Load and parse metadata for CIFAR-10, CIFAR-100 (pytorch#502)

* cifar10.meta['label_names']

* cifar100.meta['fine_label_names']

* cifar100.meta['coarse_label_names']

Author: xenosoz

torchvision/datasets/cifar.py

@@ -46,6 +46,10 @@ class CIFAR10(data.Dataset):
         ['test_batch', '40351d587109b95175f43aff81a1287e'],
     ]
 
+    meta_list = [
+        ['batches.meta', '5ff9c542aee3614f3951f8cda6e48888'],
+    ]
+
     def __init__(self, root, train=True,
                  transform=None, target_transform=None,
                  download=False):
@@ -100,6 +104,16 @@ def __init__(self, root, train=True,
             self.test_data = self.test_data.reshape((10000, 3, 32, 32))
             self.test_data = self.test_data.transpose((0, 2, 3, 1))  # convert to HWC
 
+        f = self.meta_list[0][0]
+        file = os.path.join(self.root, self.base_folder, f)
+        fo = open(file, 'rb')
+        if sys.version_info[0] == 2:
+            entry = pickle.load(fo)
+        else:
+            entry = pickle.load(fo, encoding='latin1')
+        fo.close()
+        self.meta = entry
+
     def __getitem__(self, index):
         """
         Args:
@@ -133,7 +147,7 @@ def __len__(self):
 
     def _check_integrity(self):
         root = self.root
-        for fentry in (self.train_list + self.test_list):
+        for fentry in (self.train_list + self.test_list + self.meta_list):
             filename, md5 = fentry[0], fentry[1]
             fpath = os.path.join(root, self.base_folder, filename)
             if not check_integrity(fpath, md5):
@@ -187,3 +201,7 @@ class CIFAR100(CIFAR10):
     test_list = [
         ['test', 'f0ef6b0ae62326f3e7ffdfab6717acfc'],
     ]
+
+    meta_list = [
+        ['meta', '7973b15100ade9c7d40fb424638fde48'],
+    ]