Jingwei

Jingwei is an open-source testbed for evaluating methods for image tag assignment, tag refinement and tag-based image retrieval. It is developed as part of our survey effort, aiming to provide a timely reflection of the state-of-the-art in the field.

Methods implemented

Python Dependencies

• NLTK and nltk_data for SemanticField and RobustPCA.
• h5py and Numpy / Scipy for MATLAB<->Python data exchange of TagProp and RobustPCA.
• We recommend Anaconda (Python 2.7), a free Python distribution which include NLTK, h5py, numpy, scipy, etc.
• web.py for web demo

Training and Test Data

• A mini-version for hands-on experience
• Data dashboard, Florence data server

Setup

• Modify Paths in start.sh (for linux/mac) and start.bat for windows.

This file includes several environment variables that the methods depend on, to select proper input and output folders. From a shell, you can prepare the environment for using the framework with:

$ source start.sh 

• Configuration and Dependencies.

Depending on the method to be run, several different dependencies must be met and some external packages must be downloaded. The script setup.sh will report ready to run methods, depending on the available system packages. For some methods, it will also try to download and compile the provided libraries.

$ bash setup.sh 

Use a specific method

• Scripts in doit provide step-by-step usages of each method.
• Tutorials in samples show how to leverage the framework for solving varied tasks.

Citation

If you publish work that uses Jingwei, please cite our survey paper: Xirong Li, Tiberio Uricchio, Lamberto Ballan, Marco Bertini, Cees G. M. Snoek, Alberto Del Bimbo: Socializing the Semantic Gap: A Comparative Survey on Image Tag Assignment, Refinement and Retrieval, ACM Computing Surveys (CSUR), Volume 49, Issue 1, 14:1-14:39, June 2016

Contact

• Xirong Li, Renmin University of China ([email protected])
• Tiberio Uricchio, University of Florence ([email protected])