https://github.com/Nithin-GK/MaxFusion

[ECCV'24] MaxFusion: Plug & Play multimodal generation in text to image diffusion models
https://github.com/Nithin-GK/MaxFusion

diffusion-models multimodal plug-and-play txt2img

Last synced: 2 months ago
JSON representation

[ECCV'24] MaxFusion: Plug & Play multimodal generation in text to image diffusion models

• Host: GitHub
• URL: https://github.com/Nithin-GK/MaxFusion
• Owner: Nithin-GK
• License: mit
• Created: 2024-03-18T14:17:13.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-07-23T17:10:21.000Z (5 months ago)
• Last Synced: 2024-08-01T18:35:45.529Z (5 months ago)
• Topics: diffusion-models, multimodal, plug-and-play, txt2img
• Language: Jupyter Notebook
• Homepage: https://nithin-gk.github.io/maxfusion.github.io/
• Size: 8.7 MB
• Stars: 15
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-diffusion-categorized - [Code

README

        


 MaxFusion: Plug & Play multimodal generation in text to image diffusion models



 If you like our project, please give us a star ⭐ on GitHub for latest update.  




    

[![hf_space](https://img.shields.io/badge/🤗-Open%20In%20Spaces-blue.svg)]()

[![project page](https://img.shields.io/badge/Project%20Page-8A2BE2)](https://nithin-gk.github.io/maxfusion.github.io/)

[![arXiv](https://img.shields.io/badge/Arxiv-2404.09977-b31b1b.svg?logo=arXiv)](https://arxiv.org/pdf/2404.09977) 


## Applications





Keywords: Multimodal Generation,   Text to image generation, Plug and Play

We propose **MaxFusion**, a plug and play framework for multimodal generation using text to image diffusion models.

    *(a) Multimodal generation*. We address the problem of conflicting spatial conditioning for text to iamge models .

    *(b) Saliency in variance maps*. We discover that the variance maps of different feature layers expresses the strength og conditioning.




### Contributions:

- We tackle the need for training with paired data for multi-task conditioning using diffusion models.

- We propose a novel variance-based feature merging strategy for diffusion models.

- Our method allows us to use combined information to influence the output, unlike individual models that are limited to a single condition.

- Unlike previous solutions, our approach is easily scalable and can be added on top of off-the-shelf models.

## Environment setup 

```

conda env create -f environment.yml

```

## Code demo:

A notebook for differnt demo conditions is provided in demo.ipynb

# Testing On custom datasets 

Will be released shortly

##  Instructions for Interactive Demo

An intractive demo can be run locally using

```

python gradio_maxfusion.py

```

This code is reliant on:

```

https://github.com/google/prompt-to-prompt/

```

## Citation

5. If you use our work, please use the following citation

```

@article{nair2024maxfusion,

  title={MaxFusion: Plug\&Play Multi-Modal Generation in Text-to-Image Diffusion Models},

  author={Nair, Nithin Gopalakrishnan and Valanarasu, Jeya Maria Jose and Patel, Vishal M},

  journal={arXiv preprint arXiv:2404.09977},

  year={2024}

}

```