https://github.com/kyegomez/gen1

My Implementation of " Structure and Content-Guided Video Synthesis with Diffusion Models" by RunwayML
https://github.com/kyegomez/gen1

diffusion opensource opensource-projects opensourceforgood stable stable-diffusion-webui texttoimage texttovideo

Last synced: about 1 month ago
JSON representation

My Implementation of " Structure and Content-Guided Video Synthesis with Diffusion Models" by RunwayML

• Host: GitHub
• URL: https://github.com/kyegomez/gen1
• Owner: kyegomez
• License: mit
• Created: 2023-08-28T23:42:21.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-01-16T17:05:57.000Z (11 months ago)
• Last Synced: 2024-10-13T15:22:15.958Z (2 months ago)
• Topics: diffusion, opensource, opensource-projects, opensourceforgood, stable, stable-diffusion-webui, texttoimage, texttovideo
• Language: Python
• Homepage: https://discord.gg/qUtxnK2NMf
• Size: 305 KB
• Stars: 26
• Watchers: 4
• Forks: 3
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
[![Multi-Modality](agorabanner.png)](https://discord.gg/qUtxnK2NMf)

# Gen1

My Implementation of " Structure and Content-Guided Video Synthesis with Diffusion Models" by RunwayML. "Input videos x are encoded to z0 with a fixed encoder E and diffused to zt. We extract a

structure representation s by encoding depth maps obtained with MiDaS, and a content representation c by encoding one of the frames

with CLIP. The model then learns to reverse the diffusion process in the latent space, with the help of s, which gets concatenated to zt, as

well as c, which is provided via cross-attention blocks. During inference (right), the structure s of an input video is provided in the same

manner. To specify content via text, we convert CLIP text embeddings to image embeddings via a prior."

# Install

`pip3 install gen1`

# Usage

```python

import torch

from gen1.model import Gen1

# Create an instance of the Gen1 model

model = Gen1()

# Generate random input images and video tensors

images = torch.randn(1, 3, 128, 128)

video = torch.randn(1, 3, 16, 128, 128)

# Pass the input images and video through the model's forward method

run_out = model.forward(images, video)

```

## Datasets

Here is a summary table of the datasets used in the Structure and Content-Guided Video Synthesis with Diffusion Models paper:

| Dataset | Type | Size | Domain | Description | Source |

|-|-|-|-|-|-|

| Internal dataset | Images | 240M | General | Uncaptioned images | Private |  

| Custom video dataset | Videos | 6.4M clips | General | Uncaptioned short video clips | Private |

| DAVIS | Videos | - | General | Video object segmentation | [Link](https://davischallenge.org/) |

| Stock footage | Videos | - | General | Diverse video clips | - |

## Citation

```

@misc{2302.03011,

Author = {Patrick Esser and Johnathan Chiu and Parmida Atighehchian and Jonathan Granskog and Anastasis Germanidis},

Title = {Structure and Content-Guided Video Synthesis with Diffusion Models},

Year = {2023},

Eprint = {arXiv:2302.03011},

```

# Todo

- [ ] Add training script

- [ ] Add in conditional text paramater to pass in text, not just images and or other videos