search

ducha-aiki / affnet

Code and weights for local feature affine shape estimation paper "Repeatability Is Not Enough: Learning Discriminative Affine Regions via Discriminability"
MIT License
265 stars 47 forks source link

Is the experiments of Oxford Dataset retrieval task using BoW available? #19

Closed appleleaves closed 5 years ago

appleleaves commented 5 years ago

I do not find any BoW codes in the repository, but you report the BoW performance. I am very interested in this experiment and hope you can help me!

ducha-aiki commented 5 years ago

Hi,

Things regarding feature extraction are here:

https://github.com/ducha-aiki/mods-light-zmq#if-you-need-to-extract-and-save-features-from-directory-with-images

These gives you features and geometry. Regarding BoW itself, unfortunately, it is lab`s proprietary matlab code.

You could try this ASMK from here, but it is not the same

http://cmp.felk.cvut.cz/~toliageo/soft.html

appleleaves commented 5 years ago

When clustering with kmneas methods into 1M vocabulary, it takes a really long time and I do not see the end. Any advises?

ducha-aiki commented 5 years ago

1)Use faiss https://github.com/facebookresearch/faiss or 2)use approximate KNN method like kd-tree for calculating nearest cluster center.