From 53e0ffde95fed4f0e6afa8845f2524a6fa057530 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Mon, 26 Oct 2020 02:41:18 -0400
Subject: [PATCH 01/16] add vggface dataset class

---
 torchvision/datasets/__init__.py |  3 +-
 torchvision/datasets/vggface2.py | 71 ++++++++++++++++++++++++++++++++
 2 files changed, 73 insertions(+), 1 deletion(-)
 create mode 100644 torchvision/datasets/vggface2.py

diff --git a/torchvision/datasets/__init__.py b/torchvision/datasets/__init__.py
index 4531d937c68..ae429e098ac 100644
--- a/torchvision/datasets/__init__.py
+++ b/torchvision/datasets/__init__.py
@@ -18,6 +18,7 @@
 from .celeba import CelebA
 from .sbd import SBDataset
 from .vision import VisionDataset
+from .vggface2 import VGGFace2
 from .usps import USPS
 from .kinetics import Kinetics400
 from .hmdb51 import HMDB51
@@ -32,4 +33,4 @@
            'Omniglot', 'SBU', 'Flickr8k', 'Flickr30k',
            'VOCSegmentation', 'VOCDetection', 'Cityscapes', 'ImageNet',
            'Caltech101', 'Caltech256', 'CelebA', 'SBDataset', 'VisionDataset',
-           'USPS', 'Kinetics400', 'HMDB51', 'UCF101', 'Places365')
+           'VGGFace2', 'USPS', 'Kinetics400', 'HMDB51', 'UCF101', 'Places365')
diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
new file mode 100644
index 00000000000..b8d17d3003b
--- /dev/null
+++ b/torchvision/datasets/vggface2.py
@@ -0,0 +1,71 @@
+import os
+from PIL import Image
+import torch
+from torchvision.datasets import VisionDataset
+
+class VGGFace2(VisionDataset):
+
+    def __init__(self, root, split="train", transform=None, target_transform=None):
+        """
+        Args:
+            root (string): Root directory of the VGGFace2 Dataset.
+            Expects the following folder structure if download=False:
+                .
+                └── vggface2
+                    ├── vggface2_train.tar.gz (or 'train' if uncompressed)
+                    ├── vggface2_test.tar.gz (or 'test' if uncompressed)
+                    ├── train_list.txt
+                    └── test_list.txt
+        split (string): One of {``train``, ``test``}.
+            The dataset split to use. Defaults to ``train``.
+        target_type (string): The type of target to use, can be one of {``identity``, ``bbox``, ``attr``.``""``}
+            Can also be a list to output a tuple with all specified target types.
+            The targets represent:
+                ``raw`` (torch.tensor shape=(10,) dtype=int): all annotations combined (bbox + attr)
+                ``bbox`` (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
+                ``attr`` (torch.tensor shape=(6,) dtype=int): label values for attributes
+                    that represent (blur, expression, illumination, occlusion, pose, invalid)
+            Defaults to ``raw``. If empty, ``None`` will be returned as target.
+        transform (callable, optional): A function/transform that  takes in a PIL image
+            and returns a transformed version. E.g, ``transforms.RandomCrop``
+        target_transform (callable, optional): A function/transform that takes in the
+            target and transforms it.
+        """
+        super(VGGFace2, self).__init__(root, transform=transform, target_transform=target_transform)
+        # check arguments
+        if split not in ('train','test'):
+            raise ValueError('split \"{}\" is not recognized.'.format(split))
+        self.split = split
+        self.img_info = []
+
+        image_list_file = 'train_list.txt' if self.split=='train' else 'test_list.txt'
+        self.image_list_file = os.path.join(self.root, image_list_file)
+
+        with open(self.image_list_file, 'r') as f:
+            for i, img_file in enumerate(f):
+                img_file = img_file.strip()  # e.g. train/n004332/0317_01.jpg
+                class_id = img_file.split("/")[0]  # like n004332
+                img_file = os.path.join(self.root, self.split, img_file)
+                self.img_info.append({
+                    'img_path': img_file,
+                    'class_id': class_id,
+                })
+                if i % 1000 == 0:
+                    print("processing: {} images for {}".format(i, self.split))
+
+    def __len__(self):
+        return len(self.img_info)
+
+    def __getitem__(self, index):
+        img_info = self.img_info[index]
+        img = Image.open(img_info['img_path'])
+        if self.transform:
+            img = self.transform(img)
+
+        target = None
+        if self.split == "test":
+            return img, target
+
+        if self.target_transform is not None:
+            target = self.target_transform(target)
+        return img, target

From 5e75f3c34a397fabf72ab86accf5e9e1615536c5 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Mon, 26 Oct 2020 03:04:35 -0400
Subject: [PATCH 02/16] fix flake8 errors

---
 torchvision/datasets/vggface2.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index b8d17d3003b..c05c433798f 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -3,6 +3,7 @@
 import torch
 from torchvision.datasets import VisionDataset
 
+
 class VGGFace2(VisionDataset):
 
     def __init__(self, root, split="train", transform=None, target_transform=None):
@@ -33,12 +34,12 @@ def __init__(self, root, split="train", transform=None, target_transform=None):
         """
         super(VGGFace2, self).__init__(root, transform=transform, target_transform=target_transform)
         # check arguments
-        if split not in ('train','test'):
+        if split not in ('train', 'test'):
             raise ValueError('split \"{}\" is not recognized.'.format(split))
         self.split = split
         self.img_info = []
 
-        image_list_file = 'train_list.txt' if self.split=='train' else 'test_list.txt'
+        image_list_file = 'train_list.txt' if self.split == 'train' else 'test_list.txt'
         self.image_list_file = os.path.join(self.root, image_list_file)
 
         with open(self.image_list_file, 'r') as f:

From dc365803d3cda9c366c26db04b9dc1f2c6ff5382 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Mon, 26 Oct 2020 18:20:51 -0400
Subject: [PATCH 03/16] add standard dataset arguments

---
 torchvision/datasets/vggface2.py | 78 ++++++++++++++++++++++----------
 1 file changed, 55 insertions(+), 23 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index c05c433798f..620dbaa5fc0 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -1,13 +1,12 @@
-import os
 from PIL import Image
+import os
 import torch
-from torchvision.datasets import VisionDataset
-
+from .vision import VisionDataset
+from .utils import check_integrity, extract_archive
 
 class VGGFace2(VisionDataset):
+    """ VGGFace2 <http://zeus.robots.ox.ac.uk/vgg_face2/>`_ Dataset.
 
