2025-08-01 10:04 🚀

app.py

@@ -2,6 +2,7 @@ import torch
 from torch import nn
 import torch.nn.functional as F
 import torchvision.transforms.functional as TF
+from torch.amp import autocast
 
 from torch import Tensor
 import spaces
@@ -28,19 +29,19 @@ pretrained_models = [
     "ZIP-B @ ShanghaiTech B @ MAE", "ZIP-B @ ShanghaiTech B @ NAE",
     "ZIP-B @ UCF-QNRF @ MAE", "ZIP-B @ UCF-QNRF @ NAE",
     "ZIP-B @ NWPU-Crowd @ MAE", "ZIP-B @ NWPU-Crowd @ NAE",
-    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
     "ZIP-S @ ShanghaiTech A @ MAE", "ZIP-S @ ShanghaiTech A @ NAE",
     "ZIP-S @ ShanghaiTech B @ MAE", "ZIP-S @ ShanghaiTech B @ NAE",
     "ZIP-S @ UCF-QNRF @ MAE", "ZIP-S @ UCF-QNRF @ NAE",
-    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
     "ZIP-T @ ShanghaiTech A @ MAE", "ZIP-T @ ShanghaiTech A @ NAE",
     "ZIP-T @ ShanghaiTech B @ MAE", "ZIP-T @ ShanghaiTech B @ NAE",
     "ZIP-T @ UCF-QNRF @ MAE", "ZIP-T @ UCF-QNRF @ NAE",
-    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
     "ZIP-N @ ShanghaiTech A @ MAE", "ZIP-N @ ShanghaiTech A @ NAE",
     "ZIP-N @ ShanghaiTech B @ MAE", "ZIP-N @ ShanghaiTech B @ NAE",
     "ZIP-N @ UCF-QNRF @ MAE", "ZIP-N @ UCF-QNRF @ NAE",
-    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
     "ZIP-P @ ShanghaiTech A @ MAE", "ZIP-P @ ShanghaiTech A @ NAE",
     "ZIP-P @ ShanghaiTech B @ MAE", "ZIP-P @ ShanghaiTech B @ NAE",
     "ZIP-P @ UCF-QNRF @ MAE", "ZIP-P @ UCF-QNRF @ NAE",
@@ -238,7 +239,7 @@ def _sliding_window_predict(
     model.eval()
     pi_maps, lambda_maps = [], []
     for i in range(0, len(windows), max_num_windows):
-        with torch.no_grad():
+        with torch.no_grad(), autocast(device_type="cuda" if torch.cuda.is_available() else "cpu"):
             image_feats = model.backbone(windows[i: min(i + max_num_windows, len(windows))])
             pi_image_feats, lambda_image_feats = model.pi_head(image_feats), model.lambda_head(image_feats)
             pi_image_feats = F.normalize(pi_image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)
@@ -535,6 +536,62 @@ option[value*="━━━━━━"] {
     text-align: center;
 }
 
+/* 下拉框滑动条样式 */
+.gr-dropdown select {
+    max-height: 200px;
+    overflow-y: auto;
+}
+
+/* 下拉框选项容器样式 */
+.gr-dropdown .choices__list {
+    max-height: 200px;
+    overflow-y: auto;
+}
+
+.gr-dropdown .choices__list--dropdown {
+    max-height: 200px;
+    overflow-y: auto;
+    border: 1px solid #e5e7eb;
+    border-radius: 6px;
+    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
+}
+
+/* 确保下拉框容器不被裁剪 */
+.gr-dropdown {
+    position: relative;
+    z-index: 1000;
+}
+
+/* 自定义滚动条样式 - WebKit浏览器 */
+.gr-dropdown select::-webkit-scrollbar,
+.gr-dropdown .choices__list::-webkit-scrollbar {
+    width: 8px;
+}
+
+.gr-dropdown select::-webkit-scrollbar-track,
+.gr-dropdown .choices__list::-webkit-scrollbar-track {
+    background: #f1f1f1;
+    border-radius: 4px;
+}
+
+.gr-dropdown select::-webkit-scrollbar-thumb,
+.gr-dropdown .choices__list::-webkit-scrollbar-thumb {
+    background: #c1c1c1;
+    border-radius: 4px;
+}
+
+.gr-dropdown select::-webkit-scrollbar-thumb:hover,
+.gr-dropdown .choices__list::-webkit-scrollbar-thumb:hover {
+    background: #a1a1a1;
+}
+
+/* Firefox滚动条样式 */
+.gr-dropdown select,
+.gr-dropdown .choices__list {
+    scrollbar-width: thin;
+    scrollbar-color: #c1c1c1 #f1f1f1;
+}
+
 /* 基础组件样式 */
 .gr-group {
     background: white;
@@ -597,7 +654,10 @@ with gr.Blocks(css=css, theme=gr.themes.Soft(), title="ZIP Crowd Counting") as d
                     choices=pretrained_models,
                     value="ZIP-B @ NWPU-Crowd @ MAE",
                     label="🎛️ Select Model & Dataset",
-                    info="Choose model variant, dataset, and evaluation metric"
+                    info="Choose model variant, dataset, and evaluation metric",
+                    allow_custom_value=False,
+                    filterable=True,
+                    max_choices=None
                 )
         
         with gr.Column(scale=1):
@@ -687,9 +747,9 @@ with gr.Blocks(css=css, theme=gr.themes.Soft(), title="ZIP Crowd Counting") as d
         if "━━━━━━" in variant_dataset_metric:
             return "⚠️ Please select a valid model configuration"
         result = update_model_if_needed(variant_dataset_metric)
-        if "Model loaded:" in result:
+        if "Model configuration set:" in result:
             return f"✅ {result}"
-        elif "Model already loaded:" in result:
+        elif "Model configuration:" in result:
             return f"🔄 {result}"
         else:
             return f"❌ {result}"