This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision. Our MAE approach is simple: we mask random patches of the input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask tokens. Second, we find that masking a high proportion of the input image, e.g., 75%, yields a nontrivial and meaningful self-supervisory task. Coupling these two designs enables us to train large models efficiently and effectively: we accelerate training (by 3x or more) and improve accuracy. Our scalable approach allows for learning high-capacity models that generalize well: e.g., a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pre-training and shows promising scaling behavior.
title={Masked autoencoders are scalable vision learners},
author={He, Kaiming and Chen, Xinlei and Xie, Saining and Li, Yanghao and Doll{\'a}r, Piotr and Girshick, Ross},
journal={arXiv preprint arXiv:2111.06377},
year={2021}
}
```
## Usage
To use other repositories' pre-trained models, it is necessary to convert keys.
We provide a script [`beit2mmseg.py`](../../tools/model_converters/beit2mmseg.py) in the tools directory to convert the key of MAE model from [the official repo](https://github.com/facebookresearch/mae) to MMSegmentation style.
Since relative position embedding requires the input length and width to be equal, the sliding window is adopted for multi-scale inference. So we set min_size=512, that is, the shortest edge is 512. So the multi-scale inference of config is performed separately, instead of '--aug-test'. For multi-scale inference:
