重写了数据集

spikingjelly/datasets/__init__.py

@@ -34,7 +34,7 @@ def integrate_events_to_frames(events, height, width, frames_num=10, split_by='t
 
     '''
     frames = np.zeros(shape=[frames_num, 2, height, width])
-
+    eps = 1e-5  # 涉及到除法的地方，被除数加上eps，防止出现除以0
     if split_by == 'time':
         # 按照脉冲的发生时刻进行等分
         events['t'] -= events['t'][0]
@@ -57,9 +57,9 @@ def integrate_events_to_frames(events, height, width, frames_num=10, split_by='t
             if normalization == 'frequency':
                 frames[i] /= dt  # 表示脉冲发放的频率
             elif normalization == 'max':
-                frames[i] /= frames[i].max()
+                frames[i] /= max(frames[i].max(), eps)
             elif normalization == 'norm':
-                frames[i] = (frames[i] - frames[i].mean()) / np.sqrt((frames[i].var() + 1e-5))
+                frames[i] = (frames[i] - frames[i].mean()) / np.sqrt(max(frames[i].var(), eps))
             elif normalization is None:
                 continue
             else:
@@ -80,10 +80,10 @@ def integrate_events_to_frames(events, height, width, frames_num=10, split_by='t
             if normalization == 'frequency':
                 frames[i] /= dt  # 表示脉冲发放的频率
             elif normalization == 'max':
-                frames[i] /= max(frames[i].max(), 1e-5)
+                frames[i] /= max(frames[i].max(), eps)
 
             elif normalization == 'norm':
-                frames[i] = (frames[i] - frames[i].mean()) / np.sqrt((frames[i].var() + 1e-5))
+                frames[i] = (frames[i] - frames[i].mean()) / np.sqrt(max(frames[i].var(), eps))
             elif normalization is None:
                 continue
             else:
@@ -220,56 +220,4 @@ def create_frames_dataset(events_data_dir: str, frames_data_dir: str, frames_num
         将 ``events_data_dir`` 文件夹下的脉冲数据全部转换成帧数据，并保存在 ``frames_data_dir``。
         转换参数的详细含义，参见 ``integrate_events_to_frames`` 函数。
         '''
-        raise NotImplementedError
-
-class SubDirDataset(Dataset):
-    def __init__(self, root, train=True, split_ratio=0.9):
-        '''
-        :param root: 保存数据集的文件夹
-        :param train: 训练还是测试
-        :param split_ratio: 训练集占数据集的比例，对于每一类的数据，按文件夹内部数据文件的命名排序，取前 ``split_ratio`` 的数据作为训练集，
-                            其余数据作为测试集
-
-        适用于包含多个子文件夹，每个子文件夹名称为类别名，子文件夹内部是npz格式的数据的数据集基类。文件结构类似如下所示：
-
-        .. code-block:: bash
-
-            dvs_cifar10_npz/
-            |-- airplane
-            |   |-- 0.npz
-            |   |-- ...
-            |-- automobile
-            |-- bird
-            |-- cat
-            |-- deer
-            |-- dog
-            |-- frog
-            |-- horse
-            |-- ship
-            `-- truck
-        '''
-
-        self.root = root
-        self.label_name = os.listdir(self.root)
-        self.file_path = []
-        self.label = []
-
-        for i in range(self.label_name.__len__()):
-            sub_dir_path = os.path.join(self.root, self.label_name[i])
-            file_names = os.listdir(sub_dir_path)
-            split_boundary = int(file_names.__len__() * split_ratio)
-            if train:
-                for j in range(0, split_boundary):
-                    self.file_path.append(os.path.join(sub_dir_path, file_names[j]))
-                    self.label.append(i)
-            else:
-                for j in range(split_boundary, file_names.__len__()):
-                    self.file_path.append(os.path.join(sub_dir_path, file_names[j]))
-                    self.label.append(i)
-
-    def __len__(self):
-        return self.file_path.__len__()
-
-    def __getitem__(self, index):
-        frame = torch.from_numpy(np.load(self.file_path[index])['arr_0']).float()
-        return frame, self.label[index]
+        raise NotImplementedError