reduce time

app.py

@@ -89,8 +89,8 @@ def zeroshot_classifier(model, classnames, templates, device):
         zeroshot_weights = torch.stack(zeroshot_weights).cuda()
     return zeroshot_weights*100
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-# device = "cuda"
+# device = "cuda" if torch.cuda.is_available() else "cpu"
+device = "cpu"
 # Instantiate model
 vis_model = CLIPModel_Super("ViT-B/16", device=device, download_root="./ckpt")
 vis_model.eval()
@@ -135,7 +135,10 @@ def add_value_decrease(smdl_mask, json_file):
     
     return attribution_map, np.array(values)
 
-def visualization(image, submodular_image_set, saved_json_file, vis_image, index=None, compute_params=True):
+def visualization(image, submodular_image_set, saved_json_file, index=None, compute_params=True):
+    
+    attribution_map, value_list = add_value_decrease(submodular_image_set, saved_json_file)
+    vis_image, heatmap = gen_cam(image, norm_image(attribution_map))
     
     insertion_ours_images = []
     # deletion_ours_images = []
@@ -166,7 +169,7 @@ def visualization(image, submodular_image_set, saved_json_file, vis_image, index
     ax1.yaxis.set_visible(False)
     ax1.set_title('Attribution Map', fontsize=54)
     ax1.set_facecolor('white')
-    ax1.imshow(vis_image.astype(np.uint8))
+    ax1.imshow(vis_image[...,::-1].astype(np.uint8))
     
     ax2.spines["left"].set_visible(False)
     ax2.spines["right"].set_visible(False)
@@ -289,19 +292,20 @@ def interpret_image(uploaded_image, slider, text_input):
         return None, 0, 0
 
     image = cv2.resize(image, (224, 224))
-    element_sets_V = SubRegionDivision(image, mode="slico", region_size=30) 
+    element_sets_V = SubRegionDivision(image, mode="slico", region_size=40) 
     
     explainer.k = len(element_sets_V)
+    print(len(element_sets_V))
     
     global submodular_image_set
     global saved_json_file
-    global im
+    # global im
     submodular_image, submodular_image_set, saved_json_file = explainer(element_sets_V, id=None)
     
-    attribution_map, value_list = add_value_decrease(submodular_image_set, saved_json_file)
-    im, heatmap = gen_cam(image, norm_image(attribution_map))
+    # attribution_map, value_list = add_value_decrease(submodular_image_set, saved_json_file)
+    # im, heatmap = gen_cam(image, norm_image(attribution_map))
     
-    image_curve, highest_confidence, insertion_auc_score, ours_best_index = visualization(image, submodular_image_set, saved_json_file, im, index=None)
+    image_curve, highest_confidence, insertion_auc_score, ours_best_index = visualization(image, submodular_image_set, saved_json_file, index=None)
 
     text_output_class = "The method explains why the CLIP (ViT-B/16) model identifies an image as {}.".format(imagenet_classes[explainer.target_label])
 
@@ -316,7 +320,7 @@ def visualization_slider(uploaded_image, slider):
 
     image = cv2.resize(image, (224, 224))
     
-    image_curve = visualization(image, submodular_image_set, saved_json_file, im, index=slider, compute_params=False)
+    image_curve = visualization(image, submodular_image_set, saved_json_file, index=slider, compute_params=False)
     
     return image_curve
 
@@ -326,6 +330,7 @@ def update_image(thumbnail_name):
 
 # 创建 Gradio 界面
 with gr.Blocks() as demo:
+    gr.Markdown("# Semantic Region Attribution via Submodular Subset Selection")  # 使用Markdown添加标题
     with gr.Row():
         with gr.Column():
             # 第一排：上传图像输入框和一个缩略图
@@ -364,7 +369,7 @@ with gr.Blocks() as demo:
             # 输出图像和控件
             image_output = gr.Image(label="Output Image")
             
-            slider = gr.Slider(minimum=0, maximum=50, step=1, label="Confidence Slider")
+            slider = gr.Slider(minimum=0, maximum=34, step=1, label="Number of Sub-regions")
             
             text_output_class = gr.Textbox(label="Explaining Category")
             with gr.Row():