csp_darknet_53¶

lucid.models.csp_darknet_53(num_classes: int = 1000, split_ratio: float = 0.5, stem_channels: int = 32, **kwargs) → CSPNet¶

The csp_darknet_53 function builds a variant of CSPNet based on DarkNet-53 architecture as used in YOLO-v4. It replaces each standard residual stage with a CSPDarkBlock, utilizing DarkNet bottlenecks and cross-stage partial connections to significantly reduce memory cost and duplicate gradient paths.

The model is constructed with 5 CSP stages, closely aligned with the original DarkNet-53 design, while maintaining CSP semantics throughout.

Total Parameters: 27,278,536

Function Signature¶

@register_model
def csp_darknet_53(
    num_classes: int = 1000,
    split_ratio: float = 0.5,
    stem_channels: int = 32,
    **kwargs
) -> CSPNet

Parameters¶

num_classes (int, optional): Number of output classes for classification. Default is 1000.
split_ratio (float, optional): Ratio of channels going through the residual transform branch in each stage. Default is 0.5.
stem_channels (int, optional): Channels used in the stem convolution layer. Default is 32.
kwargs (dict, optional): Additional customization arguments passed to CSPNet.

Returns¶

CSPNet: A CSPNet-based model using DarkNet-53 backbone structure with CSP wrapping.

Examples¶

from lucid.models import csp_darknet_53

model = csp_darknet_53(num_classes=1000)
x = lucid.random.randn(1, 3, 224, 224)
y = model(x)
print(y.shape)  # torch.Size([1, 1000])