FFF: Fixing Flawed Foundations in Contrastive Pre-Training Results in Very Strong Vision-Language Models

Benchmark Model Rank Results
zero-shot-cross-modal-retrieval-on-flickr30kFFF-ViT-B/32 (Open30M) (extra data)–Image-to-text R@1: 87.9
zero-shot-cross-modal-retrieval-on-flickr30kFFF-ViT-B/32 (Open70M) (extra data)–Text-to-image R@1: 72.9
zero-shot-transfer-image-classification-on-caltech-101FFF-ViT-B/32 (Open70M) (extra data)–Accuracy: 86.5
zero-shot-transfer-image-classification-on-fgvc-aircraftFFF-ViT-B/32 (Open70M) (extra data)–Accuracy: 5.3