Spaces:
Sleeping
Sleeping
Upload resnet.py
Browse files
resnet.py
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch.nn as nn
|
2 |
+
import torchvision
|
3 |
+
|
4 |
+
class Resnet50Flower102(nn.Module):
|
5 |
+
def __init__(self, device, pretrained=True, freeze_backbone=True):
|
6 |
+
super().__init__()
|
7 |
+
self.device = device
|
8 |
+
|
9 |
+
if pretrained:
|
10 |
+
weights = torchvision.models.ResNet50_Weights.IMAGENET1K_V1
|
11 |
+
else:
|
12 |
+
weights = None
|
13 |
+
|
14 |
+
self.model = torchvision.models.resnet50(weights=weights)
|
15 |
+
|
16 |
+
self.model.fc = nn.Sequential(
|
17 |
+
nn.Linear(2048, 1024),
|
18 |
+
nn.BatchNorm1d(1024),
|
19 |
+
nn.ReLU(),
|
20 |
+
nn.Dropout(0.2),
|
21 |
+
nn.Linear(1024, 512),
|
22 |
+
nn.BatchNorm1d(512),
|
23 |
+
nn.ReLU(),
|
24 |
+
nn.Dropout(0.2),
|
25 |
+
nn.Linear(512, 102),
|
26 |
+
)
|
27 |
+
self.model.to(device)
|
28 |
+
|
29 |
+
def forward(self, x):
|
30 |
+
return self.model(x)
|