Home

Awesome

Adma-GAN: Attribute-Driven Memory Augmented GANs for Text-to-Image Generation (ACMMM 2022)

Official Pytorch implementation for our paper [arxiv]

Pipeline

img

Experimental Results on CUB-200-2011 and COCO2014

img

Dataset

  1. Download the preprocessed CUB-200-2011 dataset. (including the extracted attribute information)
  2. Add it to data/CUB-200-2011
  3. Copy attribute/attributes_captions_embedding.pickle to data/CUB_200_2011/caption/

Training

cd Adma-GAN/code/
python train.py --config=cfg/bird.yml

Citing

If you find Adma-GAN useful in your research, please consider citing our paper:

@inproceedings{wu2022adma,
  title={Adma-GAN: Attribute-Driven Memory Augmented GANs for Text-to-Image Generation.},
  author={Wu, Xintian and Zhao, Hanbin and Zheng, Liangli and Ding, Shouhong and Li, Xi},
  booktitle={Proceedings of the 30th ACM International Conference on Multimedia},
  pages={1593--1602},
  year={2022}
}