data

YOLOV2Config

extendsModelConfig

YOLOV2Config(num_classes: int = 80, in_channels: int = 3, num_anchors: int = 5, anchors: tuple[tuple[float, float], ...] = _COCO_ANCHORS, score_thresh: float = 0.5, nms_thresh: float = 0.5, lambda_coord: float = 5.0, lambda_noobj: float = 0.5, tiny: bool = False)

source edit

Configuration for YOLOv2 / YOLO9000 (Redmon & Farhadi, CVPR 2017).

Parameters

num_classesint= 80

Number of object classes C (COCO default: 80).

in_channelsint= 3

Input image channels (default 3).

num_anchorsint= 5

Number of anchor boxes per cell A (default 5).

anchorstuple[tuple[float, float], ...]= _COCO_ANCHORS

Anchor (width, height) pairs relative to a 32-pixel cell. Must have exactly num_anchors entries.

score_threshfloat= 0.5

Minimum objectness × class-prob score for inference.

nms_threshfloat= 0.5

IoU threshold for NMS at inference.

lambda_coordfloat= 5.0

Up-weighting for box regression loss.

lambda_noobjfloat= 0.5

Down-weighting for no-object confidence loss.

tinybool= False

Use the tiny Darknet backbone.

Used by 1

lucid.models.vision.yolo

Constructors

dunder

init

→None

__init__(num_classes: int = 80, in_channels: int = 3, num_anchors: int = 5, anchors: tuple[tuple[float, float], ...] = _COCO_ANCHORS, score_thresh: float = 0.5, nms_thresh: float = 0.5, lambda_coord: float = 5.0, lambda_noobj: float = 0.5, tiny: bool = False)

source edit

data

YOLOV2Config

extendsModelConfig

YOLOV2Config(num_classes: int = 80, in_channels: int = 3, num_anchors: int = 5, anchors: tuple[tuple[float, float], ...] = _COCO_ANCHORS, score_thresh: float = 0.5, nms_thresh: float = 0.5, lambda_coord: float = 5.0, lambda_noobj: float = 0.5, tiny: bool = False)

source edit

Configuration for YOLOv2 / YOLO9000 (Redmon & Farhadi, CVPR 2017).

Parameters

num_classesint= 80

Number of object classes C (COCO default: 80).

in_channelsint= 3

Input image channels (default 3).

num_anchorsint= 5

Number of anchor boxes per cell A (default 5).

anchorstuple[tuple[float, float], ...]= _COCO_ANCHORS

Anchor (width, height) pairs relative to a 32-pixel cell. Must have exactly num_anchors entries.

score_threshfloat= 0.5

Minimum objectness × class-prob score for inference.

nms_threshfloat= 0.5

IoU threshold for NMS at inference.

lambda_coordfloat= 5.0

Up-weighting for box regression loss.

lambda_noobjfloat= 0.5

Down-weighting for no-object confidence loss.

tinybool= False

Use the tiny Darknet backbone.

Used by 1

lucid.models.vision.yolo

Constructors

dunder

init

→None

__init__(num_classes: int = 80, in_channels: int = 3, num_anchors: int = 5, anchors: tuple[tuple[float, float], ...] = _COCO_ANCHORS, score_thresh: float = 0.5, nms_thresh: float = 0.5, lambda_coord: float = 5.0, lambda_noobj: float = 0.5, tiny: bool = False)

source edit