YOLO_V3Config¶

class lucid.models.YOLO_V3Config(num_classes: int, anchors: list[tuple[int, int]] = <factory>, image_size: int = 416, darknet: lucid.nn.module.Module | None = None, darknet_out_channels_arr: list[int] | None = None)¶

YOLO_V3Config stores the class count, 9-anchor set, input image size, and optional custom 3-scale backbone used by lucid.models.YOLO_V3.

Class Signature¶

@dataclass
class YOLO_V3Config:
    num_classes: int
    anchors: list[tuple[int, int]] = field(default_factory=...)
    image_size: int = 416
    darknet: nn.Module | None = None
    darknet_out_channels_arr: list[int] | None = None

Parameters¶

num_classes (int): Number of object classes predicted at every scale.
anchors: Flat list of 9 anchor (width, height) pairs used by the three detection heads.
image_size (int): Input image size used for box decoding and loss scaling.
darknet (nn.Module | None): Optional custom backbone that returns three feature maps.
darknet_out_channels_arr (list[int] | None): Channel widths of the three backbone outputs when a custom backbone is supplied.

Validation¶

num_classes and image_size must be greater than 0.
anchors must contain exactly 9 positive integer (width, height) pairs.
darknet must be an nn.Module or None.
darknet_out_channels_arr must be None for the default backbone.
When darknet is custom, darknet_out_channels_arr must contain exactly three positive integers.

Usage¶

import lucid.models as models
import lucid.nn as nn

class ToyBackbone(nn.Module):
    def forward(self, x):
        return x, x, x

config = models.YOLO_V3Config(
    num_classes=3,
    darknet=ToyBackbone(),
    darknet_out_channels_arr=[3, 3, 3],
)
model = models.YOLO_V3(config)