### Install PyTorch and torchvision

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Installs the required deep learning libraries using pip. Run after cloning the repository.

```bash
pip install torch torchvision
```

--------------------------------

### Clone and enter the UniversalFakeDetect repository

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Clones the GitHub repository and changes into the project directory. Run this first to set up the project locally.

```bash
git clone https://github.com/Yuheng-Li/UniversalFakeDetect
cd UniversalFakeDetect
```

--------------------------------

### Dataset directory structure after unzipping

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Shows the expected directory layout under 'datasets/test' after downloading and unzipping the dataset. Each model directory (e.g., progan) contains '0_real' and '1_fake' subfolders.

```text
datasets
└── test					
      ├── progan	
      │── cyclegan   
      │── biggan
      │      .
      │      .
	  
```

--------------------------------

### Evaluate on all datasets

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Runs validation on all test domains. Requires a checkpoint file and an output folder. The results are saved in the results folder.

```bash
python validate.py  --arch=CLIP:ViT-L/14   --ckpt=pretrained_weights/fc_weights.pth   --result_folder=clip_vitl14
```

--------------------------------

### Dataset directory structure

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Shows the expected layout of the training dataset under the datasets/train directory. Each category folder contains real and fake images in 0_real and 1_fake subfolders.

```text
datasets
└── train			
      └── progan			
           ├── airplane
           │── bird
           │── boat
           │      .
           │      .
```

--------------------------------

### Train the model

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Trains the model on the Wang2020 dataset. The --fix_backbone argument is required to train only the linear layer parameters.

```bash
python train.py --name=clip_vitl14 --wang2020_data_path=datasets/ --data_mode=wang2020  --arch=CLIP:ViT-L/14  --fix_backbone
```

--------------------------------

### Evaluate on a specific generative model

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Evaluates on a single generative model by specifying the real and fake dataset paths. The results are saved in the results folder.

```bash
python validate.py  --arch=CLIP:ViT-L/14   --ckpt=pretrained_weights/fc_weights.pth   --result_folder=clip_vitl14  --real_path datasets/test/progan/0_real --fake_path datasets/test/progan/1_fake
```

--------------------------------

### BibTeX citation for UniversalFakeDetect paper

Source: https://github.com/yuheng-li/universalfakedetect/blob/main/README.md

Use this BibTeX entry to cite the UniversalFakeDetect paper in academic publications. It includes the paper title, authors, and venue (CVPR 2023).

```bibtex
@inproceedings{ojha2023fakedetect,
      title={Towards Universal Fake Image Detectors that Generalize Across Generative Models}, 
      author={Ojha, Utkarsh and Li, Yuheng and Lee, Yong Jae},
      booktitle={CVPR},
      year={2023},
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.