# CViT

**Repository Path**: a844304090/CViT

## Basic Information

- **Project Name**: CViT
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: GPL-3.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-06-30
- **Last Updated**: 2021-06-30

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# CViT
### Deepfake Video Detection Using Convolutional Vision Transformer

Implementation code for our paper. 
[link to paper](https://arxiv.org/abs/2102.11126) | [link to MS Thesis](http://etd.aau.edu.et/handle/123456789/24209)

### Requirements:
* Pytorch >=1.4

### DL library used for face extraction
   * helpers_read_video_1.py
   * helpers_face_extract_1.py
   * blazeface.py
   * blazeface.pth

### Preprocessing
extractfaces.py<br />
&nbsp;&nbsp;&nbsp;Face extraction from video. <br /> 
&nbsp;&nbsp;&nbsp;The code works for DFDC dataset. You can test it using the sample data provided. 

### Weights
deepfake_cvit_gpu_ep_50.pth - Full model weight. <br />
deepfake_cvit_gpu_inference_ep_50.pth - For detection. <br />

### Predict CViT 

python cvit_prediction.py <br />
&nbsp;&nbsp;&nbsp; Predicts whether a video is Deepfake or not.<br />
&nbsp;&nbsp;&nbsp; Prediction value <0.5 - REAL <br />
&nbsp;&nbsp;&nbsp; Prediction value >=5  - FAKE


### Train CViT
To train the model on your own you can use the following parameters:<br />
&nbsp;&nbsp;e: epoch <br/>
&nbsp;&nbsp;s: session - **(g)** - GPU or **(t)** - TPU <br/>
&nbsp;&nbsp;w: weight decay  default= 0.0000001 <br/>
&nbsp;&nbsp;l: learning rate default=0.001 <br/>
&nbsp;&nbsp;d: path file <br/>
&nbsp;&nbsp;b: batch size, defualt=32 <br/>

python cvit_train.py -e 10 -s 'g' -l 0.0001 -w 0.0000001 -d sample_train_data/

### Authors
**Deressa Wodajo** <br />
**Solomon Atnafu (PhD)**

### References
Deressa Wodajo and Solomon Atnafu, "[Deepfake Video Detection Using Convolutional Vision Transformer](https://arxiv.org/abs/2102.11126)," arXiv preprints arXiv:2102.11126, 2021.