def __init__(self, num_classes, **kwags):
super(CascadeRCNN, self).__init__(**kwags)
self.NUM_CLASSES = num_classes
# RPN configuration
# Anchor attributes
self.ANCHOR_SCALES = (32, 64, 128, 256, 512)
self.ANCHOR_RATIOS = (0.5, 1, 2)
self.ANCHOR_FEATURE_STRIDES = (4, 8, 16, 32, 64)
# Bounding box refinement mean and standard deviation
self.RPN_TARGET_MEANS = (0., 0., 0., 0.)
self.RPN_TARGET_STDS = (0.1, 0.1, 0.2, 0.2)
# RPN training configuration
self.PRN_BATCH_SIZE = 256
self.RPN_POS_FRAC = 0.5
self.RPN_POS_IOU_THR = 0.7
self.RPN_NEG_IOU_THR = 0.3
# ROIs kept configuration
self.PRN_PROPOSAL_COUNT = 2000
self.PRN_NMS_THRESHOLD = 0.7
# RCNN configuration
# Bounding box refinement mean and standard deviation
self.RCNN_TARGET_MEANS = (0., 0., 0., 0.)
self.RCNN_TARGET_STDS = (0.1, 0.1, 0.2, 0.2)
# ROI Feat Size
self.POOL_SIZE = (7, 7)
# RCNN training configuration
self.RCNN_BATCH_SIZE = 256
self.RCNN_POS_FRAC = 0.25
self.RCNN_POS_IOU_THR = [
0.5, 0.6, 0.7
] # This can be smaller like [0.3, 0.4, 0.5] / [0.4, 0.5, 0.6]
self.RCNN_NEG_IOU_THR = [0.5, 0.4, 0.3]
# Boxes kept configuration
self.RCNN_MIN_CONFIDENCE = 0.05
self.RCNN_NMS_THRESHOLD = 0.5
self.RCNN_MAX_INSTANCES = 100
# Target Generator for the second stage.
self.bbox_target1 = bbox_target.ProposalTarget(
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
num_rcnn_deltas=self.RCNN_BATCH_SIZE,
positive_fraction=self.RCNN_POS_FRAC,
pos_iou_thr=self.RCNN_POS_IOU_THR[0],
neg_iou_thr=self.RCNN_NEG_IOU_THR[0],
num_classes=self.NUM_CLASSES)
self.bbox_target2 = bbox_target.ProposalTarget(
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
num_rcnn_deltas=self.RCNN_BATCH_SIZE,
positive_fraction=self.RCNN_POS_FRAC,
pos_iou_thr=self.RCNN_POS_IOU_THR[1],
neg_iou_thr=self.RCNN_NEG_IOU_THR[1],
num_classes=self.NUM_CLASSES)
self.bbox_target3 = bbox_target.ProposalTarget(
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
num_rcnn_deltas=self.RCNN_BATCH_SIZE,
positive_fraction=self.RCNN_POS_FRAC,
pos_iou_thr=self.RCNN_POS_IOU_THR[2],
neg_iou_thr=self.RCNN_NEG_IOU_THR[2],
num_classes=self.NUM_CLASSES)
# Modules
self.backbone = resnet.ResNet(depth=101, name='res_net')
self.neck = fpn.FPN(name='fpn')
self.rpn_head = rpn_head.RPNHead(
anchor_scales=self.ANCHOR_SCALES,
anchor_ratios=self.ANCHOR_RATIOS,
anchor_feature_strides=self.ANCHOR_FEATURE_STRIDES,
proposal_count=self.PRN_PROPOSAL_COUNT,
nms_threshold=self.PRN_NMS_THRESHOLD,
target_means=self.RPN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
num_rpn_deltas=self.PRN_BATCH_SIZE,
positive_fraction=self.RPN_POS_FRAC,
pos_iou_thr=self.RPN_POS_IOU_THR,
neg_iou_thr=self.RPN_NEG_IOU_THR,
name='rpn_head')
self.roi_align = roi_align.PyramidROIAlign(pool_shape=self.POOL_SIZE,
name='pyramid_roi_align')
# first detection stage
self.bbox_head1 = bbox_head.BBoxHead(
