Implemented data augmentation and optimized the training loop

[twinkle485]

Description

This pull request implements data augmentation process in dataset.py and optimizes the training loop in train.py to improve model performance and training stability.

Changes Include:

• Data Augmentation:

  • Added various transformations such as random resizing, horizontal flipping, rotation, color jitter, Gaussian blur, and random erasing to enrich the dataset and enhance model generalization.

• Training Loop Optimization:

  • Implemented gradient clipping, learning rate scheduling (Cosine Annealing), and an early stopping mechanism to enhance training efficiency and prevent overfitting on smaller datasets.

Related Issue

• Fixes Data Augmentation #4 Cyclic Learning Rate Scheduler #6 Early Stopping #7

Additional Information

• Name: Twinkle Pal
• Email: [email protected]
• Roll No: 2305744
• Branch: CSE