stavc / mmn12computervisionuniversity

Computer Vision multilabel classification with different techniques: (KNN,PCA,LDA,BOVW,SIFT,VGG16)

MMN12ComputerVisionUniversity

Computer Vision multilabel classification with different techniques: (KNN,PCA,LDA,BOVW,SIFT,VGG16)

The full Report can be found at the Doc file.

Question 1:

Classify categories of MNIST Fashion by using KNN,PCA, LDA:

KNN Accuracy (takes alot of time to fit)

PCA with 20 Main Components(Good accuracy and fast fitting)

LDA with 8 componenets(fast fitting and ok accuracy)

transforming picture to lower dimension with PCA and inversing back

Question 2:

multilabel classification with SIFT or VGG16 as Feature extractor, Bag Of Visual Words representation and classification with LinearSVM.

DataBase: http://people.csail.mit.edu/torralba/code/spatialenvelope/spatial_envelope_256x256_static_8outdoorcategories.zip

Sift Method

VGG Method + ROC + Confusion Matrix