data

FasterRCNNConfig

extendsModelConfig

FasterRCNNConfig(num_classes: int = 91, in_channels: int = 3, backbone_layers: tuple[int, int, int, int] = (3, 4, 6, 3), backbone_bn_eps: float = 0.0, fpn_out_channels: int = 256, rpn_anchor_sizes: tuple[int, ...] = (32, 64, 128, 256, 512), rpn_anchor_ratios: tuple[float, ...] = (0.5, 1.0, 2.0), rpn_pre_nms_top_n: int = 1000, rpn_post_nms_top_n: int = 1000, rpn_nms_thresh: float = 0.7, rpn_min_size: float = 0.001, rpn_score_thresh: float = 0.0, rpn_fg_iou_thresh: float = 0.7, rpn_bg_iou_thresh: float = 0.3, roi_size: int = 7, roi_sampling_ratio: int = 2, roi_representation_size: int = 1024, roi_fg_iou_thresh: float = 0.5, roi_bg_iou_thresh: float = 0.5, bbox_reg_weights: tuple[float, float, float, float] = (10.0, 10.0, 5.0, 5.0), canonical_scale: int = 224, canonical_level: int = 4, score_thresh: float = 0.05, nms_thresh: float = 0.5, max_detections: int = 100)

source edit

Configuration for Faster R-CNN.

Faster R-CNN (Ren et al., NeurIPS 2015) replaces external proposal methods (selective search) with a learned Region Proposal Network (RPN) that shares convolutional features with the detection head. This makes the entire pipeline end-to-end trainable and near real-time.

Architecture overview (ResNet-50-FPN variant): Image → ResNet-50 backbone (frozen BN, C2-C5) → FPN (P2-P5 + pool) FPN levels → RPN head → region proposals (per-level top-k + NMS) FPN levels + proposals → MultiScale RoI Align (7×7) → TwoMLPHead (fc6/fc7) → FastRCNNPredictor (cls + per-class bbox)

Parameters

num_classesint= 91

Object classes incl. background (91 for COCO).

in_channelsint= 3

Input image channels.

backbone_layerstuple[int, int, int, int]= (3, 4, 6, 3)

ResNet bottleneck counts (default 3,4,6,3 = R50).

backbone_bn_epsfloat= 0.0

FrozenBatchNorm2d epsilon (reference uses 0).

fpn_out_channelsint= 256

Unified FPN channel count (256).

rpn_anchor_sizestuple[int, ...]= (32, 64, 128, 256, 512)

One anchor scale per FPN level (5 levels).

rpn_anchor_ratiostuple[float, ...]= (0.5, 1.0, 2.0)

Aspect ratios for anchor generation.

rpn_pre_nms_top_nint= 1000

Proposals kept per level before NMS (test 1000).

rpn_post_nms_top_nint= 1000

Proposals kept per image after NMS (test 1000).

rpn_nms_threshfloat= 0.7

RPN NMS IoU threshold.

rpn_min_sizefloat= 0.001

Minimum proposal side length (pixels).

rpn_score_threshfloat= 0.0

Minimum RPN objectness score.

rpn_fg_iou_threshfloat= 0.7

Anchor→GT IoU for foreground assignment.

rpn_bg_iou_threshfloat= 0.3

Anchor→GT IoU upper bound for background.

roi_sizeint= 7

RoI Align output spatial size (7 → 7×7).

roi_sampling_ratioint= 2

RoI Align sub-bin sampling ratio (2).

roi_representation_sizeint= 1024

TwoMLPHead hidden width (1024).

roi_fg_iou_threshfloat= 0.5

Proposal→GT IoU for fg assignment.

roi_bg_iou_threshfloat= 0.5

Proposal→GT IoU upper bound for bg.

bbox_reg_weightstuple[float, float, float, float]= (10.0, 10.0, 5.0, 5.0)

Per-component bbox delta scale (10,10,5,5).

canonical_scaleint= 224

FPN level-assignment canonical scale (224).

canonical_levelint= 4

FPN level-assignment canonical level (4).

score_threshfloat= 0.05

Minimum final class score (0.05).

nms_threshfloat= 0.5

Per-class NMS IoU threshold (0.5).

max_detectionsint= 100

Maximum detections returned per image (100).

