Data consists of a training dataset consisting of 2000 images, intersparsed
between the airplane and cat class and a test dataset of the same size.
The dimensions of the dataset are (2000, 10), 10 stands for the word to vec
encoding of the descriptors for each image.
10 clusters of the SIFT features were taken and clustering was performed.
The 10-repr of the input represents the scaled count of the number of SIFT
features per cluster.
This gives a homogeneous representation of the input irrespective of the
number of SIFT features per image.
Models
The data is trained using both kNN and SVM (linear and gaussian kernel).
Standard python machine learning libraries have been used.
sklearn (For SVM, SVR, LinearSVC, LinearSVR)
numpy (data manipulation)
pandas (intermediate and long term storage)
h5py (dense effficient storage)
None of the hyperparameters have been changed. Also, given that the
test_data is just as big as the training, and that the input vector size
is much less than the training examples, data is dense enough to prevent
overfitting.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent