Skip to content

02-source-and-setup.md

Latest commit

 

History

History
171 lines (146 loc) · 4.76 KB

02-source-and-setup.md

File metadata and controls

171 lines (146 loc) · 4.76 KB

XceptionNet — Source & Setup

Source location(s)

What I verified

  • Does the repo match the paper description? (Yes/No/Partially) Partially
  • Pretrained model available? (Yes/No) Yes
  • Dataset expected by repo: face images extracted from video dataset

Environment / dependencies

List exactly what you installed (with versions if possible):

  • OS: Ubuntu 22.04.5 LTS
  • Python: 3.10.12
  • PyTorch: 2.7.1+cu118
  • CUDA (if used): 11.8
  • OpenCV: 4.11.0
  • Other deps: opencv-python dlib==19.18.0 numpy==1.22.0 pillow>=6.2.2

Setup steps (copy/paste friendly)

1) Clone / install

python3.10 -m venv xception_env 
source xception_env/bin/activate 
git clone https://github.com/HongguLiu/Deepfake-Detection
cd Deepfake-Detection
python -m pip install -r requirements.txt

2) Any downloads needed (weights/datasets)

3) How to run

# To train
python train_CNN.py   -n xception   -tl ./data_list/train.txt   -vl ./data_list/val.txt   -bz 64   -e 20   -mn xception_ffpp_c23.pkl   --continue_train True   -mp ./ffpp_c23.pth

# To test
python test_CNN.py -bz 64 -tl data_list/test.txt -mp output/xception/best.pkl 

# To predict a video
python detect_from_video.py --video_path ./videos/003_000.mp4 --model_path ./pretrained_model/df_c0_best.pkl -o ./output --cuda

# To predict an image
python test_CNN.py -bz 32 --test_list ./data_list/Deepfakes_c0_299.txt --model_path ./pretrained_model/df_c0_best.pkl

Output / results

  • What output you saw (logs, metrics, saved files):

Output of training

(xception_env) gdgteam1@lisa:~/AI-Video-Detection/backend/models/XceptionNet/Deepfake-Detection$ python train_CNN.py -n xception -tl ./data_list/train.txt -vl ./data_list/val.txt -bz 64 -e 20 -mn xception_ffpp_c23.pkl --continue_train True -mp ./ffpp_c23.pth Using dropout 0.5 Epoch 1/20

epoch train loss: 0.0065 Acc: 0.8341 epoch val loss: 0.0041 Acc: 0.8885 Epoch 2/20

epoch train loss: 0.0029 Acc: 0.9249 epoch val loss: 0.0037 Acc: 0.9033 Epoch 3/20

epoch train loss: 0.0020 Acc: 0.9521 epoch val loss: 0.0032 Acc: 0.9274 Epoch 4/20

epoch train loss: 0.0013 Acc: 0.9672 epoch val loss: 0.0050 Acc: 0.8987 Epoch 5/20

epoch train loss: 0.0010 Acc: 0.9752 epoch val loss: 0.0033 Acc: 0.9315 Epoch 6/20

epoch train loss: 0.0005 Acc: 0.9886 epoch val loss: 0.0033 Acc: 0.9445 Epoch 7/20

epoch train loss: 0.0003 Acc: 0.9920 epoch val loss: 0.0056 Acc: 0.9289 Epoch 8/20

epoch train loss: 0.0004 Acc: 0.9935 epoch val loss: 0.0044 Acc: 0.9412 Epoch 9/20

epoch train loss: 0.0002 Acc: 0.9947 epoch val loss: 0.0041 Acc: 0.9460 Epoch 10/20

epoch train loss: 0.0003 Acc: 0.9948 epoch val loss: 0.0054 Acc: 0.9354 Epoch 11/20

epoch train loss: 0.0001 Acc: 0.9980 epoch val loss: 0.0048 Acc: 0.9462 Epoch 12/20

epoch train loss: 0.0001 Acc: 0.9983 epoch val loss: 0.0057 Acc: 0.9447 Epoch 13/20

epoch train loss: 0.0001 Acc: 0.9984 epoch val loss: 0.0054 Acc: 0.9473 Epoch 14/20

epoch train loss: 0.0001 Acc: 0.9987 epoch val loss: 0.0059 Acc: 0.9478 Epoch 15/20

epoch train loss: 0.0002 Acc: 0.9987 epoch val loss: 0.0053 Acc: 0.9483 Epoch 16/20

epoch train loss: 0.0002 Acc: 0.9993 epoch val loss: 0.0055 Acc: 0.9507 Epoch 17/20

epoch train loss: 0.0000 Acc: 0.9996 epoch val loss: 0.0054 Acc: 0.9506 Epoch 18/20

epoch train loss: 0.0001 Acc: 0.9995 epoch val loss: 0.0053 Acc: 0.9485 Epoch 19/20

epoch train loss: 0.0000 Acc: 0.9997 epoch val loss: 0.0068 Acc: 0.9479 Epoch 20/20

epoch train loss: 0.0000 Acc: 0.9996 epoch val loss: 0.0061 Acc: 0.9492 Best val Acc: 0.9507

Output testing

(xception_env) gdgteam1@lisa:~/AI-Video-Detection/backend/models/XceptionNet/Deepfake-Detection$ python test_CNN.py -bz 64 -tl data_list/test.txt -mp output/xception/best.pkl Using dropout 0.5 Test Acc: 0.9505

  • Add screenshots in assets/ if helpful.

Issues encountered + fixes

  • Issue: requirements.txt failed to install
  • Cause: Several package versions are outdated and not compatible with python 3.10
  • Fix (steps/commands): Removed outdated packages, installed newer versions of these packages

Notes for teammates

  • If someone else sets this up from scratch, what do they need to know? If you have any question, dm on Discord.