How krishnaprasanna21/deepfake_video_detection works
Project type and purpose
• This is a deepfake‐detection model creation pipeline implemented as a set of Jupyter notebooks.
• Purpose: preprocess the Celeb-DF video dataset, train a convolutional neural network (DeepCNN), save the trained model (.h5), and run inference on new video data.
File structure
• Top-level files:
• Readme.md – project overview and instructions
• *.ipynb notebooks – data preprocessing, training, model exploration, and prediction
• deepfake-detection-model.h5, deepfake-detection-model1.h5 – saved trained model artifacts
• No nested directories; notebooks encapsulate each pipeline stage.
• External dependencies implied by the README: Google Colab (execution environment) and Google Drive (storage).
README for additional insights
• Libraries: dlib, OpenCV (cv2), PIL, TensorFlow, NumPy, Pandas, scikit-learn, etc.
• Dataset: Celeb-DF (video frames → face crops)
• Pipeline stages:
a. Data ingestion & frame extraction
b. Face cropping & dataset saving
c. Model definition (DeepCNN), train/test split
d. Training & evaluation
e. Model serialization (.h5)
f. Prediction on unseen data using saved .h5
a. Main components
• Data Source (Celeb-DF videos, Google Drive storage)
• Preprocessing Notebook (frame extraction, face cropping)
• Feature Store (folder of cropped images)
• Training Notebook (DeepCNN model, train/test split, evaluation)
• Model Artifacts (.h5 files)
• Prediction Notebook (load .h5, run inference)
• Execution Environment (Google Colab)
b. Relationships and interactions
• From Feature Store → Training → Model Artifacts
• From Model Artifacts + New Video Input → Prediction
• Google Drive ↔ Execution Environment (for loading/saving data and models)
c. Architectural patterns and design principles
• Linear ML pipeline (ETL → training → inference)
• Clear separation of concerns: data prep, training, inference
• Notebooks as orchestration layers rather than monolithic scripts
d. Key technologies and frameworks
• Python Jupyter Notebooks
• dlib & OpenCV for face detection/cropping
• TensorFlow (or PyTorch if referenced) for model definition and training
• Google Colab (compute) and Google Drive (storage)
• Supporting libraries: NumPy, Pandas, scikit-learn, matplotlib/seaborn for analysis and visualization