Requirements

python 3.6

pip3 install sklearn

pip3 install tensorflow

pip3 install opencv-python

pip3 install imutils

Dataset: http://www.vision.caltech.edu/Image_Datasets/Caltech256/

How to run

move the images into "training_images" directory

choose several test images and move them into "test_images" directory

run retrain.py

Then run test.py to see the results

Result

KNN

3 neighbors accuracy: 12.61%
4 neighbors accuracy: 12.94%
5 neighbors accuracy: 12.98%
6 neighbors accuracy: 13.13%
7 neighbors accuracy: 13.27%
8 neighbors accuracy: 13.53%
9 neighbors accuracy: 13.33%
10 neighbors accuracy: 13.55%

5 fold cross validation

best n_neighbors: 13, best accuracy: 13.20%

SVM

linear accuracy: 16.23%
poly accuracy: 16.05%
rbf accuracy: 0.74%
sigmoid accuracy: 0.65%

transfer learning

10000 training steps: 84.3%
8000 training steps: 83.9%
6000 training steps: 83.4%
4000 training steps: 82.1%

BP neural network

100 tanh accuracy: 11.81%
500 tanh accuracy: 13.46%
1000 tanh accuracy: 13.66%
2000 tanh accuracy: 12.59%
100 relu accuracy: 12.92%
500 relu accuracy: 14.16%
1000 relu accuracy: 14.68%
2000 relu accuracy: 15.79%

CNN

100 epochs: 24.93%
300 epochs: 27.25%