-    def __init__(self, root, split="train", transform=None, target_transform=None):
-        """
         Args:
             root (string): Root directory of the VGGFace2 Dataset.
             Expects the following folder structure if download=False:
@@ -17,22 +16,46 @@ def __init__(self, root, split="train", transform=None, target_transform=None):
                     ├── vggface2_test.tar.gz (or 'test' if uncompressed)
                     ├── train_list.txt
                     └── test_list.txt
-        split (string): One of {``train``, ``test``}.
-            The dataset split to use. Defaults to ``train``.
-        target_type (string): The type of target to use, can be one of {``identity``, ``bbox``, ``attr``.``""``}
-            Can also be a list to output a tuple with all specified target types.
-            The targets represent:
-                ``raw`` (torch.tensor shape=(10,) dtype=int): all annotations combined (bbox + attr)
-                ``bbox`` (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
-                ``attr`` (torch.tensor shape=(6,) dtype=int): label values for attributes
-                    that represent (blur, expression, illumination, occlusion, pose, invalid)
-            Defaults to ``raw``. If empty, ``None`` will be returned as target.
-        transform (callable, optional): A function/transform that  takes in a PIL image
-            and returns a transformed version. E.g, ``transforms.RandomCrop``
-        target_transform (callable, optional): A function/transform that takes in the
-            target and transforms it.
+            split (string): One of {``train``, ``test``}.
+                The dataset split to use. Defaults to ``train``.
+            target_type (string): The type of target to use, can be one of {``identity``, ``bbox``, ``attr``.``""``}
+                Can also be a list to output a tuple with all specified target types.
+                The targets represent:
+                    ``raw`` (torch.tensor shape=(10,) dtype=int): all annotations combined (bbox + attr)
+                    ``bbox`` (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
+                    ``attr`` (torch.tensor shape=(6,) dtype=int): label values for attributes
+                        that represent (blur, expression, illumination, occlusion, pose, invalid)
+                Defaults to ``raw``. If empty, ``None`` will be returned as target.
+            transform (callable, optional): A function/transform that  takes in a PIL image
+                and returns a transformed version. E.g, ``transforms.RandomCrop``
+            target_transform (callable, optional): A function/transform that takes in the
+                target and transforms it.
         """
+    
+    base_folder = "vggface2"
+    file_list = [
+        # Filename                MD5 Hash                            Uncompressed filename
+        ("vggface2_train.tar.gz", "88813c6b15de58afc8fa75ea83361d7f", "train"),
+        ("vggface2_test.tar.gz",  "bb7a323824d1004e14e00c23974facd3", "test"),
+    ]
+
+    def __init__(
+        self,
+        root,
+        split = "train",
+        download = False,
+        transform = None,
+        target_transform = None):
+
+        root = os.path.join(root, self.base_folder)
         super(VGGFace2, self).__init__(root, transform=transform, target_transform=target_transform)
+
+        if download:
+            msg = ("The dataset is not publicly accessible. You must login and "
+                   "download the archives externally and place them in the root "
+                   "directory.")
+            raise RuntimeError(msg)
+
         # check arguments
         if split not in ('train', 'test'):
             raise ValueError('split \"{}\" is not recognized.'.format(split))
@@ -42,17 +65,22 @@ def __init__(self, root, split="train", transform=None, target_transform=None):
         image_list_file = 'train_list.txt' if self.split == 'train' else 'test_list.txt'
         self.image_list_file = os.path.join(self.root, image_list_file)
 
+        # are files downloaded and extracted?
+        for (filename, md5, extracted_folder) in self.file_list:
+            extracted_folder_path = os.path.join(self.root, extracted_folder)
+            if not os.path.isdir(extracted_folder_path):
+                raise RuntimeError('Can not find folder \"{}\"'.format(extracted_folder_path))
+
         with open(self.image_list_file, 'r') as f:
             for i, img_file in enumerate(f):
-                img_file = img_file.strip()  # e.g. train/n004332/0317_01.jpg
+                img_file = img_file.strip()        # e.g. n004332/0317_01.jpg
                 class_id = img_file.split("/")[0]  # like n004332
-                img_file = os.path.join(self.root, self.split, img_file)
+                img_file = os.path.join(self.root, self.split, img_file) # like root/vggface2/train/n000002/0001_01.jpg
+                print("img_file: " + img_file)
                 self.img_info.append({
                     'img_path': img_file,
                     'class_id': class_id,
                 })
-                if i % 1000 == 0:
-                    print("processing: {} images for {}".format(i, self.split))
 
     def __len__(self):
         return len(self.img_info)
@@ -70,3 +98,7 @@ def __getitem__(self, index):
         if self.target_transform is not None:
             target = self.target_transform(target)
         return img, target
+
+    def extra_repr(self) -> str:
+        lines = ["Split: {split}"]
+        return '\n'.join(lines).format(**self.__dict__)

From ab31f883eeec2ba2f5ca03739688141f48953381 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Tue, 27 Oct 2020 00:23:16 -0400
Subject: [PATCH 04/16] fix code formatting and standardize dataset class

---
 torchvision/datasets/vggface2.py | 117 +++++++++++++++++++++----------
 1 file changed, 81 insertions(+), 36 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 620dbaa5fc0..62d613363ec 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -1,8 +1,11 @@
+from functools import partial
 from PIL import Image
 import os
 import torch
-from .vision import VisionDataset
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 from .utils import check_integrity, extract_archive
+from .vision import VisionDataset
+
 
 class VGGFace2(VisionDataset):
     """ VGGFace2 <http://zeus.robots.ox.ac.uk/vgg_face2/>`_ Dataset.
@@ -15,41 +18,50 @@ class VGGFace2(VisionDataset):
                     ├── vggface2_train.tar.gz (or 'train' if uncompressed)
                     ├── vggface2_test.tar.gz (or 'test' if uncompressed)
                     ├── train_list.txt
-                    └── test_list.txt
+                    ├── test_list.txt
+                    └── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
             split (string): One of {``train``, ``test``}.
                 The dataset split to use. Defaults to ``train``.
-            target_type (string): The type of target to use, can be one of {``identity``, ``bbox``, ``attr``.``""``}
+            target_type (string): The type of target to use. One of
+                {``class_id``, ``image_id``, ``face_id``, ``bbox``, ``landmarks``.``""``}
                 Can also be a list to output a tuple with all specified target types.
                 The targets represent:
-                    ``raw`` (torch.tensor shape=(10,) dtype=int): all annotations combined (bbox + attr)
+                    ``class_id`` (string)
+                    ``image_id`` (string)
+                    ``face_id`` (string)
                     ``bbox`` (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
-                    ``attr`` (torch.tensor shape=(6,) dtype=int): label values for attributes
-                        that represent (blur, expression, illumination, occlusion, pose, invalid)
-                Defaults to ``raw``. If empty, ``None`` will be returned as target.
+                    ``landmarks`` (torch.tensor shape=(10,) dtype=float): values that
+                        represent five points (P1X, P1Y, P2X, P2Y, P3X, P3Y, P4X, P4Y, P5X, P5Y)
+                Defaults to ``bbox``. If empty, ``None`` will be returned as target.
             transform (callable, optional): A function/transform that  takes in a PIL image
                 and returns a transformed version. E.g, ``transforms.RandomCrop``
             target_transform (callable, optional): A function/transform that takes in the
                 target and transforms it.
         """
-    
+
     base_folder = "vggface2"
     file_list = [
         # Filename                MD5 Hash                            Uncompressed filename
         ("vggface2_train.tar.gz", "88813c6b15de58afc8fa75ea83361d7f", "train"),
-        ("vggface2_test.tar.gz",  "bb7a323824d1004e14e00c23974facd3", "test"),
+        ("vggface2_test.tar.gz", "bb7a323824d1004e14e00c23974facd3", "test"),
+        ("bb_landmark.tar.gz", "26f7ba288a782862d137348a1cb97540", "bb_landmark")
     ]
 
     def __init__(
         self,
-        root,
-        split = "train",
-        download = False,
-        transform = None,
-        target_transform = None):
-
-        root = os.path.join(root, self.base_folder)
-        super(VGGFace2, self).__init__(root, transform=transform, target_transform=target_transform)
+        root: str,
+        split: str = "train",
+        target_type: Union[List[str], str] = "bbox",
+        transform: Optional[Callable] = None,
+        target_transform: Optional[Callable] = None,
+        download: bool = False,
+    ) -> None:
+        import pandas
+        super(VGGFace2, self).__init__(root=os.path.join(root, self.base_folder),
+                                       transform=transform,
+                                       target_transform=target_transform)
 
+        # stay consistent with other datasets and check for a download option
         if download:
             msg = ("The dataset is not publicly accessible. You must login and "
                    "download the archives externally and place them in the root "
@@ -60,45 +72,78 @@ def __init__(
         if split not in ('train', 'test'):
             raise ValueError('split \"{}\" is not recognized.'.format(split))
         self.split = split
-        self.img_info = []
+        self.img_info: List[Dict[str, object]] = []
+
+        if isinstance(target_type, list):
+            self.target_type = target_type
+        else:
+            self.target_type = [target_type]
+
+        if not (all(x in ["class_id", "image_id", "face_id", "bbox", "landmarks", ""] for x in self.target_type)):
+            raise ValueError("target_type \"{}\" is not recognized.".format(self.target_type))
+        if not self.target_type and self.target_transform is not None:
+            raise RuntimeError('target_transform is specified but target_type is empty')
 
         image_list_file = 'train_list.txt' if self.split == 'train' else 'test_list.txt'
         self.image_list_file = os.path.join(self.root, image_list_file)
 
-        # are files downloaded and extracted?
-        for (filename, md5, extracted_folder) in self.file_list:
-            extracted_folder_path = os.path.join(self.root, extracted_folder)
-            if not os.path.isdir(extracted_folder_path):
-                raise RuntimeError('Can not find folder \"{}\"'.format(extracted_folder_path))
+        # prepare dataset
+        for (filename, _, extracted_dir) in self.file_list:
+            filename = os.path.join(self.root, filename)
+            extracted_dir_path = os.path.join(self.root, extracted_dir)
+            if not os.path.isdir(extracted_dir_path):
+                extract_archive(filename)
+
+        # process dataset
+        fn = partial(os.path.join, self.root, self.file_list[2][2])
+        bbox_frames = [pandas.read_csv(fn("loose_bb_train.csv"), index_col=0),
+                       pandas.read_csv(fn("loose_bb_test.csv"), index_col=0)]
+        self.bbox = pandas.concat(bbox_frames)
+        landmark_frames = [pandas.read_csv(fn("loose_landmark_train.csv"), index_col=0),
+                           pandas.read_csv(fn("loose_landmark_test.csv"), index_col=0)]
+        self.landmarks = pandas.concat(landmark_frames)
 
         with open(self.image_list_file, 'r') as f:
             for i, img_file in enumerate(f):
-                img_file = img_file.strip()        # e.g. n004332/0317_01.jpg
-                class_id = img_file.split("/")[0]  # like n004332
-                img_file = os.path.join(self.root, self.split, img_file) # like root/vggface2/train/n000002/0001_01.jpg
-                print("img_file: " + img_file)
+                img_file = img_file.strip()
+                img_filename, ext = os.path.splitext(img_file)  # e.g. ["n004332/0317_01", "jpg"]
+                class_id, image_face_id = img_filename.split("/")
+                image_id, face_id = image_face_id.split("_")
+                img_filepath = os.path.join(self.root, self.split, img_file)
                 self.img_info.append({
-                    'img_path': img_file,
+                    'img_path': img_filepath,
                     'class_id': class_id,
+                    'image_id': image_id,
+                    'face_id': face_id,
+                    'bbox': torch.tensor(self.bbox.loc[img_filename].values),
+                    'landmarks': torch.tensor(self.landmarks.loc[img_filename].values),
                 })
 
