Tag
SimLoss uses embedding-space contrastive supervision to enable single-pass fine-grained image captioning that matches multi-stage quality at much lower latency, with variants like SimLoss FFT achieving high precision.