dd
Browse files
app.py
CHANGED
@@ -3,12 +3,30 @@ import gradio as gr
|
|
3 |
|
4 |
import torch
|
5 |
import torchvision
|
|
|
6 |
|
7 |
checkpoint = torch.load('v4-epoch=19-val_loss=0.6964-val_accuracy=0.8964.ckpt', map_location=torch.device('cpu'))
|
8 |
state_dict = checkpoint["state_dict"]
|
9 |
model_weights = state_dict
|
10 |
for key in list(model_weights):
|
11 |
model_weights[key.replace("backbone.", "")] = model_weights.pop(key)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
model.load_state_dict(model_weights).eval()
|
13 |
|
14 |
import requests
|
|
|
3 |
|
4 |
import torch
|
5 |
import torchvision
|
6 |
+
import timm
|
7 |
|
8 |
checkpoint = torch.load('v4-epoch=19-val_loss=0.6964-val_accuracy=0.8964.ckpt', map_location=torch.device('cpu'))
|
9 |
state_dict = checkpoint["state_dict"]
|
10 |
model_weights = state_dict
|
11 |
for key in list(model_weights):
|
12 |
model_weights[key.replace("backbone.", "")] = model_weights.pop(key)
|
13 |
+
|
14 |
+
|
15 |
+
def get_model():
|
16 |
+
model = timm.create_model('tf_efficientnet_b1', pretrained=True, num_classes=2, global_pool='catavgmax')
|
17 |
+
num_in_features = model.get_classifier().in_features
|
18 |
+
from torch import nn
|
19 |
+
|
20 |
+
model.fc = nn.Sequential(
|
21 |
+
nn.Linear(in_features=num_in_features, out_features=1024, bias=False),
|
22 |
+
nn.ReLU(),
|
23 |
+
nn.Linear(in_features=1024, out_features=2, bias=False),
|
24 |
+
)
|
25 |
+
|
26 |
+
return model
|
27 |
+
|
28 |
+
model = get_model()
|
29 |
+
|
30 |
model.load_state_dict(model_weights).eval()
|
31 |
|
32 |
import requests
|