- a triplet sampling module
- 3 copies of a deep CNN image encoder to convert images into the corresponding latent vectors
- an image ranking layer; use max margin triplet loss measured in Euclidean distance in latent space
- a visualization module to plot the top 10 and bottom 10 similar images w.r.t. a query image
richardkxu/image-similarity-ranking
Image similarity ranking using deep CNN and triplet sampling
Python