Quickshift++: Provably Good Initializations for Sample-Based Mean Shift

About

We provide initial seedings to the Quick Shift clustering algorithm, which approximate the locally high-density regions of the data. Such seedings act as more stable and expressive cluster-cores than the singleton modes found by Quick Shift. We establish statistical consistency guarantees for this modification. We then show strong clustering performance on real datasets as well as promising applications to image segmentation.

Heinrich Jiang, Jennifer Jang, Samory Kpotufe · 2018

Related benchmarks

Task	Dataset	Result
Clustering	Wine	ARI0.39	53
Clustering	FMNIST	ARI16	43
Clustering	Cancer	ARI41	25
Clustering	Face	NMI0.79	20
Clustering	MNIST	NMI45	15
Clustering	Codon	F1 Score24	6
Clustering	Hepatitis	F1 Score0.71	6
Clustering	Frogs	F1 Score50	6