1 paper · 1 filter
Donghuo Zeng
Contrastive loss and triplet loss are widely used objectives in deep metric learning, yet their effects on representation quality remain insufficiently understood. We present a the…