Adaptive Masked Autoencoder Transformer for image classification

Xiangru Chen, Chenjing Liu, Peng Hu, Jie Lin, Yunhong Gong, Yingke Chen, Dezhong Peng, Xue Geng

Abstract: Highlights•Propose AMAT for pre-training ViT-based models in image classification.•Develop an adaptive masking mechanism in the pre-training and fine-tuning stages.•Propose a novel objective function to facilitate the training of the AMAT method.•AMAT gains 40% FLOPs reduction compared with the original ViT with higher accuracy.