Object-wise Masked Autoencoders for Fast Pre-training

Jiantao Wu, Shentong Mo

Self-supervised pre-training for images without labels has recently achieved promising performance in image classification. The success of transformer-based methods, ViT and MAE, d…