TagMe! Machine Learning Contest by IISC Bangalore
Duration: February - March 2014
Status: Completed
Members: Pranav Sodhani, Akshat Bordia
Guide: Dr. Suresh Sundaram
Overview
TagMe! was a machine learning programming contest organized by IISC Bangalore, India. The goal was to design an algorithm for tagging or classifying a collection of images belonging to different sets. The 5 categories of image sets included - 'Human Faces', 'Flowers', 'Buildings', 'Cars', and 'Shoes'. This is a multi-class supervised learning problem. Our approach exploited SURF for feature extraction and k-means for clustering. A modified bag-of-words algorithms was employed to form a voting strategy. The proposed algorithm demonstrated a classification accuracy of 92% on the training set.