Automatic Image Annotation: the quirks and what works

This repository contains the code for our paper published in Multimedia Tools and Applications: https://doi.org/10.1007/s11042-018-6247-3

• Non deep learning annotation models like 2PKNN, Tagprop, Tagrel, SVM, JEC
• Deep learning annotation models based on multi label loss functions like Softmax, Sigmoid, Pairwise Ranking, WARP, LSEP
• Empirical Experiments as per the paper
  • Per Label vs Per Image Evaluation Criteria
  • Dataset Specific Biases

Setup and Non-deep learning annotation models

Refer to run.md

Deep learning annotation models

Repo: https://github.com/ayushidutta/cnn-image-classification

Empirical analysis

Refer to analysis.md

Datasets

The 'data' folder contains the train/test splits of all datasets used in this experiment. For images, please refer to the individual dataset's page.

Requirements

• MATLAB
• Python 2
• Tensorflow 1.3 for Deep learning annotation models