Frequency Masking for Universal DeepFake Detection

Paper

We study universal deepfake detection. Our goal is to detect synthetic images from a range of generative AI approaches, particularly emerging ones which are unseen during training of the deepfake detector. Universal deepfake detection requires outstanding generalization capability. Motivated by recently proposed masked image modeling which has demonstrated excellent generalization in self-supervised pre-training, we make the first attempt to explore masked image modeling for universal deepfake detection. We study spatial and frequency domain masking in training deepfake detector. Based on empirical analysis, we propose a novel deepfake detector via frequency masking. Our focus on frequency domain is different from most spatial domain detection. Comparative analyses reveal substantial performance gains over existing methods.

PLEASE READ DOCUMENTATION BELOW

★ Datasets

Follow strictly the naming and structure of folders below:

Wang_CVPR2020

• training/validation: link
• testing: link

Wang_CVPR2020/
├── testing/
│   ├── crn/
│   │   ├── 1_fake/
│   │   └── 0_real/
│   └── ...
├── training/
│   ├── car/
│   │   ├── 1_fake/
│   │   └── 0_real/
│   └── ...
└── validation/
    ├── car/
    │   ├── 1_fake/
    │   └── 0_real/
    └── ...

Ojha_CVPR2023

• download: link

Ojha_CVPR2023/
├── guided/
│   ├── 1_fake/
│   └── 0_real/
├── ldm_200_cfg/
│   ├── 1_fake/
│   └── 0_real/
├── ldm_100/
│   ├── 1_fake/
│   └── 0_real/
└── ...

Make sure to change the path based on where you saved the data: link1 link2

★ Model Weights

You can download the model weights here. Put the files under the repository and don't change the name or anything!

File structure should strictly be like this:

FakeImageDetection/checkpoints/
├── mask_0/
│   ├── rn50ft.pth (Wang et al.)
│   ├── rn50_modft.pth (Gragnaniello et al.)
│   ├── clipft.pth (Ojha et al.)
│   └── ...
├── mask_15/
│   ├── rn50ft_midspectralmask.pth
│   ├── rn50ft_lowspectralmask.pth
│   ├── rn50ft_highspectralmask.pth
│   ├── rn50ft_pixelmask.pth
│   ├── rn50ft_patchmask.pth
│   ├── rn50ft_spectralmask.pth (Wang et al. + Ours)
│   ├── rn50_modft_spectralmask.pth (Gragnaniello et al. + Ours)
│   ├── clipft_spectralmask.pth (Ojha et al. + Ours)
│   └── ...
└── ...

★ Testing Script (test.py)

Description

The script test.py is designed for evaluating trained models on multiple datasets. The script leverages metrics such as Average Precision, Accuracy, and Area Under the Curve (AUC) for evaluation.

Basic Command

python -m torch.distributed.launch --nproc_per_node=GPU_NUM test.py -- [options]

Command-Line Options

--model_name : Type of the model. Choices include various ResNet and ViT variants.
--mask_type  : Type of mask generator for data augmentation. Choices include 'patch', 'spectral', etc.
--pretrained : Use pretrained model.
--ratio      : Masking ratio for data augmentation.
--band       : Frequency band to randomly mask.
--batch_size : Batch size for evaluation. Default is 64.
--data_type  : Type of dataset for evaluation. Choices are 'Wang_CVPR20' and 'Ojha_CVPR23'.
--device     : Device to use for evaluation (default: auto-detect).

Bash Command

Edit testing bash script test.sh:

#!/bin/bash

# Define the arguments for your test script
GPUs="$1"
NUM_GPU=$(echo $GPUs | awk -F, '{print NF}')
DATA_TYPE="Wang_CVPR20"  # Wang_CVPR20 or Ojha_CVPR23
MODEL_NAME="clip" # clip, RN50_mod or RN50
MASK_TYPE="nomask" # spectral, pixel, patch or nomask
BAND="all" # all, low, mid, high
RATIO=15
BATCH_SIZE=64

# Set the CUDA_VISIBLE_DEVICES environment variable to use GPUs
export CUDA_VISIBLE_DEVICES=$GPUs

echo "Using $NUM_GPU GPUs with IDs: $GPUs"

