Bag of SIFT representation and linear SVM classifier

Implementation

Features were extracted using a bag of SIFT representation. First we built a vocabulary by taking a subset of sift features of a sampling from the training images, then did k means samplings on this list of features to make K (I used 400 to balance time and accuracy) centroids of features. However, the actual features for each image are a histogram of samplings of sift features for that image binned into the nearest neighbor vocabulary word (centroid). Each test image was classified using a linear SVM classifier, with optimal lambda being .0001.

Results

Category name	Accuracy	Sample training images		Sample true positives		False positives with true label		False negatives with wrong predicted label
Kitchen	0.560					Bedroom	Industrial	Industrial	Bedroom
Store	0.520					Street	TallBuilding	Industrial	Office
Bedroom	0.490					Street	Kitchen	Kitchen	Store
LivingRoom	0.400					Bedroom	Store	Store	Kitchen
Office	0.880					Bedroom	Kitchen	LivingRoom	LivingRoom
Industrial	0.610					Street	TallBuilding	Kitchen	Mountain
Suburb	0.950					OpenCountry	Industrial	LivingRoom	OpenCountry
InsideCity	0.450					Suburb	LivingRoom	Industrial	LivingRoom
TallBuilding	0.720					Street	OpenCountry	Mountain	Kitchen
Street	0.600					TallBuilding	InsideCity	Industrial	Industrial
Highway	0.810					Coast	Industrial	OpenCountry	Street
OpenCountry	0.550					Coast	Coast	Suburb	Coast
Coast	0.800					OpenCountry	OpenCountry	Highway	OpenCountry
Mountain	0.790					Forest	Bedroom	OpenCountry	Bedroom
Forest	0.910					OpenCountry	OpenCountry	Store	Mountain
Category name	Accuracy	Sample training images		Sample true positives		False positives with true label		False negatives with wrong predicted label