Doubling down: sparse grounding with an additional, almost-matching caption for detection-oriented multimodal pretraining.

Giacomo Nebbia, Adriana Kovashka

Browse the full CVPR paper archive.