12 April - 13 April
Task Completed: Model Training and Transfer Learning
This week was focused on training the selected CNN model using the collected and preprocessed datasets. The process involved:
Data Splitting:
- The dataset was divided into:
- Training: 70%
- Validation: 15%
- Testing: 15%
Transfer Learning:
- Used pre-trained ResNet50 weights.
- Fine-tuned the final layers for our specific deepfake classification task.
Training Parameters:
- Learning Rate: 0.0001
- Optimizer: Adam
- Loss Function: Binary Cross-Entropy
- Epochs: 20 (initial testing)
- Batch Size: 32
The model was trained using TensorFlow/Keras, and training progress was monitored through loss and accuracy metrics.
Early stopping and model checkpointing techniques were applied to avoid overfitting and preserve the best-performing model.
The training results showed encouraging performance, with accuracy reaching above 90% on the validation dataset after tuning.