Home

Awesome

MicroAST(AAAI 2023 Oral)

[update 2/7/2023]

Official Pytorch code for "MicroAST: Towards Super-Fast Ultra-Resolution Arbitrary Style Transfer"

News:

We have released the codes for calculating SSIM/Style Loss/Content Loss metrics! See details at path metrics/.

Introduction:

MicroAST is a lightweight model that completely abandons the use of cumbersome pre-trained Deep Convolutional Neural Networks (e.g., VGG) at inference. Instead, two micro encoders (content and style encoders) and one micro decoder are utilized for style transfer. The content encoder aims at extracting the main structure of the content image. The style encoder, coupled with a modulator, encodes the style image into learnable dual-modulation signals that modulate both intermediate features and convolutional filters of the decoder, thus injecting more sophisticated and flexible style signals to guide the stylizations. In addition, to boost the ability of the style encoder to extract more distinct and representative style signals, it also introduces a new style signal contrastive loss. MicroAST is 5-73 times smaller and 6-18 times faster than the state of the art, for the first time enabling super-fast (about 0.5 seconds) arbitrary style transfer at 4K ultra-resolutions.

show

Environment:

Getting Started:

Clone this repo:

git clone https://github.com/EndyWon/MicroAST
cd MicroAST

Test:

Train:

Citation:

If you find the ideas and codes useful for your research, please cite the paper:

@inproceedings{wang2023microast,
  title={MicroAST: Towards Super-Fast Ultra-Resolution Arbitrary Style Transfer},
  author={Wang, Zhizhong and Zhao, Lei and Zuo, Zhiwen and Li, Ailin and Chen, Haibo and Xing, Wei and Lu, Dongming},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2023}
}

Acknowledgement:

We refer to some codes and ideas from AdaIN and DIN. Great thanks to them!