-    def __len__(self):
+    def __len__(self) -> int:
         return len(self.img_info)
 
-    def __getitem__(self, index):
+    def __getitem__(self, index) -> Tuple[Any, Any]:
         img_info = self.img_info[index]
+
+        # prepare image
         img = Image.open(img_info['img_path'])
         if self.transform:
             img = self.transform(img)
 
-        target = None
-        if self.split == "test":
-            return img, target
+        # prepare target
+        target: Any = []
+        for t in self.target_type:
+            if t == "":
+                target = None
+                break
+            target.append(img_info[t])
+        if target:
+            target = tuple(target) if len(target) > 1 else target[0]
+            if self.target_transform is not None:
+                target = self.target_transform(target)
 
-        if self.target_transform is not None:
-            target = self.target_transform(target)
         return img, target
 
     def extra_repr(self) -> str:
-        lines = ["Split: {split}"]
+        lines = ["Target type: {target_type}", "Split: {split}"]
         return '\n'.join(lines).format(**self.__dict__)

From 9d3590db7f1b44e574ac9ad75c11ee913463a18b Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Tue, 27 Oct 2020 13:05:58 -0400
Subject: [PATCH 05/16] add dataset citation

---
 torchvision/datasets/vggface2.py | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 62d613363ec..495b6f9de02 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -10,6 +10,13 @@
 class VGGFace2(VisionDataset):
     """ VGGFace2 <http://zeus.robots.ox.ac.uk/vgg_face2/>`_ Dataset.
 
+        Citation:
+        @inproceedings{Cao18,
+            author    = "Cao, Q. and Shen, L. and Xie, W. and Parkhi, O. M. and Zisserman, A.",
+            title     = "VGGFace2: A dataset for recognising faces across pose and age",
+            booktitle = "International Conference on Automatic Face and Gesture Recognition",
+            year      = "2018"}
+
         Args:
             root (string): Root directory of the VGGFace2 Dataset.
             Expects the following folder structure if download=False:
@@ -37,6 +44,9 @@ class VGGFace2(VisionDataset):
                 and returns a transformed version. E.g, ``transforms.RandomCrop``
             target_transform (callable, optional): A function/transform that takes in the
                 target and transforms it.
+            download (bool, optional): If true, downloads the dataset from the internet and
+                puts it in root directory. If dataset is already downloaded, it is not
+                downloaded again.
         """
 
     base_folder = "vggface2"

From 0ea263a9e5b64a51b11158a6e6aa89c25c31e91b Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Tue, 27 Oct 2020 13:08:55 -0400
Subject: [PATCH 06/16] more formatting fixes

---
 torchvision/datasets/vggface2.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 495b6f9de02..500740e5fcb 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -27,8 +27,8 @@ class VGGFace2(VisionDataset):
                     ├── train_list.txt
                     ├── test_list.txt
                     └── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
-            split (string): One of {``train``, ``test``}.
-                The dataset split to use. Defaults to ``train``.
+            split (string): The dataset split to use. One of {``train``, ``test``}.
+                Defaults to ``train``.
             target_type (string): The type of target to use. One of
                 {``class_id``, ``image_id``, ``face_id``, ``bbox``, ``landmarks``.``""``}
                 Can also be a list to output a tuple with all specified target types.

From 7bb168eed65899607d0e2848aa9d8fb7d0d68abb Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Fri, 30 Oct 2020 01:13:00 -0400
Subject: [PATCH 07/16] docstring update

---
 torchvision/datasets/vggface2.py | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 500740e5fcb..813c8917da2 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -22,21 +22,21 @@ class VGGFace2(VisionDataset):
             Expects the following folder structure if download=False:
                 .
                 └── vggface2
+                    ├── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
                     ├── vggface2_train.tar.gz (or 'train' if uncompressed)
                     ├── vggface2_test.tar.gz (or 'test' if uncompressed)
                     ├── train_list.txt
-                    ├── test_list.txt
-                    └── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
+                    └── test_list.txt
             split (string): The dataset split to use. One of {``train``, ``test``}.
                 Defaults to ``train``.
             target_type (string): The type of target to use. One of
                 {``class_id``, ``image_id``, ``face_id``, ``bbox``, ``landmarks``.``""``}
                 Can also be a list to output a tuple with all specified target types.
                 The targets represent:
-                    ``class_id`` (string)
-                    ``image_id`` (string)
-                    ``face_id`` (string)
-                    ``bbox`` (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
+                    ``class_id``  (string)
+                    ``image_id``  (string)
+                    ``face_id``   (string)
+                    ``bbox``      (torch.tensor shape=(4,) dtype=int): bounding box (x, y, width, height)
                     ``landmarks`` (torch.tensor shape=(10,) dtype=float): values that
                         represent five points (P1X, P1Y, P2X, P2Y, P3X, P3Y, P4X, P4Y, P5X, P5Y)
                 Defaults to ``bbox``. If empty, ``None`` will be returned as target.
@@ -116,8 +116,9 @@ def __init__(
         with open(self.image_list_file, 'r') as f:
             for i, img_file in enumerate(f):
                 img_file = img_file.strip()
-                img_filename, ext = os.path.splitext(img_file)  # e.g. ["n004332/0317_01", "jpg"]
-                class_id, image_face_id = img_filename.split("/")
+                img_filename, ext = os.path.splitext(img_file)  #   e.g. ["n004332/0317_01", "jpg"]
+                class_id, image_face_id = img_filename.split("/") # e.g. ["n004332", "0317_01"]
+                class_id = class_id[1:]
                 image_id, face_id = image_face_id.split("_")
                 img_filepath = os.path.join(self.root, self.split, img_file)
                 self.img_info.append({

