Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import random
|
3 |
+
random.seed(999)
|
4 |
+
import torch
|
5 |
+
from torchvision.transforms import transforms
|
6 |
+
import gradio as gr
|
7 |
+
|
8 |
+
model = torch.load('model.pth', map_location=torch.device('cpu'))
|
9 |
+
model.eval()
|
10 |
+
transform = transforms.Compose([
|
11 |
+
transforms.Resize((384, 384)),
|
12 |
+
transforms.ToTensor(),
|
13 |
+
transforms.Normalize(
|
14 |
+
mean=[
|
15 |
+
0.5,
|
16 |
+
0.5,
|
17 |
+
0.5,
|
18 |
+
], std=[
|
19 |
+
0.5,
|
20 |
+
0.5,
|
21 |
+
0.5,
|
22 |
+
])
|
23 |
+
])
|
24 |
+
|
25 |
+
with open("tags_9940.json", "r") as file:
|
26 |
+
allowed_tags = json.load(file)
|
27 |
+
|
28 |
+
allowed_tags = sorted(allowed_tags)
|
29 |
+
allowed_tags.append("explicit")
|
30 |
+
allowed_tags.append("questionable")
|
31 |
+
allowed_tags.append("safe")
|
32 |
+
|
33 |
+
def create_tags(image, threshold):
|
34 |
+
img = image.convert('RGB')
|
35 |
+
tensor = transform(img).unsqueeze(0)
|
36 |
+
|
37 |
+
with torch.no_grad():
|
38 |
+
logits = model(tensor)
|
39 |
+
probabilities = torch.nn.functional.sigmoid(logits[0])
|
40 |
+
indices = torch.where(probabilities > threshold)[0]
|
41 |
+
values = probabilities[indices]
|
42 |
+
|
43 |
+
temp = []
|
44 |
+
tag_score = dict()
|
45 |
+
for i in range(indices.size(0)):
|
46 |
+
temp.append([allowed_tags[indices[i]], values[i].item()])
|
47 |
+
tag_score[allowed_tags[indices[i]]] = values[i].item()
|
48 |
+
# temp = sorted(temp, key=lambda x: x[1], reverse=True)
|
49 |
+
# print("Before adding implicated tags, there are " + str(len(temp)) + " tags")
|
50 |
+
temp = [t[0] for t in temp]
|
51 |
+
text_no_impl = " ".join(temp)
|
52 |
+
return text_no_impl, tag_score
|
53 |
+
|
54 |
+
demo = gr.Interface(
|
55 |
+
create_tags,
|
56 |
+
inputs=[gr.Image(label="Source", sources=['upload', 'webcam'], type='pil'), gr.Slider(minimum=0.00, maximum=1.00, step=0.01, value=0.30, label="Threshold")],
|
57 |
+
outputs=[
|
58 |
+
gr.Textbox(label="Tag String"),
|
59 |
+
gr.Label(label="Tag Predictions", num_top_classes=200),
|
60 |
+
],
|
61 |
+
allow_flagging="never",
|
62 |
+
)
|
63 |
+
|
64 |
+
demo.launch()
|