Clustering image dataset consisting of 100,000 small traffic images in different dimensions into 14 clusters.

The accuracy is measured by an external index metric Normalized Mutual Information Score (NMI)

• Unifying dimensions based on data distribution
• Feature extraction (VGG16 or ResNet50)
• Optionally apply dimensionality reduction
• Experiment with several clustering algorithms and hyperparameter tuning to find the best one

• KMeans
• DBSCAN
• SpectralClustering
• AgglomerativeClustering
• OPTICS
• Cure