Upload inference.py
Browse files- inference.py +48 -0
inference.py
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import time
|
3 |
+
from PIL import Image
|
4 |
+
import torch
|
5 |
+
from torchvision.transforms import transforms
|
6 |
+
|
7 |
+
model = torch.load('/path/to/your/model.pth').to("cuda")
|
8 |
+
model.eval()
|
9 |
+
transform = transforms.Compose([
|
10 |
+
transforms.Resize((448, 448)),
|
11 |
+
transforms.ToTensor(),
|
12 |
+
transforms.Normalize(mean=[
|
13 |
+
0.48145466,
|
14 |
+
0.4578275,
|
15 |
+
0.40821073
|
16 |
+
], std=[
|
17 |
+
0.26862954,
|
18 |
+
0.26130258,
|
19 |
+
0.27577711
|
20 |
+
])
|
21 |
+
])
|
22 |
+
|
23 |
+
with open("tags_8041.json", "r") as file:
|
24 |
+
tags = json.load(file)
|
25 |
+
allowed_tags = sorted(tags)
|
26 |
+
allowed_tags.insert(0, "placeholder0")
|
27 |
+
allowed_tags.append("placeholder1")
|
28 |
+
allowed_tags.append("explicit")
|
29 |
+
allowed_tags.append("questionable")
|
30 |
+
allowed_tags.append("safe")
|
31 |
+
|
32 |
+
image_path = "/path/to/your/image.jpg"
|
33 |
+
start = time.time()
|
34 |
+
img = Image.open(image_path).convert('RGB')
|
35 |
+
tensor = transform(img).unsqueeze(0).to("cuda") # transform and add batch dimension
|
36 |
+
|
37 |
+
with torch.no_grad():
|
38 |
+
out = model(tensor)
|
39 |
+
probabilities = torch.nn.functional.sigmoid(out[0])
|
40 |
+
indices = torch.where(probabilities > 0.3)[0]
|
41 |
+
values = probabilities[indices]
|
42 |
+
|
43 |
+
for i in range(indices.size(0)):
|
44 |
+
print(allowed_tags[indices[i]], values[i].item())
|
45 |
+
|
46 |
+
end = time.time()
|
47 |
+
print(f'Executed in {end - start} seconds')
|
48 |
+
print("\n\n", end="")
|