Pre:post
It is MY first time to see quite elegant a solution to seek a subspace for a group of local features. I list two related papers for your reference: 「Local Feature Discriminant Projection」 and 「Binary Set Embedding for Cross-Modal Retrieval」. This post is just a note which might leave out some important details and information. Following I will first introduce the LFDP algorithm then the BSE algorithm for cross-modality learning.ui