https://github.com/ExplainableML/ImageSelect

Code for the paper "If at First You Don't Succeed, Try, Try Again: Faithful Diffusion-based Text-to-Image Generation by Selection"
https://github.com/ExplainableML/ImageSelect

Last synced: 13 days ago
JSON representation

Code for the paper "If at First You Don't Succeed, Try, Try Again: Faithful Diffusion-based Text-to-Image Generation by Selection"

• Host: GitHub
• URL: https://github.com/ExplainableML/ImageSelect
• Owner: ExplainableML
• License: mit
• Created: 2023-05-19T11:54:41.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-07-10T15:51:41.000Z (over 1 year ago)
• Last Synced: 2024-08-01T18:37:29.052Z (3 months ago)
• Language: Python
• Homepage:
• Size: 3.51 MB
• Stars: 27
• Watchers: 7
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-diffusion-categorized - [Code

README

        
# Faithful Diffusion-based Text-to-Image Generation by Selection

This is the  PyTorch code our work [If at First You Don't Succeed, Try, Try Again: Faithful Diffusion-based Text-to-Image Generation by Selection](https://arxiv.org/abs/2305.13308) .



  



## Setup

You can setup the environments required by [Stable Diffusion](https://huggingface.co/blog/stable_diffusion) and [ImageReward](https://github.com/THUDM/ImageReward)

## Demo

To generate an image for a given prompt, all you need to do is create the directories `./all_images` and `./best_images` and then run the command `python3 src/imageselect_demo.py --num_seeds  --prompt `. The best image selected by ImageReward will be saved in the directory `./best_images`. 

## Diverse1k Dataset

The prompts for the `Diverse1k` dataset can be found in `./data/1k_prompts.json` and the corresponding QA data (which can be used for the [TIFA](https://github.com/Yushi-Hu/tifa) evaluation) collected from GPT3.5 is provided in `./data/qa.json`. 

## References

If you find this work useful, please cite:

```

@article{karthik2023if,

  title={If at First You Don't Succeed, Try, Try Again: Faithful Diffusion-based Text-to-Image Generation by Selection},

  author={Karthik, Shyamgopal and Roth, Karsten and Mancini, Massimiliano and Akata, Zeynep},

  journal={arXiv preprint arXiv:2305.13308},

  year={2023}

}

```