增加角度显示计算，优化CNN架构

data_units.py

@@ -19,10 +19,10 @@ def layout_to_tensor(layout_path, target_size=(256, 256)):
     img = img.resize(target_size, resample=Image.BILINEAR)
     return np.array(img) / 255.0                 # 归一化到[0,1]
 
-def tile_layout(large_layout, block_size=64):
+def tile_layout(large_layout, block_size=64, overlap_ratio=0.5):
     """将大版图分割为小块（滑动窗口方式）"""
     height, width = large_layout.shape
-    stride = block_size // 2                     # 步长设置重叠区域
+    stride = int(block_size * (1 - overlap_ratio))                    # 步长设置重叠区域
     tiles = []
     for y in range(0, height - block_size +1, stride):
         for x in range(0, width - block_size +1, stride):

inference.py

@@ -1,15 +1,8 @@
 import faiss
 import numpy as np
 import torch
-# 导入 models.rotation_cnn 模块中的 RotationInvariantNet 类
-from models.rotation_cnn import RotationInvariantNet
-
-from models.rotation_cnn import get_rotational_features
-
-# 导入 data_utils 中的 layout_to_tensor 函数（假设该函数存在）
-from data_units import layout_to_tensor  # 如果 data_utils.py 存在此函数
-
-from data_units import tile_layout
+from models.rotation_cnn import RotationInvariantNet, get_rotational_features
+from data_units import layout_to_tensor, tile_layout
 
 def main():
     # 配置参数（需根据实际调整）
@@ -17,7 +10,6 @@ def main():
     target_module_path = "target.png"
     large_layout_path = "layout_large.png"
 
-    # 加载模型
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     model = RotationInvariantNet().to(device)
     model.load_state_dict(torch.load("rotation_cnn.pth"))
@@ -33,16 +25,27 @@ def main():
     # 构建特征索引（使用Faiss加速）
     index = faiss.IndexFlatL2(64)  # 特征维度由模型决定
     features_db = []
-    for (x,y,tile) in tiles:
+    for (x, y, tile) in tiles:
         feat = get_rotational_features(model, torch.tensor(tile).to(device))
         features_db.append(feat)
     index.add(np.stack(features_db))
 
     # 检索相似区域
-    D, I = index.search(target_feat[np.newaxis,:], k=10)
-    for idx in I[0]:
-        x,y,_ = tiles[idx]
-        print(f"匹配区域坐标: ({x}, {y}), 相似度: {D[0][idx]}")
+    D, I = index.search(target_feat[np.newaxis, :], k=10)
 
+    for idx in I[0]:
+        x, y, _ = tiles[idx]
+
+        # 计算最佳匹配角度的显式计算
+        min_angle, min_dist = 90, float('inf')
+        target_vec = target_feat
+        feat = features_db[idx]
+        for a in [0, 1, 2, 3]:  # 代表0°、90°、180°、270°
+            rotated_feat = np.rot90(feat.reshape(block_size, block_size), k=a)
+            dist = np.linalg.norm(target_vec - rotated_feat.flatten())
+            if dist < min_dist:
+                min_dist, min_angle = dist, a * 90
+
+        print(f"坐标({x},{y}), 最佳旋转方向{min_angle}度，距离: {min_dist}")
 if __name__ == "__main__":
     main()

models/rotation_cnn.py

@@ -3,7 +3,7 @@ import torch.nn as nn
 
 class RotationInvariantNet(nn.Module):
     """轻量级旋转不变特征提取网络"""
-    def __init__(self, input_channels=1, num_features=64):
+    def __init__(self, input_channels=1):
         super().__init__()
         self.cnn = nn.Sequential(
             # 基础卷积层
@@ -12,13 +12,14 @@ class RotationInvariantNet(nn.Module):
             nn.MaxPool2d(2),                     # 下采样
             nn.Conv2d(32, 64, kernel_size=3, padding=1),
             nn.ReLU(),
-            nn.AdaptiveAvgPool2d((1,1))          # 全局池化获取全局特征
+            nn.Conv2d(64, 64, kernel_size=3, stride=2),  # 更大感受野
+            nn.AdaptiveAvgPool2d((4,4)),         # 全局池化获取全局特征，调整输出尺寸为4x4
+            nn.Flatten(),                        # 展平为一维向量
+            nn.Linear(64*16, 128)                # 增加全连接层以降低维度到128
         )
-        
+
     def forward(self, x):
-        features = self.cnn(x)
-        return torch.flatten(features, 1)         # 展平为特征向量
-    
+        return self.cnn(x)
 def get_rotational_features(model, input_image):
     """计算输入图像所有旋转角度的特征平均值"""
     rotations = [0, 90, 180, 270]