The paper proposes a novel pretraining framework called Cross-Attention Masked Autoencoders (CrossMAE) that leverages cross-attention between masked and visible tokens, resulting in improved representation learning with reduced decoding compute. CrossMAE outperforms MAE on ImageNet classification and COCO instance segmentation.

https://arxiv.org/abs//2401.14391

YouTube: https://www.youtube.com/@ArxivPapers

TikTok: https://www.tiktok.com/@arxiv_papers

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers