README.md

---
viewer: false
tags:
- deepfakes
- gen-ai
- text-to-video
pretty_name: DeepAction Dataset v1.0
size_categories:
- 1K<n<10K
task_categories:
- video-classification
---

<style>
        * {
        font-family: Helvetica, sans-serif;
          }
        code {
            font-family: IBM Plex Mono,ui-monospace,SFMono-Regular,Menlo,Monaco,Consolas,Liberation Mono,Courier New,monospace !important;
        }
        a {
        color: #FFA500;
        }
        .container {
            display: flex;
            justify-content: space-between; /* Ensures even space between items */
            align-items: stretch; /* Ensures boxes have the same height */
            width: 100%;
            margin: 20px auto;
            gap: 20px; /* Consistent gap between boxes */
        }
        .warning-box {
            background-color: rgba(255, 200, 100, 0.5); /* Lighter orange with more translucency */
            border-radius: 10px;
            padding: 20px;
            flex: 1;
            box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
            font-family: Arial, sans-serif;
            color: #333;
            display: flex;
            flex-direction: column;
            justify-content: flex-start; /* Align items to the top */
        }
        .warning-sign {
            font-weight: bold;
            font-size: 1em;
            margin-bottom: 10px;
        }
        .warning-text {
            font-size: 1em;
        }
        .button {
            display: inline-block;
            padding: 10px 20px;
            margin: 5px;
            background-color: #FFA500;
            color: white;
            text-decoration: none;
            border-radius: 5px;
        }
        .button span {
            margin-right: 10px;
        }
        .button:hover {
            background-color: #E69500;
        }
        .warning {
            background-color: rgba(255, 165, 0, 0.2);
            border-left: 5px solid #FFA500;
            border-radius: 5px;
            padding: 10px;
            margin: 10px 0;
            color: #000 !important;
        }
        .warning .title {
            color: #FFA500;
            font-weight: bold;
            display: flex;
            align-items: center;
        }
        .warning .title span {
            margin-right: 10px;
        }

.warning-banner {
    display: flex;
    align-items: center;
    justify-content: start; /* Adjusted to align content to the start */
    background-color: #FFCC80; /* Adjusted to a darker shade of orange for better contrast */
    color: #333;
    padding: 10px 30px;
    border-radius: 8px;
    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1); /* Lighter shadow for subtlety */
    margin: 20px auto;
    width: 95%; /* Adjust width as needed */
    font-family: Helvetica, sans-serif;
}

.warning-icon {
    font-size: 1.5em;
    margin-right: 15px;
    color: #E65100; /* Darker orange for the icon */
}

.warning-message {
    font-size: 1em;
    font-weight: bold;
    flex: 1; /* Ensures message uses available space */
}

.warning-link {
    color: #0056b3; /* Standard link color for visibility */
    text-decoration: none; /* Removes underline */
}

.warning-link:hover {
    text-decoration: underline; /* Adds underline on hover for better interaction */
}
    </style>


<img src="https://data.matsworld.io/ucbresearch/deepaction.gif" style="width: 100%">

The DeepAction dataset contains over 3,000 videos generated by six text-to-video AI models, along with real videos matched in terms of the action depicted. These videos show people performing ordinary actions such as walking, running, and cooking. The AI models used to generate these videos include, in alphabetic order, AnimateDiff, CogVideoX5B, Pexels, RunwayML, StableDiffusion, Veo (pre-release version), and VideoPoet. Refer to our <a href='https://arxiv.org/abs/2412.00526'>our pre-print</a> for details.

<br>

## Getting Started

To get started, log into Hugging Face in your CLI environment, and run:

```python
from datasets import load_dataset
dataset = load_dataset("faridlab/deepaction_v1", trust_remote_code=True)
```

<br>

## Data

The data is structured into seven folders, with six folders corresponding to text-to-video AI models and one folder for real videos. Each of these folders has 100 subfolders corresponding to human action classes. All videos in a given subfolder were generated using the same prompt (see the list of prompts <a href='https://huggingface.co/datasets/faridlab/deepaction_v1/blob/main/captions.csv'>here</a>).

Included below are example videos generated using the prompt "a person taking a selfie". Note that, since each text-to-video AI model generates videos with different ratios and resolutions, these videos were normalized 512x512.

<table class="video-table">
    <tr>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/Pexels.mp4" controls></video>
            <p style="text-align: center;">Real</p>
        </td>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/BDAnimateDiffLightning.mp4" controls ></video>
            <p style="text-align: center;">AnimateDiff</p>
        </td>
    </tr>
    <tr>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/CogVideoX5B.mp4" controls></video>
            <p style="text-align: center;">CogVideoX5B</p>
        </td>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/RunwayML.mp4" controls ></video>
            <p style="text-align: center;">RunwayML</p>
        </td>
    </tr>
  <tr>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/StableDiffusion.mp4" controls></video>
            <p style="text-align: center;">StableDiffusion</p>
        </td>
        <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/Veo.mp4" controls ></video> 
            <p style="text-align: center;">Veo (pre-release version)</p>
        </td>
    </tr>
    <tr>
      <td style="width: 50%;">
            <video src="https://data.matsworld.io/ucbresearch/deepaction/VideoPoet.mp4" controls></video>
            <p style="text-align: center;">VideoPoet</p>
        </td>
    </tr>
</table>


<br>

## Licensing

The AI-generated videos (BDAnimateDiffLightning, CogVideoX5B, RunwayML, StableDiffusion, Veo, and VideoPoet folders) are released under <a href='https://creativecommons.org/licenses/by/4.0/deed.en'>the CC BY 4.0 license</a>. The real videos (Pexels folder) are released under <a href='https://www.pexels.com/license/'>the Pexels license</a>.

<br>

## Misc

Please use the following citation when referring to this dataset:

```bib
@misc{bohacek2024human,
      title={Human Action CLIPS: Detecting AI-generated Human Motion}, 
      author={Matyas Bohacek and Hany Farid},
      year={2024},
      eprint={2412.00526},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.00526}, 
}
```

This work was done during the first author's (Matyas Bohacek) internship at Google.