Adaptive Graph Guided Embedding for Multi-label Annotation

This repository contains code for our International Joint Conferences on Artificial Intelligence (IJCAI) 2018 paper: Adaptive Graph Guided Embedding for Multi-label Annotation (AG2E). AG2E utiluzes existing small scale multi-label datasets to recovery/annotate the large scale images in semi-supervised scenario. It designed an adaptive graph and a domain-invariable projection, which trained simultaneously to achieve the high performance.

Multi-label annotation is challenging since a large amount of well-labeled training data are required to achieve promising performance. However, providing such data is expensive while unlabeled data are widely available. To this end, we propose a novel Adaptive Graph Guided Embedding (AG2E) approach for multi-label annotation in a semi-supervised fashion, which utilizes limited labeled data associating with large-scale unlabeled data to facilitate learning performance. Specifically, a multi-label propagation scheme and an effective embedding are jointly learned to seek a latent space where unlabeled instances tend to be well assigned multiple labels. Furthermore, a locality structure regularizer is designed to preserve the intrinsic structure and enhance the multi-label annotation. We evaluate our model in both conventional multi-label learning and zero-shot learning scenario. Experimental results demonstrate that our approach outperforms other compared state-of-the-art methods.

The annotation results above demonstrates the effectiveness and high robustness of our approach.

Running the code

The code is MATLAB code works in Ubuntu system. Windows system may need minor revision in the folder name. (Change the folder name xx/xx (Linux system) to xx\xx (Windows version)) should works well to switch the version between these to systems.

Code file introduction:

AG2E_demo.m -- Directly run this demo file in MATLAB could show the performance similar to our paper. It contains the dataset loading, AG2E approach function, and results output sections.

AG2E.m -- It's the implementation of AG2E approach. Please find the specific input/output instructions in the function comments.

Authors

Welcome to send us Emails if you have any questions about the code and our work :-)

• Lichen Wang Website
• Zhengming Ding Website
• Yun Raymond Fu Website

Citation

This code is corresponding to our IJCAI 2018 paper below:

@inproceedings{AG2E_IJCAI18_Lichen,
  title={Adaptive Graph Guided Embedding for Multi-label Annotation.},
  author={Wang, Lichen and Ding, Zhengming and Fu, Yun},
  booktitle={International Joint Conferences on Artificial Intelligence, IJCAI},
  pages={2798--2804},
  year={2018}
}

Please cite our paper if you like or use our work for your research, thank you very much!