Conditional Similarity Networks

TL;DR

Conditional Similarity Networks (CSNs) enable learning interpretable, multi-faceted image embeddings by disentangling features into semantic subspaces, outperforming specialized models in capturing diverse similarity notions.

Contribution

The paper introduces CSNs, a novel approach that learns disentangled, multi-faceted image embeddings with semantic subspaces and masks, capturing multiple similarity notions in a single model.

Findings

CSNs learn interpretable, semantically meaningful subspaces.

CSNs outperform specialized networks on multiple similarity tasks.

Embeddings are visually and semantically disentangled.

Abstract

What makes images similar? To measure the similarity between images, they are typically embedded in a feature-vector space, in which their distance preserve the relative dissimilarity. However, when learning such similarity embeddings the simplifying assumption is commonly made that images are only compared to one unique measure of similarity. A main reason for this is that contradicting notions of similarities cannot be captured in a single space. To address this shortcoming, we propose Conditional Similarity Networks (CSNs) that learn embeddings differentiated into semantically distinct subspaces that capture the different notions of similarities. CSNs jointly learn a disentangled embedding where features for different similarities are encoded in separate dimensions as well as masks that select and reweight relevant dimensions to induce a subspace that encodes a specific similarity notion. We show that our approach learns interpretable image representations with visually relevant semantic subspaces. Further, when evaluating on triplet questions from multiple similarity notions our model even outperforms the accuracy obtained by training individual specialized networks for each notion separately.