detector_codes/README.md

Setup

conda create -n AIGIBench -y python=3.9
conda activate AIGIBench
# install pytorch 
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

There is a corresponding requirements.txt under each code, so you only need to use the following command after installing pytorch and other environments

pip install -r requirements.txt 

Train & Test

a. CNNDetection-master

You need to change [dataroot] and [model_path] in eval_config.py when testing.

train: nohup python train.py --name 5class-car-cat-chair-horse-sdv1.4 --blur_prob 0.1 --blur_sig 0.0,3.0 --jpg_prob 0.1 --jpg_method cv2,pil --jpg_qual 30,100 --dataroot [training datasets path]  --classes car,cat,chair,horse,sdv1.4 > nohup.out 2>&1 &
test:  python eval.py

b. AIDE-main

You need to change [RESUME_PATH] and [eval_datasets] in eval_config.py when testing.

train: nohup python main_finetune.py --model AIDE --batch_size 32 --blr 1e-4 --epochs 25 --data_path [training datasets path] --eval_data_path [evaling datasets path] > nohup.out 2>&1 &
test:  bash scripts/eval.sh

c. SAFE-main

You need to change [RESUME_PATH] and [eval_datasets] in scripts/eval.sh when testing.

train: nohup python main_finetune.py --model SAFE --data_path [training datasets path] --eval_data_path [evaling datasets path] --epochs 20 > nohup.out 2>&1 &
test:  bash scripts/eval.sh

d. Effort-AIGI-Detection

You need to change [CLIPModel_Path] in models/clip_models.py and [DetectionTests] in test.py when testing.

train: nohup python train.py --name 5class-car-cat-chair-horse-sdv1.4 --dataroot [training datasets path] --classes car,cat,chair,horse,sdv1.4 --use_svd > nohup.out 2>&1 &
test:  python test.py --model_path [training checkpoints path] --use_svd

e. Others

All other methods work the same way. You need to change [DetectionTests] in test.py when testing.

train: nohup python train.py --name 5class-car-cat-chair-horse-sdv1.4 --dataroot [training datasets path] --classes car,cat,chair,horse,sdv1.4 > nohup.out 2>&1 &
test:  python test.py --model_path [training checkpoints path]