.pdf

— Reply to this email directly or view it on GitHub https://github.com/kukuruza/City-Project/issues/36.

kukuruza commented 9 years ago

Hey, I fixed my first comment

kukuruza commented 9 years ago

I'm closing this issue, after the discussion at Jose's meeting. @satwikkottur noted that we don't have such a diversity as the original paper, so the method will probably fail. Right now the training gives just simple noise, and accuracy around 1/1000.