README.md

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 as a baseline.
• 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 (ACPR) (pp. 3-16). Springer, Cham.
• CHILLAX Brust, C. A., Barz, B., & Denzler, J. (2021, January). Making Every Label Count: Handling Semantic Imprecision by Integrating Domain Knowledge. In 2020 25th International Conference on Pattern Recognition (ICPR) (pp. 6866-6873). IEEE.
• Self-Supervised CHILLAX Brust, C. A., Barz, B., & Denzler, J. (2021, April). Self-Supervised Learning from Semantically Imprecise Data. arXiv preprint arXiv:2104.10901.
• 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 == 0.14.0
• tensorflow == 2.4.3

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},
}