Billion-Scale Pretraining with Vision Transformers for Multi-Task Visual Representations

Benchmark Model Rank Results
image-classification-on-objectnetResNet-101 (JFT-300M)–Top-1 Accuracy: 49.1
image-classification-on-objectnetResNet-50 (JFT-300M)–Top-1 Accuracy: 42.5
image-classification-on-objectnetViT-B/16 (ANN-1.3B)–Top-1 Accuracy: 50.7
image-classification-on-objectnetViT-B/32–Top-1 Accuracy: 48.4