A comprehensive research and implementation of the novel SleepNet method, designed for both text and image classification tasks.
- Vision Transformer (ViT) Integration: Enhances feature extraction capabilities by leveraging the Vision Transformer (ViT) model, providing state-of-the-art performance in image classification.
- ResNet18 Architecture: Utilizes the proven ResNet18 architecture to achieve high accuracy in image classification tasks.
- Advanced Data Processing: Employs data augmentation and normalization techniques to improve generalization and robustness across diverse datasets.
- Efficient GPU Memory Management: Periodically clears GPU memory to optimize resource utilization, especially during extended training sessions.
- Adaptive Learning Rate Scheduling: Includes a dynamic learning rate scheduler to facilitate smoother and more effective model convergence.
- Python 3.8 or later
- PyTorch
- torchvision
- Hugging Face's
transformerslibrary - CUDA-compatible GPU (recommended for faster training)
-
Setup & Installation:
Install all necessary dependencies with:
pip install -r requirements.txt
-
Training the Text Classifier:
To train the SleepNet model with default parameters:
python trainer.py
For customized training settings, use:
python trainer.py --dataset 'ag_news' --epochs 50 --num_classes 10 -
Training the Vision Classifier:
To train the SleepNet model for image classification:
python train_vision.py --dataset 'cifar100' --epochs 50 --num_classes 10