Review — DeepCluster: Deep Clustering for Unsupervised Learning of Visual Features

DeepCluster, K-Mean Clustering to Generate Pseudo-Labels, a Pretext Task for Self-Supervised Learning

Illustration of the Proposed DeepCluster

Outline

1. Notations for Supervised Learning

2. DeepCluster as Pretext Task in Self-Supervised Learning

Top: k-Means Clustering on Vectors Produced by CNN; Bottom: Using the clustering results as psuedo labels for backpropagation

2.1. DeepCluster Procedures

2.2. Avoiding Trivial Solutions

2.2.1. Empty Cluster

2.2.2. Trivial Parametrization

3. DeepCluster Analysis

3.1. Normalized Mutual Information (NMI)

(a): Evolution of the clustering quality along training epochs; (b): evolution of cluster reassignments at each clustering step; (c): validation mAP classification performance for various choices of k

3.2. Visualizations

Filter visualization and top 9 activated images from a subset of 1 million images from YFCC100M
Top 9 activated images from a random subset of 10 millions images from YFCC100M for target filters in the last convolutional layer.

4. DeepCluster Performance

4.1. Linear Classification on Activations on ImageNet & Places

Linear classification on ImageNet and Places using activations from the convolutional layers of an AlexNet as features

4.1.1. ImageNet

4.1.2. Places

4.2. Pascal VOC

Comparison of the proposed approach to state-of-the-art unsupervised feature learning on classification, detection and segmentation on Pascal VOC

4.3. YFCC100M

Impact of the training set on the performance of DeepCluster measured on the Pascal VOC transfer tasks

4.4. AlexNet vs VGGNet

Pascal VOC 2007 object detection with AlexNet and VGG16

4.5. Image Retrieval

mAP on instance-level image retrieval on Oxford and Paris dataset with a VGG16

PhD, Researcher. I share what I've learnt and done. :) My LinkedIn: https://www.linkedin.com/in/sh-tsang/, My Paper Reading List: https://bit.ly/33TDhxG