In unsupervised classification, because it is difficult to make sense of clusters in k-means and, the authors propose a two-step process: 1. to make the distance of the representation closer to the augmented own data, and 2. to classify the nearest neighbor data into the same cluster. 83.5% Accuracy in CIFAR10.
TL;DR
In unsupervised classification, because it is difficult to make sense of clusters in k-means and, the authors propose a two-step process: 1. to make the distance of the representation closer to the augmented own data, and 2. to classify the nearest neighbor data into the same cluster. 83.5% Accuracy in CIFAR10.
Why it matters:
Paper URL
https://arxiv.org/abs/2005.12320
Submission Dates(yyyy/mm/dd)
2020/05/25
Authors and institutions
Wouter Van Gansbeke, Simon Vandenhende, Stamatios Georgoulis, Marc Proesmans, Luc Van Gool
Methods
Results
Comments
ECCV2020