Update app.py

app.py

@@ -10,7 +10,6 @@ import matplotlib.pyplot as plt
 import concurrent.futures
 from model.CyueNet_models import MMS
 from utils1.data import transform_image
-import plotly.graph_objects as go
 from datetime import datetime
 import io
 import base64
@@ -18,7 +17,7 @@ import base64
 # GPU/CPU设置
 device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
 
-# 更新后的CSS样式
+# CSS样式设置
 custom_css = """
     :root {
         --primary-color: #2196F3;
@@ -138,7 +137,6 @@ custom_css = """
         border: 1px solid rgba(255, 255, 255, 0.18);
     }
 """
-
 class ImageProcessor:
     def __init__(self):
         self.model = None
@@ -263,12 +261,17 @@ class ImageProcessor:
         time_start = time.time()
         
         with torch.no_grad():
-            with torch.cuda.amp.autocast() if device.type == 'cuda' else torch.cpu.amp.autocast():
-                _, res, *_ = self.model(image_tensor)
+            if device.type == 'cuda':
+                with torch.cuda.amp.autocast():
+                    _, res, *_ = self.model(image_tensor)
+            else:
+                with torch.amp.autocast(device_type='cpu'):
+                    _, res, *_ = self.model(image_tensor)
         
         time_end = time.time()
         
-        res = res.sigmoid().data.cpu().numpy().squeeze()
+        # 确保转换为float32类型
+        res = res.to(torch.float32).sigmoid().cpu().numpy().squeeze()
         res = (res - res.min()) / (res.max() - res.min() + 1e-8)
         
         h, w = image.shape[:2]
@@ -279,7 +282,6 @@ class ImageProcessor:
         self.cache[cache_key] = result
         
         return result
-
     def process_image(self, image, threshold=0.5, testsize=256, 
                      enhance_contrast=False, denoise=False, 
                      brightness=0, contrast=0, filter_type="None",
@@ -288,7 +290,7 @@ class ImageProcessor:
         if image is None:
             return [None] * 9 + ["Please provide a valid image"]
 
-        # Quick mode
+        # Quick mode check
         if process_mode == "Quick Mode":
             saliency_map, time_info = self.quick_process(image, threshold, testsize)
             return (image, saliency_map, None, None, None, None, time_info, None, None)
@@ -334,14 +336,18 @@ class ImageProcessor:
         time_start = time.time()
         
         with torch.no_grad():
-            with torch.cuda.amp.autocast() if device.type == 'cuda' else torch.cpu.amp.autocast():
-                x1, res, s1_sig, edg1, edg_s, s2, e2, s2_sig, e2_sig, s3, e3, s3_sig, e3_sig, s4, e4, s4_sig, e4_sig, s5, e5, s5_sig, e5_sig, sk1, sk1_sig, sk2, sk2_sig, sk3, sk3_sig, sk4, sk4_sig, sk5, sk5_sig = self.model(image_tensor)
+            if device.type == 'cuda':
+                with torch.cuda.amp.autocast():
+                    x1, res, s1_sig, edg1, edg_s, s2, e2, s2_sig, e2_sig, s3, e3, s3_sig, e3_sig, s4, e4, s4_sig, e4_sig, s5, e5, s5_sig, e5_sig, sk1, sk1_sig, sk2, sk2_sig, sk3, sk3_sig, sk4, sk4_sig, sk5, sk5_sig = self.model(image_tensor)
+            else:
+                with torch.amp.autocast(device_type='cpu'):
+                    x1, res, s1_sig, edg1, edg_s, s2, e2, s2_sig, e2_sig, s3, e3, s3_sig, e3_sig, s4, e4, s4_sig, e4_sig, s5, e5, s5_sig, e5_sig, sk1, sk1_sig, sk2, sk2_sig, sk3, sk3_sig, sk4, sk4_sig, sk5, sk5_sig = self.model(image_tensor)
         
         time_end = time.time()
         inference_time = time_end - time_start
         
-        # Process outputs
-        res = res.sigmoid().data.cpu().numpy().squeeze()
+        # 确保转换为float32类型
+        res = res.to(torch.float32).sigmoid().cpu().numpy().squeeze()
         res = (res - res.min()) / (res.max() - res.min() + 1e-8)
         
         h, w = original_image.shape[:2]
@@ -423,7 +429,6 @@ class ImageProcessor:
             return f"PDF report saved as saliency_report_{timestamp}.pdf"
         else:
             return f"Results exported as {format_type.lower()} files"
-
 # Create processor instance
 processor = ImageProcessor()
 
@@ -720,11 +725,6 @@ with gr.Blocks(title="Advanced Saliency Object Detection System", css=custom_css
             process_mode: "Full Analysis"
         }
     
-    def update_comparison(original, overlay, slider_val):
-        if original is not None and overlay is not None:
-            return create_comparison_view(original, overlay, slider_val)
-        return None
-    
     # Set up event handling
     submit_btn.click(
         fn=run_demo,
@@ -769,7 +769,7 @@ with gr.Blocks(title="Advanced Saliency Object Detection System", css=custom_css
     
     # Interactive comparison
     comparison_slider.change(
-        fn=update_comparison,
+        fn=create_comparison_view,
         inputs=[original_output, overlayed_output, comparison_slider],
         outputs=[interactive_comparison]
     )
@@ -787,6 +787,5 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         share=True,
-        inbrowser=True,
         show_error=True
     )