num_classes=self.NUM_CLASSES,
pool_size=self.POOL_SIZE,
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RCNN_TARGET_STDS,
min_confidence=self.RCNN_MIN_CONFIDENCE,
nms_threshold=self.RCNN_NMS_THRESHOLD,
max_instances=self.RCNN_MAX_INSTANCES,
name='b_box_head1')
# second detection stage
self.bbox_head2 = bbox_head.BBoxHead(
num_classes=self.NUM_CLASSES,
pool_size=self.POOL_SIZE,
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RCNN_TARGET_STDS,
min_confidence=self.RCNN_MIN_CONFIDENCE,
nms_threshold=self.RCNN_NMS_THRESHOLD,
max_instances=self.RCNN_MAX_INSTANCES,
name='b_box_head2')
# third detection stage
self.bbox_head3 = bbox_head.BBoxHead(
num_classes=self.NUM_CLASSES,
pool_size=self.POOL_SIZE,
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RCNN_TARGET_STDS,
min_confidence=self.RCNN_MIN_CONFIDENCE,
nms_threshold=self.RCNN_NMS_THRESHOLD,
max_instances=self.RCNN_MAX_INSTANCES,
name='b_box_head3')
def __init__(self, num_classes, **kwags):
super(FasterRCNN, self).__init__(**kwags)
self.NUM_CLASSES = num_classes
# Anchor attributes
self.ANCHOR_SCALES = (32, 64, 128, 256, 512)
self.ANCHOR_RATIOS = (0.5, 1, 2)
# The strides of each layer of the FPN Pyramid.
self.FEATURE_STRIDES = (4, 8, 16, 32, 64)
# Bounding box refinement mean and standard deviation
self.RPN_TARGET_MEANS = (0., 0., 0., 0.)
self.RPN_TARGET_STDS = (0.1, 0.1, 0.2, 0.2)
self.PRN_PROPOSAL_COUNT = 2000
self.PRN_NMS_THRESHOLD = 0.7
self.ROI_BATCH_SIZE = 512
# Bounding box refinement mean and standard deviation
self.RCNN_TARGET_MEANS = (0., 0., 0., 0.)
self.RCNN_TARGET_STDS = (0.1, 0.1, 0.2, 0.2)
self.POOL_SIZE = (7, 7)
self.backbone = resnet.ResNet(depth=101, name='res_net')
self.neck = fpn.FPN(name='fpn')
self.rpn_head = rpn_head.RPNHead(
anchors_per_location=len(self.ANCHOR_RATIOS),
proposal_count=self.PRN_PROPOSAL_COUNT,
nms_threshold=self.PRN_NMS_THRESHOLD,
target_means=self.RPN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
name='rpn_head')
self.roi_align = roi_align.PyramidROIAlign(pool_shape=self.POOL_SIZE,
name='pyramid_roi_align')
self.bbox_head = bbox_head.BBoxHead(num_classes=self.NUM_CLASSES,
pool_size=self.POOL_SIZE,
name='b_box_head')
self.generator = anchor_generator.AnchorGenerator(
scales=self.ANCHOR_SCALES,
ratios=self.ANCHOR_RATIOS,
feature_strides=self.FEATURE_STRIDES)
self.anchor_target = anchor_target.AnchorTarget(
target_means=self.RPN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS)
self.bbox_target = bbox_target.ProposalTarget(
target_means=self.RCNN_TARGET_MEANS,
target_stds=self.RPN_TARGET_STDS,
num_rcnn_deltas=self.ROI_BATCH_SIZE)
self.rpn_class_loss = losses.rpn_class_loss
self.rpn_bbox_loss = losses.rpn_bbox_loss
self.rcnn_class_loss = losses.rcnn_class_loss
self.rcnn_bbox_loss = losses.rcnn_bbox_loss