diff --git a/data/data.py b/data/data.py
new file mode 100644
index 0000000..78153ad
--- /dev/null
+++ b/data/data.py
@@ -0,0 +1,68 @@
+import os
+import torch
+from torch.utils.data import Dataset
+import numpy as np
+
+class BRepSDFDataset(Dataset):
+    def __init__(self, brep_data_dir:str, sdf_data_dir:str, split:str='train'):
+        self.brep_data_dir = brep_data_dir
+        self.sdf_data_dir = sdf_data_dir
+        self.split = split
+        self.brep_data_list = self._load_brep_data_list()
+        self.sdf_data_list = self._load_sdf_data_list()
+
+    def _load_brep_data_list(self):
+        data_list = []
+        split_dir = os.path.join(self.brep_data_dir, self.split)
+        for sample_dir in os.listdir(split_dir):
+            sample_path = os.path.join(split_dir, sample_dir)
+            if os.path.isdir(sample_path):
+                data_list.append(sample_path)
+        return data_list
+
+    def _load_sdf_data_list(self):
+        data_list = []
+        split_dir = os.path.join(self.sdf_data_dir, self.split)
+        for sample_dir in os.listdir(split_dir):
+            sample_path = os.path.join(split_dir, sample_dir)
+            if os.path.isdir(sample_path):
+                data_list.append(sample_path)
+        return data_list
+
+    def __len__(self):
+        return len(self.brep_data_list)
+
+    def __getitem__(self, idx):
+        sample_path = self.brep_data_list[idx]
+        
+        # 解析 .step 文件
+        step_file = os.path.join(sample_path, 'model.step')
+        brep_features = self._parse_step_file(step_file)
+        
+        # 加载 .sdf 文件
+        sdf_file = os.path.join(sample_path, 'sdf.npy')
+        sdf = np.load(sdf_file)
+        
+        # 转换为 torch 张量
+        brep_features = torch.tensor(brep_features, dtype=torch.float32)
+        sdf = torch.tensor(sdf, dtype=torch.float32)
+        
+        return {
+            'brep_features': brep_features,
+            'sdf': sdf
+        }
+
+    def _parse_step_file(self, step_file):
+        # 解析 .step 文件的逻辑
+        # 返回 B-rep 特征
+        pass
+
+# 使用示例
+if __name__ == '__main__':
+    dataset = BRepSDFDataset(brep_data_dir='path/to/brep_data', sdf_data_dir='path/to/sdf_data', split='train')
+    dataloader = torch.utils.data.DataLoader(dataset, batch_size=4, shuffle=True)
+
+    for batch in dataloader:
+        print(batch['brep_features'].shape)
+        print(batch['sdf'].shape)
+        break
\ No newline at end of file