In this project, we solved the task of image search engine problem specifying in using natural language query to retrieve relevant images. We leveraged image features extracted by Residual Network (ResNet) combined them with natural language processing methods. We applied Random Forest to map query vectors to object tag space, Ridge Regression to map descriptions to image features and proposed an ensemble algorithm to blend our models. This paper gives a detailed description of data prepossessing, modeling and experiment results.
laiterry / image-search-engine Goto Github PK
View Code? Open in Web Editor NEWLicense: MIT License