From 7510edfc898ddc4a359c1f07c4859b948669d4ce Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Fri, 30 Oct 2020 01:17:41 -0400
Subject: [PATCH 08/16] formatting update

---
 torchvision/datasets/vggface2.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 813c8917da2..1a45a71d3bd 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -116,8 +116,8 @@ def __init__(
         with open(self.image_list_file, 'r') as f:
             for i, img_file in enumerate(f):
                 img_file = img_file.strip()
-                img_filename, ext = os.path.splitext(img_file)  #   e.g. ["n004332/0317_01", "jpg"]
-                class_id, image_face_id = img_filename.split("/") # e.g. ["n004332", "0317_01"]
+                img_filename, ext = os.path.splitext(img_file)  # e.g. ["n004332/0317_01", "jpg"]
+                class_id, image_face_id = img_filename.split("/")  # e.g. ["n004332", "0317_01"]
                 class_id = class_id[1:]
                 image_id, face_id = image_face_id.split("_")
                 img_filepath = os.path.join(self.root, self.split, img_file)

From 170eff19be8414db00aa71a0036c58db0dfd83a9 Mon Sep 17 00:00:00 2001
From: Josh Bradley <jgbradley1@gmail.com>
Date: Fri, 30 Oct 2020 13:33:27 -0400
Subject: [PATCH 09/16] remove unused variable

---
 torchvision/datasets/vggface2.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 1a45a71d3bd..0bf298ad97f 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -114,7 +114,7 @@ def __init__(
         self.landmarks = pandas.concat(landmark_frames)
 
         with open(self.image_list_file, 'r') as f:
-            for i, img_file in enumerate(f):
+            for img_file in f:
                 img_file = img_file.strip()
                 img_filename, ext = os.path.splitext(img_file)  # e.g. ["n004332/0317_01", "jpg"]
                 class_id, image_face_id = img_filename.split("/")  # e.g. ["n004332", "0317_01"]

From d8d5e02c5de0afd9b16ac75d2e400b45aeddf184 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sat, 31 Oct 2020 04:16:56 -0400
Subject: [PATCH 10/16] use double quoted strings

---
 torchvision/datasets/vggface2.py | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 0bf298ad97f..f5a97453a23 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -79,7 +79,7 @@ def __init__(
             raise RuntimeError(msg)
 
         # check arguments
-        if split not in ('train', 'test'):
+        if split not in ("train", "test"):
             raise ValueError('split \"{}\" is not recognized.'.format(split))
         self.split = split
         self.img_info: List[Dict[str, object]] = []
@@ -92,9 +92,9 @@ def __init__(
         if not (all(x in ["class_id", "image_id", "face_id", "bbox", "landmarks", ""] for x in self.target_type)):
             raise ValueError("target_type \"{}\" is not recognized.".format(self.target_type))
         if not self.target_type and self.target_transform is not None:
-            raise RuntimeError('target_transform is specified but target_type is empty')
+            raise RuntimeError("target_transform is specified but target_type is empty")
 
-        image_list_file = 'train_list.txt' if self.split == 'train' else 'test_list.txt'
+        image_list_file = "train_list.txt" if self.split == "train" else "test_list.txt"
         self.image_list_file = os.path.join(self.root, image_list_file)
 
         # prepare dataset
@@ -122,12 +122,12 @@ def __init__(
                 image_id, face_id = image_face_id.split("_")
                 img_filepath = os.path.join(self.root, self.split, img_file)
                 self.img_info.append({
-                    'img_path': img_filepath,
-                    'class_id': class_id,
-                    'image_id': image_id,
-                    'face_id': face_id,
-                    'bbox': torch.tensor(self.bbox.loc[img_filename].values),
-                    'landmarks': torch.tensor(self.landmarks.loc[img_filename].values),
+                    "img_path": img_filepath,
+                    "class_id": class_id,
+                    "image_id": image_id,
+                    "face_id": face_id,
+                    "bbox": torch.tensor(self.bbox.loc[img_filename].values),
+                    "landmarks": torch.tensor(self.landmarks.loc[img_filename].values),
                 })
 
     def __len__(self) -> int:
@@ -137,7 +137,7 @@ def __getitem__(self, index) -> Tuple[Any, Any]:
         img_info = self.img_info[index]
 
         # prepare image
-        img = Image.open(img_info['img_path'])
+        img = Image.open(img_info["img_path"])
         if self.transform:
             img = self.transform(img)
 
@@ -157,4 +157,4 @@ def __getitem__(self, index) -> Tuple[Any, Any]:
 
     def extra_repr(self) -> str:
         lines = ["Target type: {target_type}", "Split: {split}"]
-        return '\n'.join(lines).format(**self.__dict__)
+        return "\n".join(lines).format(**self.__dict__)

From 58dfd04150d72381b5abc1ce1060a965d0544b06 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sat, 31 Oct 2020 04:26:02 -0400
Subject: [PATCH 11/16] add vggface2 unit test

---
 test/fakedata_generation.py | 96 +++++++++++++++++++++++++++++++++++++
 test/test_datasets.py       | 14 +++++-
 2 files changed, 109 insertions(+), 1 deletion(-)

diff --git a/test/fakedata_generation.py b/test/fakedata_generation.py
index e17a4309cdc..452822d7bbe 100644
--- a/test/fakedata_generation.py
+++ b/test/fakedata_generation.py
@@ -171,6 +171,102 @@ def _make_devkit_archive(root):
         yield root
 
