def __init__(self,
num_classes,
mode="training",
feat_stride=16,
anchor_scales=[8, 16, 32],
ratios=[0.5, 1, 2],
backbone='vgg'):
super(FasterRCNN, self).__init__()
self.feat_stride = feat_stride
if backbone == 'vgg':
self.extractor, classifier = decom_vgg16()
self.rpn = RegionProposalNetwork(512,
512,
ratios=ratios,
anchor_scales=anchor_scales,
feat_stride=self.feat_stride,
mode=mode)
self.head = VGG16RoIHead(n_class=num_classes + 1,
roi_size=7,
spatial_scale=1,
classifier=classifier)
elif backbone == 'resnet50':
self.extractor, classifier = resnet50()
self.rpn = RegionProposalNetwork(1024,
512,
ratios=ratios,
anchor_scales=anchor_scales,
feat_stride=self.feat_stride,
mode=mode)
self.head = Resnet50RoIHead(n_class=num_classes + 1,
roi_size=14,
spatial_scale=1,
classifier=classifier)
def __init__(self, num_classes = 20, pretrain = False):
super(CenterNet_Resnet50, self).__init__()
# 512,512,3 -> 16,16,2048
self.backbone = resnet50(pretrain=pretrain)
# 16,16,2048 -> 128,128,64
self.decoder = resnet50_Decoder(2048)
#-----------------------------------------------------------------#
# 对获取到的特征进行上采样，进行分类预测和回归预测
# 128, 128, 64 -> 128, 128, 64 -> 128, 128, num_classes
# -> 128, 128, 64 -> 128, 128, 2
# -> 128, 128, 64 -> 128, 128, 2
#-----------------------------------------------------------------#
self.head = resnet50_Head(channel=64, num_classes=num_classes)
def __init__(self,
num_classes,
mode="training",
loc_normalize_mean=(0., 0., 0., 0.),
loc_normalize_std=(0.1, 0.1, 0.2, 0.2),
feat_stride=16,
anchor_scales=[8, 16, 32],
ratios=[0.5, 1, 2],
backbone='vgg'):
super(FasterRCNN, self).__init__()
self.loc_normalize_mean = loc_normalize_mean
self.loc_normalize_std = loc_normalize_std
self.feat_stride = feat_stride
if backbone == 'vgg':
self.extractor, self.classifier = decom_vgg16()
self.rpn = RegionProposalNetwork(512,
512,
ratios=ratios,
anchor_scales=anchor_scales,
feat_stride=self.feat_stride,
mode=mode)
self.head = VGG16RoIHead(n_class=num_classes + 1,
roi_size=7,
spatial_scale=(1. / self.feat_stride),
classifier=self.classifier)
elif backbone == 'resnet50':
self.extractor, self.classifier = resnet50()
# 完成RPN的3×3、两个1×1卷积、Proposal，返回 输出结果为 rpn_locs, rpn_scores, rois, roi_indices, anchor
# 其中rois为从先验框rpn_locs中选出的得分较高的候选框
self.rpn = RegionProposalNetwork(1024,
512,
ratios=ratios,
anchor_scales=anchor_scales,
feat_stride=self.feat_stride,
mode=mode)
self.head = Resnet50RoIHead(n_class=num_classes + 1,
roi_size=14,
spatial_scale=(1. / self.feat_stride),
classifier=self.classifier)
def __init__(self,num_classes,
mode='training',
loc_normalize_mean=(0.,0.,0.,0.),
loc_normalize_std=(0.1,0.1,0.2,0.2),
feat_stride=16,
anchor_scales=[8, 16, 32],
rotios = [0.5, 1, 2],
backbone='vgg'):
super(FasterRCNN,self).__init__()
self.loc_normalize_mean = loc_normalize_mean
self.loc_normalize_std = loc_normalize_std
self.feat_stride = feat_stride
if backbone == 'vgg' :
self.extractor,classifier = decom_vgg16()
self.rpn = RegionProposalNetwork(512,512,
ratios = rotios,
anchor_scales = anchor_scales,
feat_stride=feat_stride,
mode=mode)
self.head = VGG16RoIHead(n_class=num_classes+1,
roi_size=7,
spatial_scale=(1. / self.feat_stride),
classifier=classifier)
elif backbone == 'resnet50':
self.extractor,classifier = resnet50()
self.rpn = RegionProposalNetwork(1024,512,
ratios=rotios,
anchor_scales = anchor_scales,
feat_stride=feat_stride,
mode=mode)
self.head = Resnet50RoIHead(n_class=num_classes+1,
roi_size=14,
spatial_scale=(1. / self.feat_stride),
classifier=classifier)
def __init__(self, num_classes=20, pretrain=False):
super(CenterNet_Resnet50, self).__init__()
self.backbone = resnet50(pretrain=pretrain)
self.decoder = resnet50_Decoder(2048)
self.head = resnet50_Head(channel=64, num_classes=num_classes)