requirement

app.py

@@ -1,3 +1,4 @@
+import gradio as gr
 import torch
 from torchvision import transforms, models
 from art.attacks.evasion import FastGradientMethod
@@ -17,10 +18,7 @@ def load_model():
     model.eval()
     return model
 
-def process_image(inputs: dict):
-    input_image = inputs["inputs"]
-    eps_value = inputs.get("eps", 0.3)
-
+def process_image(input_image, eps_value=0.3):
     model = load_model()
     device = torch.device("cpu")
     model = model.to(device)
@@ -42,8 +40,7 @@ def process_image(inputs: dict):
                              std=[0.229, 0.224, 0.225])
     ])
 
-    img = Image.open(io.BytesIO(base64.b64decode(input_image))).convert('RGB')
-    img_tensor = transform(img).unsqueeze(0).to(device)
+    img_tensor = transform(input_image).unsqueeze(0).to(device)
 
     attack = FastGradientMethod(estimator=classifier, eps=eps_value)
     adv_img_tensor = attack.generate(x=img_tensor.cpu().numpy())
@@ -69,4 +66,14 @@ def process_image(inputs: dict):
     bwm.embed(img_bytes)
     result_image = base64.b64encode(img_bytes.getvalue()).decode('utf-8')
 
-    return {"image": result_image}
+    return result_image
+
+# Gradio 인터페이스 정의
+def inference(image, eps_value):
+    return process_image(image, eps_value)
+
+gr.Interface(
+    fn=inference,
+    inputs=[gr.inputs.Image(type="pil"), gr.inputs.Slider(0.1, 1.0, step=0.1, default=0.3, label="Epsilon")],
+    outputs="image"
+).launch()