 
+@contextlib.contextmanager
+def vggface2_root():
+
+    class_id = 'n000001'
+    image_id = '0001'
+    face_id = '01'
+    basefolder = 'vggface2'
+
+    def _make_image(file):
+        PIL.Image.fromarray(np.zeros((32, 32, 3), dtype=np.uint8)).save(file)
+
+    def _make_tar(archive, content, arcname=None, compress=False):
+        mode = 'w:gz' if compress else 'w'
+        if arcname is None:
+            arcname = os.path.basename(content)
+        with tarfile.open(archive, mode) as fh:
+            fh.add(content, arcname=arcname)
+
+    def _make_image_list_files(root):
+        image_list_contents = os.path.join(class_id, image_id + '_' + face_id + '.jpg')  # e.g. n000001/0001_01.jpg
+        # train image list
+        image_list_file = os.path.join(root, "train_list.txt")
+        with open(image_list_file, "w") as txt_file:
+            txt_file.write(image_list_contents)
+        # test image list
+        image_list_file = os.path.join(root, "test_list.txt")
+        with open(image_list_file, "w") as txt_file:
+            txt_file.write(image_list_contents)
+
+    def _make_train_archive(root):
+        with get_tmp_dir() as tmp:
+            extracted_dir = os.path.join(tmp, 'train', class_id)
+            os.makedirs(extracted_dir)
+            print("EXTRACTED_DIR: " + extracted_dir)
+            _make_image(os.path.join(extracted_dir, image_id + '_' + face_id + '.jpg'))
+            train_archive = os.path.join(root, 'vggface2_train.tar.gz')
+            top_level_dir = os.path.join(tmp, 'train')
+            _make_tar(train_archive, top_level_dir, arcname='train', compress=True)
+
+    def _make_test_archive(root):
+        with get_tmp_dir() as tmp:
+            extracted_dir = os.path.join(tmp, 'test', class_id)
+            os.makedirs(extracted_dir)
+            _make_image(os.path.join(extracted_dir, image_id + '_' + face_id + '.jpg'))
+            test_archive = os.path.join(root, 'vggface2_test.tar.gz')
+            top_level_dir = os.path.join(tmp, 'test')
+            _make_tar(test_archive, top_level_dir, arcname='test', compress=True)
+
+    def _make_bb_landmark_archive(root):
+        train_bb_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
+        test_bb_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
+        train_landmark_contents = ('NAME_ID,P1X,P1Y,P2X,P2Y,P3X,P3Y,P4X,P4Y,P5X,P5Y\n'
+                                   '"n000001/0001_01",75.81253,110.2077,103.1778,104.6074,'
+                                   '90.06353,133.3624,85.39182,149.4176,114.9009,144.9259')
+        test_landmark_contents = ('NAME_ID,P1X,P1Y,P2X,P2Y,P3X,P3Y,P4X,P4Y,P5X,P5Y\n'
+                                  '"n000001/0001_01",75.81253,110.2077,103.1778,104.6074,'
+                                  '90.06353,133.3624,85.39182,149.4176,114.9009,144.9259')
+
+        with get_tmp_dir() as tmp:
+            extracted_dir = os.path.join(tmp, 'bb_landmark')
+            os.makedirs(extracted_dir)
+
+            # bbox training file
+            bbox_file = os.path.join(extracted_dir, "loose_bb_train.csv")
+            with open(bbox_file, "w") as csv_file:
+                csv_file.write(train_bb_contents)
+
+            # bbox testing file
+            bbox_file = os.path.join(extracted_dir, "loose_bb_test.csv")
+            with open(bbox_file, "w") as csv_file:
+                csv_file.write(test_bb_contents)
+
+            # landmark training file
+            landmark_file = os.path.join(extracted_dir, "loose_landmark_train.csv")
+            with open(landmark_file, "w") as csv_file:
+                csv_file.write(train_landmark_contents)
+
+            # landmark testing file
+            landmark_file = os.path.join(extracted_dir, "loose_landmark_test.csv")
+            with open(landmark_file, "w") as csv_file:
+                csv_file.write(test_landmark_contents)
+
+            archive = os.path.join(root, 'bb_landmark.tar.gz')
+            _make_tar(archive, extracted_dir, compress=True)
+
+    with get_tmp_dir() as root:
+        root_base = os.path.join(root, basefolder)
+        os.makedirs(root_base)
+        _make_train_archive(root_base)
+        _make_test_archive(root_base)
+        _make_image_list_files(root_base)
+        _make_bb_landmark_archive(root_base)
+
+        yield root
+
+
 @contextlib.contextmanager
 def cityscapes_root():
 
diff --git a/test/test_datasets.py b/test/test_datasets.py
index af092e1845d..ea9fdd5a350 100644
--- a/test/test_datasets.py
+++ b/test/test_datasets.py
@@ -9,7 +9,7 @@
 import torchvision
 from common_utils import get_tmp_dir
 from fakedata_generation import mnist_root, cifar_root, imagenet_root, \
-    cityscapes_root, svhn_root, voc_root, ucf101_root, places365_root
+    cityscapes_root, svhn_root, voc_root, ucf101_root, places365_root, vggface2_root
 import xml.etree.ElementTree as ET
 from urllib.request import Request, urlopen
 import itertools
@@ -139,6 +139,18 @@ def test_imagenet(self, mock_verify):
             dataset = torchvision.datasets.ImageNet(root, split='val')
             self.generic_classification_dataset_test(dataset)
 
+    def test_vggface2(self):
+        with vggface2_root() as root:
+            dataset = torchvision.datasets.VGGFace2(root, split='train')
+            self.assertEqual(len(dataset), 1)
+            img, target = dataset[0]
+            self.assertTrue(isinstance(img, PIL.Image.Image))
+
+            dataset = torchvision.datasets.VGGFace2(root, split='test')
+            self.assertEqual(len(dataset), 1)
+            img, target = dataset[0]
+            self.assertTrue(isinstance(img, PIL.Image.Image))
+
     @mock.patch('torchvision.datasets.cifar.check_integrity')
     @mock.patch('torchvision.datasets.cifar.CIFAR10._check_integrity')
     def test_cifar10(self, mock_ext_check, mock_int_check):

