DeCoF

KO | EN

gitlite — search

#react #ai #typescript #go #android #rust #linux #java #nodejs #windows #cli #kubernetes #awesome-list #openai

(★ 16)

Detecting AI-Generated Video via Frame Consistency

DeCoF Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Detecting AI-Generated Video via Frame Consistency
<div align="center">
    <a href='https://arxiv.org/abs/2402.02085'><img src='https://img.shields.io/badge/ArXiv-2402.02085-red'></a>
</div>

![Overview](https://raw.githubusercontent.com/LongMa-2025/DeCoF/main/pics/model-1_1.png)

## News 🚀

**[2024/6/7]**  The prompts used to generate videos, the attribute partitioning involved in prompts, and the partitioning of the dataset  have been open-source. You can access it in the `/datas/` folder. Unfortunately, we are unable to directly provide real videos. You can download them from the original dataset based on the video_id of prompts.

**[2024/6/16]**  All generated videos can be downloaded from [here](https://drive.google.com/drive/folders/1X4Gw4hkWfka42IaBQ6ImkDTGAeA9Wlk4?usp=drive_link), The emergence speed of video generation models far exceeds our imagination. If you expand a subset based on our dataset, we sincerely invite you to release the corresponding generated videos.

**[2024/7.1]** Thanks to [Kling](https://kling.kuaishou.com/). We have extended the DecoF dataset based on the video generation model **Kling**, and the test dataset for Kling will be made public soon.

**[2026/2/28]**  Regarding pretrained weights: due to **server migration**, the original pretrained weights are unfortunately no longer available. However, we have released **new pretrained models trained on our updated dataset**. Please refer to our [repository](https://github.com/LongMa-2025/AIGVDBench) for details.

## Abstract
> The increasing realism of AI-generated videos has raised potential security concerns, making it difficult for humans to distinguish them from the naked eye. Despite these concerns, limited research has been dedicated to detecting such videos effectively.
To this end, we propose an open-source AI-generated video detection dataset. Our dataset spans diverse objects, scenes, behaviors, and actions by organizing input prompts into independent dimensions. 
It also includes various generation models with different generative models, featuring popular commercial models such as OpenAI's Sora, Google's Veo, and Kwai's Kling.
Furthermore,  we propose a simple yet effective **De**tection model based on **Co**ncistency of **F**rame (DeCoF), which learns robust temporal artifacts across different generation methods.
Extensive experiments demonstrate the generality and efficacy of the proposed DeCoF in detecting AI-generated videos, including those from nowadays' mainstream commercial generators.

## Setup

1. Clone this repository 
```bash
git clone https://github.com/wuwuwuyue/DeCoF
cd DeCoF
```
2. Prepare the dataset
Download datasets and place them in `./data/` folder.  Extract frames from the first 32 frames of the video, evenly select 8 frames, crop the frames into squares at the center, and store them as follows. The files for dividing the training set, testing set, and validation set are stored in  `datas/split`.
For example, download **Text2Video_Zero** and **Real**,:
```
└── data
    └── Text2Video_Zero
        ├── Train 
        ├── Val
        ├── Test
        │   └── 0_real
        |       └── -_hbPLsZvvo_19_25
        |           └── 000.jpg
        |           └── 001.jpg
        │   └── 1_fake
        |       └── -_hbPLsZvvo_19_25
        |           └── 000.jpg
        |           └── 001.jpg    
```
3. Install the necessary libraries
```bash
pip install torch torchvision
pip install tqdm  einops  numpy scikit-learn pillow ftfy regex  opencv-python  scipy  opencv-python
```

## Train
Modify `subdatasets_name`  to the name of the sub dataset used in the `src/configs/base.json`.

Change the `root`  in the `src/utils/initialize.py` to your dataset path.
```bash
CUDA_VISIBLE_DEVICES=* python3 src/train.py \
src/configs/base.json \
-n DeCoF
    
```
## Test
Change the `load_file`  in the `src/test.py` to your checkpoint path.
```bash
CUDA_VISIBLE_DEVICES=* python3 src/test.py \
src/configs/base.json \
-n DeCoF

```

## Citation
If you find our work useful for your research, please consider citing our paper
```bibtex
@article{ma2024detecting,
  title={Detecting AI-Generated Video via Frame Consistency},
  author={Long Ma and Zhiyuan Yan and Qinglang Guo and Yong Liao and Haiyang Yu and Pengyuan Zhou},
  journal={arXiv preprint arXiv:2402.02085},
  year={2024}
}
```

DeCoF

DeCoF Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

DeCoF

DeCoF Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README