For tiny images I went ahead and normalized the images by subtracting the mean and dividing by the standard deviation. For build_vocab I iterated through all the images and got sift features for each then clustered. Similary for bag_of_sift I just did a double for loop through the images, got the distances and incremented the appropiate part of the histogram for each feature. For the svm I just grouped the data based on its label, trained the relevant svm parameters and saved the result of applying the svm's to the data. Then I grabbed the svm with the highest output.
Implementation: For tiny images with knn I used a k value of ten. I got this by just testing different k values to find the best one. For the best performance of svm I had a vocab of 500 and a step size of 5 for the vocab building and the bag_of_sift step. Lambda for the best run was set to .00001. K= 10 still seemed to work the best for the knn plus bag_of_sifts model (using the same parameters as for svm), so that's what I used for the best knn plus bag_of_sifts model. Shown below are my best results. In order: tiny images with nn, svm with bag of sifts, and nn with bag of sifts. For the fast runs: I ran bag of sifts with a step size of 15 leaving the other paramters for nn and svm the same. Tiny has no change.
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label | ||||
---|---|---|---|---|---|---|---|---|---|
Kitchen | 0.090 | Store |
Street |
Suburb |
OpenCountry |
||||
Store | 0.030 | Forest |
InsideCity |
Coast |
Coast |
||||
Bedroom | 0.140 | Street |
Kitchen |
Highway |
Highway |
||||
LivingRoom | 0.100 | Bedroom |
TallBuilding |
Mountain |
Street |
||||
Office | 0.190 | Kitchen |
Kitchen |
Bedroom |
Coast |
||||
Industrial | 0.080 | Store |
Office |
Office |
Highway |
||||
Suburb | 0.490 | Street |
Coast |
Street |
Street |
||||
InsideCity | 0.070 | Coast |
Mountain |
OpenCountry |
LivingRoom |
||||
TallBuilding | 0.180 | Store |
LivingRoom |
Bedroom |
OpenCountry |
||||
Street | 0.510 | Office |
Store |
Mountain |
Suburb |
||||
Highway | 0.770 | Kitchen |
Industrial |
OpenCountry |
Mountain |
||||
OpenCountry | 0.370 | InsideCity |
Suburb |
InsideCity |
Suburb |
||||
Coast | 0.290 | TallBuilding |
Bedroom |
Highway |
Highway |
||||
Mountain | 0.160 | Coast |
Highway |
Office |
Highway |
||||
Forest | 0.090 | Mountain |
InsideCity |
Street |
Bedroom |
||||
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label |
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label | ||||
---|---|---|---|---|---|---|---|---|---|
Kitchen | 0.440 | Office |
LivingRoom |
Store |
Bedroom |
||||
Store | 0.580 | Industrial |
Industrial |
Highway |
InsideCity |
||||
Bedroom | 0.400 | Street |
LivingRoom |
TallBuilding |
OpenCountry |
||||
LivingRoom | 0.390 | TallBuilding |
OpenCountry |
Store |
Street |
||||
Office | 0.830 | Kitchen |
Store |
LivingRoom |
TallBuilding |
||||
Industrial | 0.560 | Store |
Coast |
Suburb |
LivingRoom |
||||
Suburb | 0.940 | Coast |
InsideCity |
OpenCountry |
Industrial |
||||
InsideCity | 0.480 | LivingRoom |
Industrial |
Store |
Industrial |
||||
TallBuilding | 0.670 | Mountain |
Bedroom |
Industrial |
Bedroom |
||||
Street | 0.690 | TallBuilding |
Kitchen |
Highway |
TallBuilding |
||||
Highway | 0.810 | Store |
Industrial |
Bedroom |
LivingRoom |
||||
OpenCountry | 0.540 | TallBuilding |
Industrial |
Highway |
Store |
||||
Coast | 0.710 | OpenCountry |
OpenCountry |
OpenCountry |
OpenCountry |
||||
Mountain | 0.740 | OpenCountry |
Highway |
Forest |
OpenCountry |
||||
Forest | 0.930 | Store |
Store |
Mountain |
OpenCountry |
||||
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label |
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label | ||||
---|---|---|---|---|---|---|---|---|---|
Kitchen | 0.430 | Suburb |
Street |
LivingRoom |
InsideCity |
||||
Store | 0.740 | InsideCity |
InsideCity |
Bedroom |
LivingRoom |
||||
Bedroom | 0.390 | Store |
Suburb |
LivingRoom |
TallBuilding |
||||
LivingRoom | 0.520 | Kitchen |
Kitchen |
Office |
Bedroom |
||||
Office | 0.740 | LivingRoom |
Bedroom |
Kitchen |
Kitchen |
||||
Industrial | 0.250 | OpenCountry |
TallBuilding |
OpenCountry |
Suburb |
||||
Suburb | 0.840 | OpenCountry |
Office |
Office |
Street |
||||
InsideCity | 0.270 | Street |
Street |
Store |
Store |
||||
TallBuilding | 0.240 | Bedroom |
Forest |
Street |
Street |
||||
Street | 0.590 | TallBuilding |
Mountain |
Store |
InsideCity |
||||
Highway | 0.830 | OpenCountry |
Mountain |
Store |
Street |
||||
OpenCountry | 0.390 | Coast |
Coast |
Suburb |
Suburb |
||||
Coast | 0.380 | OpenCountry |
Mountain |
Store |
Highway |
||||
Mountain | 0.510 | Coast |
Forest |
Store |
Forest |
||||
Forest | 0.950 | OpenCountry |
OpenCountry |
Suburb |
TallBuilding |
||||
Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label |
Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.