# Run the test command
python -m torch.distributed.launch --nproc_per_node=$NUM_GPU test.py \
  -- \
  --data_type $DATA_TYPE \
  --pretrained \
  --model_name $MODEL_NAME \
  --mask_type $MASK_TYPE \
  --band $BAND \
  --ratio $RATIO \
  --batch_size $BATCH_SIZE \
  --other_model

Now, use this to run testing:

bash test.sh "0" # gpu id/s to use

Results

You can find the results in this structure:

results/
├── mask_15/
│   ├── ojha_cvpr23/
│   │   ├── rn50ft_spectralmask.txt
│   │   └── ...
│   └── wang_cvpr20/
│       ├── rn50ft_spectralmask.txt
│       └── ...
└── ...

★ Training Script (train.py)

Description

This script (train.py) is designed for distributed training and evaluation of models. The script is highly configurable through command-line arguments and provides advanced features such as WandB integration, early stopping, and various masking options for data augmentation.

Basic Command

To run the script in a distributed environment:

python -m torch.distributed.launch --nproc_per_node=GPU_NUM train.py -- [options]

Command-Line Options

--local_rank     : Local rank for distributed training. 
--num_epochs     : Number of epochs for training. 
--model_name     : Type of the model. Choices include various ResNet and ViT variants.
--wandb_online   : Run WandB in online mode. Default is offline.
--project_name   : Name of the WandB project.
--wandb_run_id   : WandB run ID.
--resume_train   : Resume training from last or best epoch.
--pretrained     : Use pretrained model.
--early_stop     : Enable early stopping.
--mask_type      : Type of mask generator for data augmentation. Choices include 'patch', 'spectral', etc.
--batch_size     : Batch size for training. Default is 64.
--ratio          : Masking ratio for data augmentation.
--band           : Frequency band to randomly mask.

Bash Command

Edit training bash script train.sh:

#!/bin/bash

# Get the current date
current_date=$(date)

# Print the current date
echo "The current date is: $current_date"

# Define the arguments for your training script
GPUs="$1"
NUM_GPU=$(echo $GPUs | awk -F, '{print NF}')
NUM_EPOCHS=10000
PROJECT_NAME="Frequency-Masking"
MODEL_NAME="clip" # RN50_mod, RN50, clip
MASK_TYPE="spectral" # nomask, spectral, pixel, patch
BAND="all" # all, low, mid, high
RATIO=15
BATCH_SIZE=128
WANDB_ID="2w0btkas"
RESUME="from_last" # from_last or from_best

# Set the CUDA_VISIBLE_DEVICES environment variable to use GPUs
export CUDA_VISIBLE_DEVICES=$GPUs

echo "Using $NUM_GPU GPUs with IDs: $GPUs"

# Run the distributed training command
python -m torch.distributed.launch --nproc_per_node=$NUM_GPU train.py \
  -- \
  --num_epochs $NUM_EPOCHS \
  --project_name $PROJECT_NAME \
  --model_name $MODEL_NAME \
  --mask_type $MASK_TYPE \
  --band $BAND \
  --ratio $RATIO \
  --batch_size $BATCH_SIZE \
  --early_stop \
  --pretrained \
  # --resume_train $RESUME \
  # --debug \
  # --wandb_online \
  # --wandb_run_id $WANDB_ID \

Now, use this to run training:

bash train.sh "0,1,2,4" # gpu ids to use

Important:

• When starting the training (from 1st epoch), please comment out --resume_train $RESUME \ and --debug \. And if you don't want to use wandb logging, comment out --wandb_online \ and --wandb_run_id $WANDB_ID \. In short, just comment out the last three lines in the bash script.
• If you notice that the training process stalls during an epoch (e.g., epoch 20+ or 30+), please interrupt it by pressing ctrl+c. The bash script is configured to resume training from the last saved epoch (if you uncomment --resume_train $RESUME \).

★ License

This project is licensed under the Apache License.

★ Citation

If you use this code in your research, please consider citing it. Below is the BibTeX entry for citation:

@misc{doloriel2024frequency,
      title={Frequency Masking for Universal Deepfake Detection}, 
      author={Chandler Timm Doloriel and Ngai-Man Cheung},
      year={2024},
      eprint={2401.06506},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}