data
DenseNetConfig
extends
ModelConfigDenseNetConfig(num_classes: int = 1000, in_channels: int = 3, growth_rate: int = 32, block_config: tuple[int, ...] = (6, 12, 24, 16), num_init_features: int = 64, bn_size: int = 4, dropout_rate: float = 0.0, memory_efficient: bool = False)Configuration for all DenseNet variants (121/169/201/264).
growth_rate (k) — number of feature maps each dense layer contributes.
block_config — number of dense layers per block (4 blocks total).
num_init_features — channels after the initial conv stem.
bn_size — bottleneck expansion factor (each layer uses bn_size*k filters
in its 1×1 branch before the 3×3 branch).
dropout_rate — dropout after each dense layer (0 = disabled).
memory_efficient — toggles checkpointing in dense blocks (unused here,
kept for API parity with other frameworks).