增加训练阶段的数据增强; 将输入数据标准化; 删除main中的config不需要的量并修改输入维度;

data_module.py

@@ -22,11 +22,11 @@ class DataModule(pl.LightningDataModule):
         k_fold_dataset_list = self.get_k_fold_dataset_list()
         if stage == 'fit' or stage is None:
             dataset_train, dataset_val = self.get_fit_dataset_lists(k_fold_dataset_list)
-            self.train_dataset = CustomDataset(self.dataset_path, dataset_train, self.config, 'train')
-            self.val_dataset = CustomDataset(self.dataset_path, dataset_val, self.config, 'train')
+            self.train_dataset = CustomDataset(self.dataset_path, dataset_train,  'train', self.config,)
+            self.val_dataset = CustomDataset(self.dataset_path, dataset_val,  'val', self.config,)
         if stage == 'test' or stage is None:
             dataset_test = self.get_test_dataset_lists(k_fold_dataset_list)
-            self.test_dataset = CustomDataset(self.dataset_path, dataset_test, self.config, 'test')
+            self.test_dataset = CustomDataset(self.dataset_path, dataset_test, 'test', self.config,)
 
     def get_k_fold_dataset_list(self):
         # 得到用于K折分割的数据的list, 并生成文件夹进行保存
@@ -72,13 +72,31 @@ class DataModule(pl.LightningDataModule):
 
 
 class CustomDataset(Dataset):
-    def __init__(self, dataset_path, dataset, config, type):
+    def __init__(self, dataset_path, dataset, stage, config, ):
         super().__init__()
         self.dataset = dataset
-        self.trans = transforms.ToTensor()
-        self.labels = open(dataset_path + '/' + type + '/label.txt').readlines()
+        # 此处的均值和方差来源于ImageNet
+        normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                         std=[0.229, 0.224, 0.225])
+        if stage == 'train':
+            self.trans = transforms.Compose([
+                transforms.RandomHorizontalFlip(),
+                transforms.RandomCrop(config['dim_in'], 4),
+                transforms.ToTensor(),
+                normalize, ])
+        elif stage == 'val':
+            stage = 'train'
+            self.trans = transforms.Compose([
+                transforms.ToTensor(),
+                normalize, ])
+        else:
+            self.trans = transforms.Compose([
+                transforms.ToTensor(),
+                normalize, ])
+        self.labels = open(dataset_path + '/' + stage + '/label.txt').readlines()
 
     def __getitem__(self, idx):
+        # 注意: 为了满足初始化权重算法的要求, 需要输入参数的均值为0. 可以使用transforms.Normalize()
         image_path = self.dataset[idx]
         image_name = os.path.basename(image_path)
         image = Image.open(image_path)

main.py

@@ -59,12 +59,7 @@ def main(stage,
     # TODO 获得最优的batch size
     num_workers = cpu_count()
     # 获得非通用参数
-    config = {'dim_in': 5,
-              'dim': 10,
-              'res_coef': 0.5,
-              'dropout_p': 0.1,
-              'n_layers': 3,
-              'dataset_len': 100000}
+    config = {'dim_in': 32, }
     for kth_fold in range(kth_fold_start, k_fold):
         load_checkpoint_path = get_ckpt_path(version_nth, kth_fold)
         logger = pl_loggers.TensorBoardLogger('logs/')