config_example.yaml

# Example configuration file for HierarchialViT

# Model Configuration
model:
  name: "hvt_xl"  # Model variant (base, large, xl)
  patch_size: 16  # Size of image patches
  embed_dim: 768  # Initial embedding dimension
  depth: [2, 2, 6, 2]  # Number of transformer blocks per stage
  num_heads: [3, 6, 12, 24]  # Number of attention heads per stage
  mlp_ratio: 4  # MLP expansion ratio
  drop_path: 0.1  # Stochastic depth rate

# Training Configuration
training:
  batch_size: 64
  epochs: 100
  optimizer:
    name: "adamw"
    lr: 1e-4
    weight_decay: 0.05
  scheduler:
    name: "cosine"
    warmup_epochs: 5

# Data Configuration
data:
  train_path: "path/to/train/data"
  val_path: "path/to/val/data"
  input_size: 224
  augmentation:
    random_crop: true
    random_flip: true
    color_jitter: 0.4
    auto_augment: true

# Logging Configuration
logging:
  log_dir: "logs/"
  save_freq: 10
  eval_freq: 5