Table 1 ViT Configuration parameters
From: Vi-GeN 1.0: GAN-Augmented Vision Transformer Pipeline for Diversified Pneumonia Classification
Parameter | Value |
|---|---|
Input Image Size | 256*256 |
Patch Size | 16*16 |
Number of Transformer Layers | 12 |
Number of Attention Heads | 8 |
Embedding Dimensions | 768 |
MLP size | 3072 |
Optimizer | AdaMW |
Learning Rate | 0.00005 |
Batch Size | 32 |
Loss Function | Categorical Cross-Entropy |
Epochs | 50–80 (with Early Stopping on validation loss) |
Early Stopping | Yes (based on validation loss improvement) |
Regularization | Weight Decay (AdamW default), Dropout in attention layer |
Attention Mechanism | Multi-Head Self-Attention (MHSA) |
Classifier Head | MLP Head with Softmax Activation |