From baf22b79f582737ac218e377eee4c66a0617e7ac Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sat, 31 Oct 2020 13:49:09 -0400
Subject: [PATCH 12/16] add pandas check

---
 test/test_datasets.py | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/test/test_datasets.py b/test/test_datasets.py
index ea9fdd5a350..a1a173b5d95 100644
--- a/test/test_datasets.py
+++ b/test/test_datasets.py
@@ -15,6 +15,12 @@
 import itertools
 
 
+try:
+    import pandas
+    HAS_PANDAS = True
+except ImportError:
+    HAS_PANDAS = False
+
 try:
     import scipy
     HAS_SCIPY = True
@@ -139,6 +145,7 @@ def test_imagenet(self, mock_verify):
             dataset = torchvision.datasets.ImageNet(root, split='val')
             self.generic_classification_dataset_test(dataset)
 
+    @unittest.skipIf(not HAS_PANDAS, "pandas unavailable")
     def test_vggface2(self):
         with vggface2_root() as root:
             dataset = torchvision.datasets.VGGFace2(root, split='train')

From 1256a351d57f4b3fe481a9c3bb3f8db6c828a93e Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sat, 31 Oct 2020 14:17:09 -0400
Subject: [PATCH 13/16] add docstring to vggface fakedata generator

---
 test/fakedata_generation.py | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/test/fakedata_generation.py b/test/fakedata_generation.py
index 452822d7bbe..36f2e4b898e 100644
--- a/test/fakedata_generation.py
+++ b/test/fakedata_generation.py
@@ -173,6 +173,18 @@ def _make_devkit_archive(root):
 
 @contextlib.contextmanager
 def vggface2_root():
+    """
+    Generates a dataset with the following folder structure and returns the path root:
+    <root>
+    └── vggface2
+        ├── bb_landmark.tar.gz ('bb_landmark' when uncompressed)
+        ├── vggface2_train.tar.gz ('train' when uncompressed)
+        ├── vggface2_test.tar.gz ('test' when uncompressed)
+        ├── train_list.txt
+        └── test_list.txt
+
+    The dataset consist of 1 image in the train set and 1 image in the test set.
+    """
 
     class_id = 'n000001'
     image_id = '0001'

From 941682b4bcd3243df65b26b1546ed318a581f80c Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sun, 1 Nov 2020 18:26:31 -0500
Subject: [PATCH 14/16] fix docstring indentation

---
 torchvision/datasets/vggface2.py | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index f5a97453a23..63c9416782d 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -20,13 +20,13 @@ class VGGFace2(VisionDataset):
         Args:
             root (string): Root directory of the VGGFace2 Dataset.
             Expects the following folder structure if download=False:
-                .
-                └── vggface2
-                    ├── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
-                    ├── vggface2_train.tar.gz (or 'train' if uncompressed)
-                    ├── vggface2_test.tar.gz (or 'test' if uncompressed)
-                    ├── train_list.txt
-                    └── test_list.txt
+                <root>
+                    └── vggface2
+                        ├── bb_landmark.tar.gz (or 'bb_landmark' if uncompressed)
+                        ├── vggface2_train.tar.gz (or 'train' if uncompressed)
+                        ├── vggface2_test.tar.gz (or 'test' if uncompressed)
+                        ├── train_list.txt
+                        └── test_list.txt
             split (string): The dataset split to use. One of {``train``, ``test``}.
                 Defaults to ``train``.
             target_type (string): The type of target to use. One of

From a39251410be01d1010cd6a95c15f5ade3aba4b06 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Sun, 1 Nov 2020 20:03:31 -0500
Subject: [PATCH 15/16] use local variable scope and fixed minor docstring
 formatting

---
 test/fakedata_generation.py      | 20 ++++++++++----------
 torchvision/datasets/vggface2.py |  4 ++--
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/test/fakedata_generation.py b/test/fakedata_generation.py
index 36f2e4b898e..430452db100 100644
--- a/test/fakedata_generation.py
+++ b/test/fakedata_generation.py
@@ -176,12 +176,12 @@ def vggface2_root():
     """
     Generates a dataset with the following folder structure and returns the path root:
     <root>
-    └── vggface2
-        ├── bb_landmark.tar.gz ('bb_landmark' when uncompressed)
-        ├── vggface2_train.tar.gz ('train' when uncompressed)
-        ├── vggface2_test.tar.gz ('test' when uncompressed)
-        ├── train_list.txt
-        └── test_list.txt
+        └── vggface2
+            ├── bb_landmark.tar.gz ('bb_landmark' when uncompressed)
+            ├── vggface2_train.tar.gz ('train' when uncompressed)
+            ├── vggface2_test.tar.gz ('test' when uncompressed)
+            ├── train_list.txt
+            └── test_list.txt
 
     The dataset consist of 1 image in the train set and 1 image in the test set.
     """
@@ -232,8 +232,8 @@ def _make_test_archive(root):
             _make_tar(test_archive, top_level_dir, arcname='test', compress=True)
 
     def _make_bb_landmark_archive(root):
-        train_bb_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
-        test_bb_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
+        train_bbox_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
+        test_bbox_contents = 'NAME_ID,X,Y,W,H\n"n000001/0001_01",161,140,224,324'
         train_landmark_contents = ('NAME_ID,P1X,P1Y,P2X,P2Y,P3X,P3Y,P4X,P4Y,P5X,P5Y\n'
                                    '"n000001/0001_01",75.81253,110.2077,103.1778,104.6074,'
                                    '90.06353,133.3624,85.39182,149.4176,114.9009,144.9259')
@@ -248,12 +248,12 @@ def _make_bb_landmark_archive(root):
             # bbox training file
             bbox_file = os.path.join(extracted_dir, "loose_bb_train.csv")
             with open(bbox_file, "w") as csv_file:
-                csv_file.write(train_bb_contents)
+                csv_file.write(train_bbox_contents)
 
