How davidgh96/20-21_dagahe works
This repository implements a facial presentation-attack classifier, with two ways to train and run it: separate local scripts or a self-contained Google Colab script. The workflow is image datasets → VGG19-based model training → saved model and weights → per-image prediction as attack or real. The README also documents using a previously trained model instead of training. The Colab script combines model creation, training, and prediction; the separate scripts divide those responsibilities. Arrows reflect README statements and the shown source; no separate dataset or model files are present in the tree.
Main entry points: fitModel.py Local trainer · predictModel.py Image predictor · ModuloCompletoGoogleColab.py Combined Colab script · createModelVGG19.py VGG19 model
Components 9
Model training
- VGG19 model
createModelVGG19.py - Local trainer
fitModel.py
Image inference
- Image predictor
predictModel.py
Colab workflow
- Combined Colab script
ModuloCompletoGoogleColab.py
Other components
- User
- Labeled image sets
- Model and weights
- Input image
- Attack or real
Connections 12
- User → Local trainer: runs training
README.md - Labeled image sets → Local trainer: provides images
fitModel.py - Local trainer → VGG19 model: creates model via
fitModel.py - Local trainer → Model and weights: saves model
fitModel.py - User → Image predictor: runs prediction
README.md - Model and weights → Image predictor: loads model
predictModel.py - Input image → Image predictor: supplies image
predictModel.py - Image predictor → Attack or real: returns class
predictModel.py - User → Combined Colab script: runs workflow
README.md - Labeled image sets → Combined Colab script: provides images
ModuloCompletoGoogleColab.py - Input image → Combined Colab script: supplies image
ModuloCompletoGoogleColab.py - Combined Colab script → Attack or real: prints class
ModuloCompletoGoogleColab.py