Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -76,7 +76,45 @@ advanced_css = f"""
|
|
76 |
background: linear-gradient(90deg, {theme_color} 0%, #C06C84 100%);
|
77 |
}}
|
78 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
with gr.Blocks(theme=gr.themes.Soft(), css=advanced_css) as demo:
|
81 |
# 标题区
|
82 |
with gr.Column(elem_classes="header-section"):
|
|
|
76 |
background: linear-gradient(90deg, {theme_color} 0%, #C06C84 100%);
|
77 |
}}
|
78 |
"""
|
79 |
+
# Define CNN model
|
80 |
+
class Classifier(nn.Module):
|
81 |
+
def __init__(self):
|
82 |
+
super(Classifier, self).__init__()
|
83 |
+
self.cnn_layers = resnet18(weights=ResNet18_Weights.DEFAULT)
|
84 |
+
self.fc_layers = nn.Sequential(
|
85 |
+
nn.Linear(1000, 512),
|
86 |
+
nn.Dropout(0.3),
|
87 |
+
nn.Linear(512, 128),
|
88 |
+
nn.ReLU(),
|
89 |
+
nn.Linear(128, 5),
|
90 |
+
)
|
91 |
|
92 |
+
def forward(self, x):
|
93 |
+
x = self.cnn_layers(x)
|
94 |
+
x = self.fc_layers(x)
|
95 |
+
return x
|
96 |
+
|
97 |
+
# Pre-process
|
98 |
+
preprocess = transforms.Compose([
|
99 |
+
transforms.Resize(224),
|
100 |
+
transforms.ToTensor(),
|
101 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
102 |
+
])
|
103 |
+
|
104 |
+
# Load model
|
105 |
+
model = Classifier()
|
106 |
+
model.load_state_dict(torch.load("classify_nsfw_v3.0.pth", map_location="cpu"))
|
107 |
+
model.eval()
|
108 |
+
|
109 |
+
def predict(image_path):
|
110 |
+
img = Image.open(image_path).convert("RGB")
|
111 |
+
img = preprocess(img).unsqueeze(0)
|
112 |
+
|
113 |
+
with torch.no_grad():
|
114 |
+
prediction = torch.nn.functional.softmax(model(img)[0], dim=0)
|
115 |
+
|
116 |
+
result = {labels[i]: float(prediction[i]) for i in range(5)}
|
117 |
+
return result
|
118 |
with gr.Blocks(theme=gr.themes.Soft(), css=advanced_css) as demo:
|
119 |
# 标题区
|
120 |
with gr.Column(elem_classes="header-section"):
|