Developing a platform for automatic real estate image tagging and keyword based search
The goal is to build a platform for potential real estate sellers to upload their property images and have them tagged automatically (with no or very little manual effort) according to the scene attributes such as room type, architecture, styling, etc. These generated tags can then be utilized to develop an efficient keyword based search of real estate properties.
Multiple feature engineering techniques are used to train a robust house image classifier. Different CNN architectures are used along with multiple open source and proprietary image datasets to enhance the classification accuracy.