Used by 3

Constructors

dunder

init

→None

__init__(num_classes: int = 91, in_channels: int = 3, backbone_layers: tuple[int, int, int, int] = (3, 4, 6, 3), backbone_bn_eps: float = 0.0, fpn_out_channels: int = 256, rpn_anchor_sizes: tuple[int, ...] = (32, 64, 128, 256, 512), rpn_anchor_ratios: tuple[float, ...] = (0.5, 1.0, 2.0), rpn_pre_nms_top_n: int = 1000, rpn_post_nms_top_n: int = 1000, rpn_nms_thresh: float = 0.7, rpn_min_size: float = 0.001, rpn_score_thresh: float = 0.0, rpn_fg_iou_thresh: float = 0.7, rpn_bg_iou_thresh: float = 0.3, roi_size: int = 7, roi_sampling_ratio: int = 2, roi_representation_size: int = 1024, roi_fg_iou_thresh: float = 0.5, roi_bg_iou_thresh: float = 0.5, bbox_reg_weights: tuple[float, float, float, float] = (10.0, 10.0, 5.0, 5.0), canonical_scale: int = 224, canonical_level: int = 4, score_thresh: float = 0.05, nms_thresh: float = 0.5, max_detections: int = 100)

source edit

FasterRCNNConfig(num_classes: int = 91, in_channels: int = 3, backbone_layers: tuple[int, int, int, int] = (3, 4, 6, 3), backbone_bn_eps: float = 0.0, fpn_out_channels: int = 256, rpn_anchor_sizes: tuple[int, ...] = (32, 64, 128, 256, 512), rpn_anchor_ratios: tuple[float, ...] = (0.5, 1.0, 2.0), rpn_pre_nms_top_n: int = 1000, rpn_post_nms_top_n: int = 1000, rpn_nms_thresh: float = 0.7, rpn_min_size: float = 0.001, rpn_score_thresh: float = 0.0, rpn_fg_iou_thresh: float = 0.7, rpn_bg_iou_thresh: float = 0.3, roi_size: int = 7, roi_sampling_ratio: int = 2, roi_representation_size: int = 1024, roi_fg_iou_thresh: float = 0.5, roi_bg_iou_thresh: float = 0.5, bbox_reg_weights: tuple[float, float, float, float] = (10.0, 10.0, 5.0, 5.0), canonical_scale: int = 224, canonical_level: int = 4, score_thresh: float = 0.05, nms_thresh: float = 0.5, max_detections: int = 100)

__init__(num_classes: int = 91, in_channels: int = 3, backbone_layers: tuple[int, int, int, int] = (3, 4, 6, 3), backbone_bn_eps: float = 0.0, fpn_out_channels: int = 256, rpn_anchor_sizes: tuple[int, ...] = (32, 64, 128, 256, 512), rpn_anchor_ratios: tuple[float, ...] = (0.5, 1.0, 2.0), rpn_pre_nms_top_n: int = 1000, rpn_post_nms_top_n: int = 1000, rpn_nms_thresh: float = 0.7, rpn_min_size: float = 0.001, rpn_score_thresh: float = 0.0, rpn_fg_iou_thresh: float = 0.7, rpn_bg_iou_thresh: float = 0.3, roi_size: int = 7, roi_sampling_ratio: int = 2, roi_representation_size: int = 1024, roi_fg_iou_thresh: float = 0.5, roi_bg_iou_thresh: float = 0.5, bbox_reg_weights: tuple[float, float, float, float] = (10.0, 10.0, 5.0, 5.0), canonical_scale: int = 224, canonical_level: int = 4, score_thresh: float = 0.05, nms_thresh: float = 0.5, max_detections: int = 100)