1 paper
Konstantinos Georgiou, Maofeng Tang, Hairong Qi
Masked image modeling (MIM) methods typically operate in either raw pixel space (reconstructing masked patches) or latent feature space (aligning with a pre-trained teacher). We pr…