chia 2.2.0

Concept Hierarchies for Incremental and Active Learning

CHIA: Concept Hierarchies for Incremental and Active Learning

CHIA is a collection of methods and helper functions centered around hierarchical classification in a lifelong learning environment. It forms the basis for some of the experiments and tools developed at Computer Vision Group Jena.

Methods

CHIA implements:

• One-Hot Classifier
• Probabilistic Hierarchical Classifier Brust, C. A., & Denzler, J. (2019, November). Integrating domain knowledge: using hierarchies to improve deep classifiers. In Asian Conference on Pattern Recognition (pp. 3-16). Springer, Cham.
• Semantic Label Sharing Fergus, R., Bernal, H., Weiss, Y., & Torralba, A. (2010, September). Semantic label sharing for learning with many categories. In European Conference on Computer Vision (pp. 762-775). Springer, Berlin, Heidelberg.

Datasets

The following datasets are integrated into CHIA:

• CORe50
• CUB200-2011
• (i)CIFAR-100
• ImageNet ILSVRC2012
• NABirds

Requirements

CHIA depends on:

• python-configuration == 0.7.1
• nltk ~= 3.5
• imageio ~= 2.6
• pillow ~= 7.1.0
• gputil ~= 1.4.0
• networkx ~= 2.4
• numpy ~= 1.18.5
• tensorflow-addons == 0.11.1
• tensorflow == 2.3.0

Optional dependencies:

• tables ~= 3.6.1
• pandas ~= 1.0.4
• sacred ~= 0.8.1
• pyqt5 ~= 5.15.0
• scikit-image ~= 0.17.2
• scikit-learn ~= 0.23.1
• scipy == 1.4.1
• matplotlib ~= 3.2.1

Installation

To install, simply run:

pip install chia

or clone this repository, and run:

pip install -U pip setuptools
python setup.py develop

We also include the shell script quick-venv.sh, which creates a virtual environment and install CHIA for you.

Getting Started

To run the example experiment which makes sure that everything works, use the following command:

python examples/experiment.py examples/configuration.json

After a few minutes, the last lines of output should look like this:

[DEBUG] [ExceptionShroud]: Leaving exception shroud without exception
[SHUTDOWN] [Experiment] Successful: True

Citation

If you use CHIA for your research, kindly cite:

Brust, C. A., & Denzler, J. (2019, November). Integrating domain knowledge: using hierarchies to improve deep classifiers. In Asian Conference on Pattern Recognition (pp. 3-16). Springer, Cham.

You can refer to the following BibTeX:

@inproceedings{Brust2019IDK,
author = {Clemens-Alexander Brust and Joachim Denzler},
booktitle = {Asian Conference on Pattern Recognition (ACPR)},
title = {Integrating Domain Knowledge: Using Hierarchies to Improve Deep Classifiers},
year = {2019},
}