CS 143 / Project 3 / Scene Recognition with Bag of Words
Dat classification.
I implemented the barebones pipeline, without any extra credit. I got the best accuracy out of the bag of SIFT features with the SVM classifier, maxing out at 64.8%.
I used a large step size (32) in build vocabulary in order to make the speed a little more reasonable.
Results
Tiny images & Nearest neighbors
My tiny image feature feature code simply resized images into 16 x 16 without doing any cropping. Combined with my nearest neighbors implementation, this resulted in 20.5% accuracy. It's not a great accuracy, but this pipeline was pretty quick.
Bag of SIFTs & Nearest neighbors
My bag of SIFTS with nearest neighbors implementation resulted in 50.2% accuracy. MUCH BETTER!
Bags of SIFTS & SVM
My bag of SIFTS with SVM implementation resulted in 64.3%. GOOD!
Accuracy (mean of diagonal of confusion matrix) is 0.643
| Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label | ||||
|---|---|---|---|---|---|---|---|---|---|
| Kitchen | 0.520 |  |
 |
 |
 |
 InsideCity |
 Bedroom |
 Store |
 InsideCity |
| Store | 0.560 |  |
 |
 |
 |
 InsideCity |
 Industrial |
 Kitchen |
 InsideCity |
| Bedroom | 0.490 |  |
 |
 |
 |
 Kitchen |
 LivingRoom |
 LivingRoom |
 LivingRoom |
| LivingRoom | 0.420 |  |
 |
 |
 |
 Street |
 Industrial |
 Bedroom |
 Kitchen |
| Office | 0.780 |  |
 |
 |
 |
 Bedroom |
 Kitchen |
 Kitchen |
 LivingRoom |
| Industrial | 0.540 |  |
 |
 |
 |
 Bedroom |
 TallBuilding |
 Forest |
Store |
| Suburb | 0.950 |  |
 |
 |
 |
 OpenCountry |
 Office |
 OpenCountry |
 Store |
| InsideCity | 0.400 |  |
 |
 |
 |
Kitchen |
 Street |
 Highway |
 Store |
| TallBuilding | 0.690 |  |
 |
 |
 |
 LivingRoom |
 InsideCity |
 Industrial |
 Industrial |
| Street | 0.640 |  |
 |
 |
 |
 Store |
 InsideCity |
 LivingRoom |
 Industrial |
| Highway | 0.830 |  |
 |
 |
 |
 Mountain |
 Coast |
 Coast |
 Bedroom |
| OpenCountry | 0.440 |  |
 |
 |
 |
 Mountain |
 Mountain |
 Coast |
 Coast |
| Coast | 0.760 |  |
 |
 |
 |
 OpenCountry |
 InsideCity |
 OpenCountry |
 Highway |
| Mountain | 0.730 |  |
 |
 |
 |
 OpenCountry |
 Kitchen |
 Store |
 OpenCountry |
| Forest | 0.900 |  |
 |
 |
 |
 OpenCountry |
 InsideCity |
 OpenCountry |
 Street |
| Category name | Accuracy | Sample training images | Sample true positives | False positives with true label | False negatives with wrong predicted label | ||||
GOOD PERFORMANCE!