# AI-generated image detection
**(Work In Progress)**

- [ ] Refactor code
- [ ] Review dependencies
- [ ] Containerize (Docker)
- [ ] Update documentation

## AI-Generated Image detection

This part handles the detection of AI-generated images.
The current code contains two classifiers to detect AI-generated images from two types of architectures:
- GANs

## Model weights

### 1. CNN Detection

Run the `download_weights_CNN.sh` script:

```commandline
bash download_weights_CNN.sh
```

Note: you need `wget` installed on your system (it is by default for most Linux systems).

### 2. Diffusion

**TODO**


## Run the models

Make sure you have the weights available before doing so.

**TODO: environments**

### 1. CNN Detection

```commandline
python CNN_model_classifier.py
```
Available options:

- `-f / --file` (default=`'examples_realfakedir'`)
- `-m / --model_path` (default=`'weights/blur_jpg_prob0.5.pth'`)
- `-c / --crop` (default=`None`):  Specify crop size (int) by default, do not crop.
- `--use_cpu`: use cpu (by default uses GPU) -> **TODO: remove (obsolete)**

Example usage:

```commandline
python CNN_model_classifier.py -f examples/real.png -m weights/blur_jpg_prob0.5.pth
```

### 2. Diffusion detection

**TODO**

## References

Based on:
- https://github.com/hoangthuc701/GenAI-image-detection
- https://github.com/ptmaimai106/DetectGenerateImageByRealImageOnly