ImageNet Classification with Multiple Classifiers

In this project we proposed an ensemble classifier to classify over 20 thousand images sampled from ImageNet, which originally has over 10 million images. One of the challenge of this classification problem is that the images cannot be precisely represented by one type of features, such as SIFT and GIST. Hence, in this project, we use different kinds of features. Another challenge is that different classification models perform differently on different feature set. Here, we use different classifiers (Kernel Regression and SVM) on different features and ensemble the classification results in a weighting manner. The accuracy of the ensemble classifier outperforms almost all of the baselines.