def __init__(self, cfg, in_channels):
# from .loss import make_rpn_loss_evaluator
# from .inference import make_rpn_postprocessor
super(RPNModule, self).__init__()
self.cfg = cfg
rpn_box_coder = BoxCoder(
weights=cfg.RPN.BOX_REG_WEIGHTS,
lib=torch) # weights=(1.0, 1.0, 1.0, 1.0, 1.0))
self.anchor_generator = make_anchor_generator(cfg)
self.head = RPNHead(
in_channels,
self.anchor_generator.num_anchors_per_location()[0])
# raise NotImplementedError
self.box_selector_train = make_rpn_postprocessor(cfg,
rpn_box_coder,
is_train=True)
self.box_selector_test = make_rpn_postprocessor(cfg,
rpn_box_coder,
is_train=False)
self.loss_evaluator = make_rpn_loss_evaluator(cfg, rpn_box_coder)
def __init__(
self,
pre_nms_top_n,
post_nms_top_n,
nms_thresh,
min_size,
box_coder=None,
fpn_post_nms_top_n=None,
):
"""
Arguments:
pre_nms_top_n (int)
post_nms_top_n (int)
nms_thresh (float)
min_size (int)
box_coder (BoxCoder)
fpn_post_nms_top_n (int)
"""
super(RPNPostProcessor, self).__init__()
self.pre_nms_top_n = pre_nms_top_n
self.post_nms_top_n = post_nms_top_n
self.nms_thresh = nms_thresh
self.min_size = min_size
self.nms_rotate = RotateNMS(nms_threshold=nms_thresh,
post_nms_top_n=post_nms_top_n)
if box_coder is None:
box_coder = BoxCoder(weights=(1.0, 1.0, 1.0, 1.0, 1.0))
self.box_coder = box_coder
if fpn_post_nms_top_n is None:
fpn_post_nms_top_n = post_nms_top_n
self.fpn_post_nms_top_n = fpn_post_nms_top_n
def make_roi_box_post_processor(cfg):
bbox_reg_weights = cfg.ROI_HEADS.BBOX_REG_WEIGHTS
box_coder = BoxCoder(weights=bbox_reg_weights, lib=torch)
score_thresh = cfg.ROI_HEADS.SCORE_THRESH
nms_thresh = cfg.ROI_HEADS.NMS
detections_per_img = cfg.ROI_HEADS.DETECTIONS_PER_IMG
cls_agnostic_bbox_reg = cfg.CLS_AGNOSTIC_BBOX_REG
postprocessor = PostProcessor(score_thresh, nms_thresh, detections_per_img,
box_coder, cls_agnostic_bbox_reg)
return postprocessor
def __init__(self, cfg, in_channels):
super(RPNModule, self).__init__()
anchor_generator = make_anchor_generator(cfg)
head = RPNHead(
in_channels, anchor_generator.num_anchors_per_location()[0]
)
rpn_box_coder = BoxCoder(weights=(1.0, 1.0, 1.0, 1.0))
box_selector_train = make_rpn_postprocessor(cfg, rpn_box_coder, is_train=True)
box_selector_test = make_rpn_postprocessor(cfg, rpn_box_coder, is_train=False)
loss_evaluator = make_rpn_loss_evaluator(cfg, rpn_box_coder)
self.anchor_generator = anchor_generator
self.head = head
self.box_selector_train = box_selector_train
self.box_selector_test = box_selector_test
self.loss_evaluator = loss_evaluator
def make_roi_box_loss_evaluator(cfg):
from matcher import Matcher
from sampler import BalancedPositiveNegativeSampler
matcher = Matcher(
cfg.ROI_HEADS.FG_IOU_THRESHOLD,
cfg.ROI_HEADS.BG_IOU_THRESHOLD,
allow_low_quality_matches=False,
)
bbox_reg_weights = cfg.ROI_HEADS.BBOX_REG_WEIGHTS
box_coder = BoxCoder(weights=bbox_reg_weights, lib=torch)
fg_bg_sampler = BalancedPositiveNegativeSampler(
cfg.ROI_HEADS.BATCH_SIZE_PER_IMAGE, cfg.ROI_HEADS.POSITIVE_FRACTION)
cls_agnostic_bbox_reg = cfg.CLS_AGNOSTIC_BBOX_REG
loss_evaluator = FastRCNNLossComputation(matcher, fg_bg_sampler, box_coder,
cls_agnostic_bbox_reg)
return loss_evaluator
def __init__(self,
score_thresh=0.05,
nms=0.5,
detections_per_img=100,
box_coder=None,
cls_agnostic_bbox_reg=False):
"""
Arguments:
box_coder (BoxCoder)
"""
super(PostProcessor, self).__init__()
self.score_thresh = score_thresh
self.nms = nms
self.detections_per_img = detections_per_img
self.nms_rotate = RotateNMS(nms_threshold=nms,
post_nms_top_n=detections_per_img)
if box_coder is None:
box_coder = BoxCoder(weights=None, lib=torch)
self.box_coder = box_coder
self.cls_agnostic_bbox_reg = cls_agnostic_bbox_reg