victorchika/igp-group-15-anti-spoofing-app

Face Liveness Detection Project

This project follows the pipeline:

• —video input
• —extract 10 frames per video
• —detect and crop faces automatically
• —keep frames grouped per video
• —train a CNN + LSTM model
• —classify as real or fake

Folder structure

liveness_detection_project/
├── data/
│   ├── raw_videos/
│   │   ├── Real/                               ← real face videos
│   │   │   └── Selfies/                        ← images only, skipped automatically
│   │   ├── 3D_paper_mask_/                     ← fake
│   │   ├── Cutout_attacks/                     ← fake
│   │   ├── Latex_mask/                         ← fake
│   │   ├── Printouts/                          ← fake
│   │   │   ├── WITH CUTOUT/
│   │   │   └── WITHOUT CUTOUT/
│   │   ├── Replay_display_attacks/             ← fake
│   │   ├── Silicone_mask/                      ← fake
│   │   ├── Textile 3D Face Mask Attack Sample/ ← fake
│   │   └── Wrapped_3D_paper_mask/              ← fake
│   ├── extracted_frames/
│   └── face_sequences/
├── models/
├── src/
├── requirements.txt
└── README.md

How folders are classified

The script extract_frames.py uses an explicit mapping to assign each folder to real or fake. If you add a new attack folder to raw_videos/, you must also add it to the FOLDER_LABELS dictionary at the top of src/extract_frames.py, otherwise it will be skipped with a warning.

The Real/Selfies/ subfolder contains .jpg images, not videos. It is skipped automatically.

Before you start

Make sure Python is installed on your laptop.

Step-by-step setup on Windows PowerShell

1. Open PowerShell in this project folder

Use cd to move into the folder that contains this project.

2. Create a virtual environment

python -m venv venv

3. Activate it

.\venv312\Scripts\Activate

If activation is blocked, run this once in PowerShell:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass

Then activate again.

4. Install dependencies

pip install -r requirements.txt

Run the project step by step

Step 1. Extract exactly 10 frames from each video

python src\extract_frames.py

Output goes to:

data/extracted_frames/
├── real/
└── fake/

At the end it prints a summary like:

Done. Real: 95 | Fake: 112 | Skipped: 3

Step 2. Detect and crop faces using OpenCV

python src\crop_faces_opencv.py

Output goes to:

data/face_sequences/
├── real/
└── fake/

Important:

• —only video folders with at least 7 usable face crops are kept
• —folders with fewer valid faces are skipped automatically
• —each kept folder is padded or trimmed to exactly 10 frames

Step 3. Check a few output folders manually

Open a few folders in:

data/face_sequences/real/
data/face_sequences/fake/

Check that:

• —the face is visible
• —the crop is not broken
• —each kept folder has exactly 10 images
• —each folder still represents one single video

Step 4. Train the model

python src\train.py

Best model is saved to:

models/best_model.pth

Split indices are also saved to:

models/split_indices.json

Step 5. Evaluate the model

python src\evaluate.py

This prints:

• —classification report
• —confusion matrix
• —APCER
• —BPCER
• —ACER

Step 6. Webcam demo

python src\webcam_demo.py

Step 7. Launch Streamlit

streamlit run app.py

Press q to quit.

One-command run

If your dataset is already in place and dependencies are installed, you can run:

python run_pipeline.py

That will run:

1. 1.frame extraction
2. 2.face cropping
3. 3.training
4. 4.evaluation

To redo frame extraction and face cropping

If you want to clear the processed data and start fresh:

Remove-Item -Recurse -Force "data\extracted_frames\*"
Remove-Item -Recurse -Force "data\face_sequences\*"

Then rerun Steps 1 and 2.

Common issues

python not found

Try:

py --version

If py works, use py instead of python.

ModuleNotFoundError

Make sure the virtual environment is activated and dependencies were installed.

A folder is being skipped with a WARNING

If you see WARNING: Unknown folder '...' — not in FOLDER_LABELS, open src/extract_frames.py and add the folder name and its label to the FOLDER_LABELS dictionary at the top of the file.

Not enough valid face folders

That usually means face detection failed on many extracted frames. Check your raw videos and cropped outputs.

Training starts but does not save a model

Make sure models/ exists. This starter project already includes that folder.

evaluate.py raises FileNotFoundError for split_indices.json

You must run train.py before evaluate.py. Training saves the split indices that evaluation depends on.

Recommended workflow

1. 1.Create and activate venv
2. 2.Install requirements
3. 3.Confirm your dataset folders are inside data/raw_videos/ as shown in the folder structure above
4. 4.Run python src\extract_frames.py
5. 5.Run python src\crop_faces_opencv.py
6. 6.Inspect the output folders manually
7. 7.Run python src\train.py
8. 8.Run python src\evaluate.py
9. 9.Try webcam demo only after the model is saved

AFTER DOWNLOADING THE ZIP

1. Activate it

.\venv312\Scripts\Activate

If activation is blocked, run this once in PowerShell/terminal:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass

Then activate again.

2. Run webcam_demo.py

 python src\webcam_demo.py