Awesome

GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis (CVPR 2023)

A high-quality, fast, and efficient text-to-image model

Official Pytorch implementation for our paper GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis by Ming Tao, Bing-Kun Bao, Hao Tang, Changsheng Xu.

Requirements

• python 3.9
• Pytorch 1.9
• At least 1x24GB 3090 GPU (for training)
• Only CPU (for sampling)

GALIP is a small and fast generative model which can generate multiple pictures in one second even on the CPU.

Installation

Clone this repo.

git clone https://github.com/tobran/GALIP
pip install -r requirements.txt

Install CLIP

Preparation (Same as DF-GAN)

Datasets

1. Download the preprocessed metadata for birds coco and extract them to data/
2. Download the birds image data. Extract them to data/birds/
3. Download coco2014 dataset and extract the images to data/coco/images/

Training

cd GALIP/code/

Train the GALIP model

• For bird dataset: bash scripts/train.sh ./cfg/bird.yml
• For coco dataset: bash scripts/train.sh ./cfg/coco.yml

Resume training process

If your training process is interrupted unexpectedly, set state_epoch, log_dir, and pretrained_model_path in train.sh to resume training.

TensorBoard

Our code supports automate FID evaluation during training, the results are stored in TensorBoard files under ./logs. You can change the test interval by changing test_interval in the YAML file.

• For bird dataset: tensorboard --logdir=./code/logs/bird/train --port 8166
• For coco dataset: tensorboard --logdir=./code/logs/coco/train --port 8177

Evaluation

Download Pretrained Model

• GALIP for COCO. Download and save it to ./code/saved_models/pretrained/
• GALIP for CC12M. Download and save it to ./code/saved_models/pretrained/

Evaluate GALIP models

cd GALIP/code/

set pretrained_model in test.sh

• For bird dataset: bash scripts/test.sh ./cfg/bird.yml
• For COCO dataset: bash scripts/test.sh ./cfg/coco.yml
• For CC12M (zero-shot on COCO) dataset: bash scripts/test.sh ./cfg/coco.yml

Performance

The released model achieves better performance than the paper version.

Sampling

Synthesize images from your text descriptions

• the sample.ipynb can be used to sample

Citing GALIP

If you find GALIP useful in your research, please consider citing our paper:

@inproceedings{tao2023galip,
  title={GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis},
  author={Tao, Ming and Bao, Bing-Kun and Tang, Hao and Xu, Changsheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={14214--14223},
  year={2023}
}

The code is released for academic research use only. For commercial use, please contact Ming Tao (陶明) (mingtao2000@126.com).

Reference

• DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis [code]