             # bbox testing file
             bbox_file = os.path.join(extracted_dir, "loose_bb_test.csv")
             with open(bbox_file, "w") as csv_file:
-                csv_file.write(test_bb_contents)
+                csv_file.write(test_bbox_contents)
 
             # landmark training file
             landmark_file = os.path.join(extracted_dir, "loose_landmark_train.csv")
diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 63c9416782d..4613f3bf3dd 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -95,7 +95,7 @@ def __init__(
             raise RuntimeError("target_transform is specified but target_type is empty")
 
         image_list_file = "train_list.txt" if self.split == "train" else "test_list.txt"
-        self.image_list_file = os.path.join(self.root, image_list_file)
+        image_list_file = os.path.join(self.root, image_list_file)
 
         # prepare dataset
         for (filename, _, extracted_dir) in self.file_list:
@@ -113,7 +113,7 @@ def __init__(
                            pandas.read_csv(fn("loose_landmark_test.csv"), index_col=0)]
         self.landmarks = pandas.concat(landmark_frames)
 
-        with open(self.image_list_file, 'r') as f:
+        with open(image_list_file, 'r') as f:
             for img_file in f:
                 img_file = img_file.strip()
                 img_filename, ext = os.path.splitext(img_file)  # e.g. ["n004332/0317_01", "jpg"]

From 70891563de0fb7785c448b84d856064fcc7f14e0 Mon Sep 17 00:00:00 2001
From: Joshua Bradley <jgbradley1@gmail.com>
Date: Thu, 12 Nov 2020 00:04:31 -0500
Subject: [PATCH 16/16] minor style fixes

---
 torchvision/datasets/vggface2.py | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/torchvision/datasets/vggface2.py b/torchvision/datasets/vggface2.py
index 4613f3bf3dd..c49099f9c15 100644
--- a/torchvision/datasets/vggface2.py
+++ b/torchvision/datasets/vggface2.py
@@ -3,7 +3,7 @@
 import os
 import torch
 from typing import Any, Callable, Dict, List, Optional, Tuple, Union
-from .utils import check_integrity, extract_archive
+from .utils import check_integrity, extract_archive, verify_str_arg
 from .vision import VisionDataset
 
 
@@ -49,8 +49,8 @@ class VGGFace2(VisionDataset):
                 downloaded again.
         """
 
-    base_folder = "vggface2"
-    file_list = [
+    BASE_FOLDER = "vggface2"
+    FILE_LIST = [
         # Filename                MD5 Hash                            Uncompressed filename
         ("vggface2_train.tar.gz", "88813c6b15de58afc8fa75ea83361d7f", "train"),
         ("vggface2_test.tar.gz", "bb7a323824d1004e14e00c23974facd3", "test"),
@@ -67,7 +67,7 @@ def __init__(
         download: bool = False,
     ) -> None:
         import pandas
-        super(VGGFace2, self).__init__(root=os.path.join(root, self.base_folder),
+        super(VGGFace2, self).__init__(root=os.path.join(root, self.BASE_FOLDER),
                                        transform=transform,
                                        target_transform=target_transform)
 
@@ -79,18 +79,17 @@ def __init__(
             raise RuntimeError(msg)
 
         # check arguments
-        if split not in ("train", "test"):
-            raise ValueError('split \"{}\" is not recognized.'.format(split))
-        self.split = split
+        self.split = verify_str_arg(split, "split", ("train", "test"))
         self.img_info: List[Dict[str, object]] = []
 
         if isinstance(target_type, list):
             self.target_type = target_type
         else:
             self.target_type = [target_type]
+        self.target_type = [verify_str_arg(t, "target_type",
+                            ("class_id", "image_id", "face_id", "bbox", "landmarks", ""))
+                            for t in self.target_type]
 
-        if not (all(x in ["class_id", "image_id", "face_id", "bbox", "landmarks", ""] for x in self.target_type)):
-            raise ValueError("target_type \"{}\" is not recognized.".format(self.target_type))
         if not self.target_type and self.target_transform is not None:
             raise RuntimeError("target_transform is specified but target_type is empty")
 
@@ -98,14 +97,14 @@ def __init__(
         image_list_file = os.path.join(self.root, image_list_file)
 
         # prepare dataset
-        for (filename, _, extracted_dir) in self.file_list:
+        for (filename, _, extracted_dir) in self.FILE_LIST:
             filename = os.path.join(self.root, filename)
             extracted_dir_path = os.path.join(self.root, extracted_dir)
             if not os.path.isdir(extracted_dir_path):
                 extract_archive(filename)
 
         # process dataset
-        fn = partial(os.path.join, self.root, self.file_list[2][2])
+        fn = partial(os.path.join, self.root, self.FILE_LIST[2][2])
         bbox_frames = [pandas.read_csv(fn("loose_bb_train.csv"), index_col=0),
                        pandas.read_csv(fn("loose_bb_test.csv"), index_col=0)]
         self.bbox = pandas.concat(bbox_frames)
@@ -134,10 +133,8 @@ def __len__(self) -> int:
         return len(self.img_info)
 
     def __getitem__(self, index) -> Tuple[Any, Any]:
-        img_info = self.img_info[index]
-
         # prepare image
-        img = Image.open(img_info["img_path"])
+        img = Image.open(self.img_info[index]["img_path"])
         if self.transform:
             img = self.transform(img)
 
@@ -147,7 +144,7 @@ def __getitem__(self, index) -> Tuple[Any, Any]:
             if t == "":
                 target = None
                 break
-            target.append(img_info[t])
+            target.append(self.img_info[index][t])
         if target:
             target = tuple(target) if len(target) > 1 else target[0]
             if self.target_transform is not None: