diff --git a/plugins/ai_shoe_det/ai_shoe_det_node.cpp b/plugins/ai_shoe_det/ai_shoe_det_node.cpp
index f432db9..d6cdc4e 100644
--- a/plugins/ai_shoe_det/ai_shoe_det_node.cpp
+++ b/plugins/ai_shoe_det/ai_shoe_det_node.cpp
@@ -323,10 +323,17 @@ private:
         std::vector<DetBox> dets;
         if (outputs.empty() || !outputs[0].data) return dets;
 
+        if (outputs[0].type != RKNN_TENSOR_FLOAT32 || outputs[0].size < sizeof(float) * 5) {
+            LogWarn("[ai_shoe_det] unsupported output tensor type or size");
+            return dets;
+        }
+
         // 假设 YOLOv8 输出格式: [num_boxes, 5] = x, y, w, h, conf
         // 实际格式可能不同，需要根据模型调整
         const float* data = reinterpret_cast<const float*>(outputs[0].data);
-        int num_boxes = 8400;  // YOLOv8 默认
+        const size_t value_count = outputs[0].size / sizeof(float);
+        const int num_boxes = static_cast<int>(std::min<size_t>(8400, value_count / 5));
+        if (num_boxes <= 0) return dets;
 
         float scale_x = static_cast<float>(win_w) / model_w_;
         float scale_y = static_cast<float>(win_h) / model_h_;