Can HOG to work with BoW'2 with SVM or OSVM

[gkomix88]

Hi Ceasar and All,

Thanks for the great work in Accord.NET! It make everything sound possible :)

I m currently trying to apply object recognition using HOG with Bag of Visual Words and classifier by SVM or OSVM. However, it seem the HOG could not work with the BoW'2 class. For instance,

var kmodes = new KModes(numberOfWords, new Hamming());
var detector = new HistogramsOfOrientedGradients(9, 3, 6);
var hogBow = new BagOfVisualWords<FeatureDescriptor, double[]>(detector, kmodes);
hogBow.Compute(TrainImages.Values.ToArray());

Could anyone assist on this case? Thanks in advance.

[cesarsouza]

Hi there,

I will soon commit some changes that will facilitate the use of HOG and other feature extractors in the BoW framework. Sorry for the delay in answering though.

Regards,
Cesar

[cesarsouza]

Changes committed in e185da9

A new pre-release package will be available on NuGet soon.

[gkomix88]

Thanks! Cesar. Will try it now.

[hzawary]

Hi everybody :)

Is also there a way to use another descriptors such Haralick textural feature extractor or LBP and clustering algorithms such as SOM in BoW?!

Best,
H.Zavvari

[cesarsouza]

Hi Zawari,

Yes, it should have been possible to use any descriptor or clustering algorithms in BoW. However, some algorithms like SOM still haven't been updated to use the more recent clustering/classifier interfaces, and as such will not work yet.

Haralick should work though, but I am not sure if I have written a test to confirm yet.

Regards,
Cesar

[cesarsouza]

Fixed in 3.5.0.