Skip to content

Incremental Label Propagation (ILP) - Incremental Semi-Supervised Learning from Streams for Object Classification

License

Notifications You must be signed in to change notification settings

johny-c/incremental-label-propagation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Incremental Label Propagation

This repository provides the implementation of our paper "Incremental Semi-Supervised Learning from Streams for Object Classification" (Ioannis Chiotellis*, Franziska Zimmermann*, Daniel Cremers and Rudolph Triebel, IROS 2018). All results presented in our work were produced with this code.

The code was developed in python 3.5 under Ubuntu 16.04. You can clone the repo with:

git clone https://github.com/johny-c/incremental-label-propagation.git
  • KITTI

The repository includes 64-dimentional features extracted from KITTI sequences compressed in a zip file (data/kitti_features.zip). The included files will be extracted automatically if one of the included experiments is run on KITTI.

  • MNIST

A script will automatically download the MNIST dataset if an experiment is run on it.

The repository includes scripts that replicate the experiments found in the paper, including:

  • Varying the number of labeled points or the ratio of labeled points in the data.
  • Varying the number of labeled or unlabeled neighbors considered for each node.
  • Varying the hyperparameter $$\theta$$ that controls the propagation area size.

To run an experiment with varying $$\theta$$:

python ilp/experiments/var_theta.py -d mnist

You can set different experiment options in the .yaml files found in the experimens/cfg directory.

WARNING:

The included experiment scripts compute and store statistics after every new data point, therefore the resulting output files are very large.

If you use this code in your work, please cite the following paper.

Ioannis Chiotellis*, Franziska Zimmermann*, Daniel Cremers and Rudolph Triebel, "Incremental Semi-Supervised Learning from Streams for Object Classification", in proceedings of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2018). (pdf)

*equal contribution

@InProceedings{chiotellis2018incremental,
  author = "I. Chiotellis and F. Zimmermann and D. Cremers and R. Triebel",
  title = "Incremental Semi-Supervised Learning from Streams for Object Classification",
  booktitle = iros,
  year = "2018",
  month = "October",
  keywords={stream-based learning, sequential data, semi-supervised learning, object classification},
  note = {{<a href="https://github.com/johny-c/incremental-label-propagation" target="_blank">[code]</a>} },
}

This work is released under the [MIT Licence].

Contact John Chiotellis ✉️ for questions, comments and reporting bugs.

About

Incremental Label Propagation (ILP) - Incremental Semi-Supervised Learning from Streams for Object Classification

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages