frankmnb/hyperparameter-analysis-in-online-constrained-clustering

Code for running various online constrained clustering methods with different parameter combinations.

Hyperparameter-Analysis-in-Online-Constrained-Clustering

This contains MATLAB code for analysing different combinations of parameters in online constrained clustering. These are the proportion of constraints, initial number of clusters, and the batch window size.

Usage

MainIterator.m is the main file from which all the functions are called. Edit the variables to incorporate your data and your desired hyperparameters.

Data

The data must be in .csv format and stored using the following formatting:

Bounding Box (BB) Frame Data:

BB top-left x	BB top-left y	BB width	BB height	frame number	frame width	frame height	object label	frame image name
354	259	276	247	1	952	540	3	Florence_frame_00001.jpg
342	408	140	132	1	952	540	5	JP_frame_00001.jpg
218	173	371	136	1	952	540	6	Jack_frame_00001.jpg
352	72	367	240	1	952	540	8	Selwyn_frame_00001.jpg
362	276	260	228	2	952	540	3	Florence_frame_00002.jpg
341	415	129	123	2	952	540	5	JP_frame_00002.jpg

Feature Data - see link below for examples 👇

Constraints Data:

Object A	Object B	ML/CL
1	5	0.84631511
2	6	0.884163644
3	7	0.754439876
8	9	-1
8	10	-1

where the decimal is the IoU between the two BB's (Must-link), and '-1' means both objects are in the same frame (Cannot-link).

Examples of the datasets as used in the code can be found here 👉 https://zenodo.org/record/7322821 .

Methods

• kul.m - Kulshreshtha method ¹
• online_cop_kmeans.m - Online COP-Kmeans
• osl.m - Online SIngle Linkage
• bla.m - BaseLine A "same"
• blb.m - BaseLine B "different"
• blc.m - BaseLine C "random

Metrics

• normalised_mutual_information.m - Normalised Mutual Information (NMI) ²
• adjusted_rand_index.m - Adjusted Rand Index (ARI) ²
• classification_accuracy.m - Classification Accuracy "Count"
• classification_accuracy_old.m - Classification Accuracy "Hungarian"

References

Footnotes

1. Kulshreshtha, Prakhar, and Tanaya Guha. "An online algorithm for constrained face clustering in videos." 2018 25th IEEE International Conference on Image Processing (ICIP). IEEE, 2018. ↩

2. Vinh, Nguyen Xuan, Julien Epps, and James Bailey. "Information theoretic measures for clusterings comparison: is a correction for chance necessary?." Proceedings of the 26th annual international conference on machine learning. 2009. ↩ ↩²