def __init__(self, num_classes=1):
super(RetinaNet, self).__init__()
self.num_anchors = 7 * 2 # vertical offset -> *2
self.num_classes = num_classes
self.fpn = FPN50()
self.loc_head = self._make_head(self.num_anchors * 8)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
def __init__(self, num_classes=80):
super(RetinaNet, self).__init__()
self.fpn = FPN50()
self.num_classes = num_classes
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
self.focal_loss = FocalLoss()
def __init__(self, num_classes=500, num_anchors=9):
super().__init__()
self.fpn = FPN50()
self.num_classes = num_classes
self.num_anchors = num_anchors
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
def __init__(self, num_classes=20, num_anchors=9, backbone='resnet50'):
super(RetinaNet, self).__init__()
if backbone == 'resnet50':
self.fpn = FPN50()
elif backbone == 'resnet101':
self.fpn = FPN101()
else:
print('Invalid backbone network')
self.num_classes = num_classes
self.num_anchors = num_anchors
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
def __init__(self, num_classes=9):
super(RetinaNet, self).__init__()
self.fpn = FPN50()
self.num_classes = num_classes
self.classifier1 = nn.Conv2d(3,
18,
kernel_size=3,
padding=1,
bias=False)
self.classifier2 = nn.Conv2d(108,
3,
kernel_size=3,
padding=1,
bias=False)
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
def __init__(self, num_classes):
super(FPNSSD512, self).__init__()
self.num_classes = num_classes
self.extractor = FPN50()
self.loc_layers = nn.ModuleList()
self.cls_layers = nn.ModuleList()
in_channels = 256
num_anchors = (4, 6, 6, 6, 6, 4, 4)
for i in range(len(num_anchors)):
self.loc_layers += [
nn.Conv2d(in_channels,
num_anchors[i] * 4,
kernel_size=3,
padding=1)
]
self.cls_layers += [
nn.Conv2d(in_channels,
num_anchors[i] * num_classes,
kernel_size=3,
padding=1)
]
def __init__(self, layers):
super(poseNet, self).__init__()
if layers == 101:
self.fpn = FPN101()
if layers == 50:
self.fpn = FPN50()
self.latlayer4 = nn.Conv2d(256,
256,
kernel_size=1,
stride=1,
padding=0)
# D-layers
# 两个3x3卷积核，把channels降到128
self.convt1 = nn.Conv2d(256, 128, kernel_size=3, stride=1, padding=1)
self.convt2 = nn.Conv2d(256, 128, kernel_size=3, stride=1, padding=1)
self.convt3 = nn.Conv2d(256, 128, kernel_size=3, stride=1, padding=1)
self.convt4 = nn.Conv2d(256, 128, kernel_size=3, stride=1, padding=1)
self.convs1 = nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1)
self.convs2 = nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1)
self.convs3 = nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1)
self.convs4 = nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1)
self.upsample1 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=4,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
# self.upsample4 = nn.Upsample(size=(120,120),mode='bilinear',align_corners=True)
self.concat = Concat()
self.conv2 = nn.Conv2d(512, 256, kernel_size=3, stride=1, padding=1)
self.convfin = nn.Conv2d(256, 17, kernel_size=1, stride=1, padding=0)
def __init__(self, num_classes, firstinit=False):
super(RetinaNet, self).__init__()
self.name = 'RetinaNet'
self.num_classes = num_classes
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
self.best_loss = None
self.lr = None
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.weight.data.normal_(0, 0.01)
m.bias.data.zero_()
self.fpn = FPN50(firstinit)
https://download.pytorch.org/models/resnet50-19c8e357.pth
'''
import math
import torch
from torch import nn
from torch.nn import init
from fpn import FPN50
from retinanet import RetinaNet
print('Loading pretrained ResNet50 model..')
d = torch.load('./model/resnet50.pth')
print('Loading into FPN50..')
fpn = FPN50()
dd = fpn.state_dict()
for k in d.keys():
if not k.startswith('fc'): # skip fc layers
dd[k] = d[k]
print('Saving RetinaNet..')
net = RetinaNet()
for m in net.modules():
if isinstance(m, nn.Conv2d):
init.normal(m.weight, mean=0, std=0.01)
if m.bias is not None:
init.constant(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def __init__(self):
super(RetinaNet, self).__init__()
self.fpn = FPN50()
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
def __init__(self, num_classes):
super(FPNSSD512, self).__init__()
self.fpn = FPN50()
self.num_classes = num_classes
self.loc_head = self._make_head(self.num_anchors * 4)
self.cls_head = self._make_head(self.num_anchors * self.num_classes)
