Add FLIC dataset

chainercv/datasets/__init__.py

@@ -22,6 +22,8 @@
 from chainercv.datasets.cub.cub_utils import cub_label_names  # NOQA
 from chainercv.datasets.directory_parsing_label_dataset import directory_parsing_label_names  # NOQA
 from chainercv.datasets.directory_parsing_label_dataset import DirectoryParsingLabelDataset  # NOQA
+from chainercv.datasets.flic.flic_keypoint_dataset import FLICKeypointDataset  # NOQA
+from chainercv.datasets.flic.flic_utils import flic_joint_names  # NOQA
 from chainercv.datasets.mixup_soft_label_dataset import MixUpSoftLabelDataset  # NOQA
 from chainercv.datasets.online_products.online_products_dataset import online_products_super_label_names  # NOQA
 from chainercv.datasets.online_products.online_products_dataset import OnlineProductsDataset  # NOQA

chainercv/datasets/flic/flic_keypoint_dataset.py

@@ -0,0 +1,138 @@
+import glob
+import os
+
+import numpy as np
+
+import chainer
+from chainercv.datasets.flic import flic_utils
+from chainercv import utils
+
+try:
+    from scipy.io import loadmat
+    _scipy_available = True
+except (ImportError, TypeError):
+    _scipy_available = False
+
+
+class FLICKeypointDataset(chainer.dataset.DatasetMixin):
+
+    """`Frames Labaled in Cinema (FLIC)`_ dataset  with annotated keypoints.
+
+    .. _`Frames Labaled in Cinema (FLIC)`:
+        https://bensapp.github.io/flic-dataset.html
+
+    An index corresponds to each image.
+
+    When queried by an index, this dataset returns the corresponding
+    :obj:`img, keypoint`, which is a tuple of an image and keypoints
+    that indicates visible keypoints in the image.
+    The data type of the two elements are :obj:`float32, float32`.
+
+    The keypoints are packed into a two dimensional array of shape
+    :math:`(K, 2)`, where :math:`K` is the number of keypoints.
+    Note that :math:`K=29` in FLIC dataset. Also note that not all
+    keypoints are visible in an image. When a keypoint is not visible,
+    the values stored for that keypoint are :obj:`~numpy.nan`. The second axis
+    corresponds to the :math:`y` and :math:`x` coordinates of the
+    keypoints in the image.
+
+    The torso bounding box is a one-dimensional array of shape :math:`(4,)`.
+    The elements of the bounding box corresponds to
+    :math:`(y_{min}, x_{min}, y_{max}, x_{max})`, where the four attributes are
+    coordinates of the top left and the bottom right vertices.
+    This information can optionally be retrieved from the dataset
+    by setting :obj:`return_torsobox = True`.
+
+    Args:
+        data_dir (string): Path to the root of the training data. If this is
+            :obj:`auto`, this class will automatically download data for you
+            under :obj:`$CHAINER_DATASET_ROOT/pfnet/chainercv/flic/FLIC-full`.
+        split ({'train', 'test'}): Select from dataset splits used in
+            the FLIC dataset.
+        return_torsobox (bool): If :obj:`True`, this returns a bounding box
+            around the torso. The default value is :obj:`False`.
+        use_bad (bool): If :obj:`False`, the data which have :obj:`isbad = 1`
+            will be ignored. The default is :obj:`False`.
+        use_unchecked (bool): If :obj:`False`, the data which have
+            :obj:`isunchecked = 1` will be ignored. The default is
+            :obj:`False`.
+
+    """
+
+    def __init__(self, data_dir='auto', split='train', return_torsobox=False,
+                 use_bad=False, use_unchecked=False):
+        super(FLICKeypointDataset, self).__init__()
+        if split not in ['train', 'test']:
+            raise ValueError(
+                '\'split\' argment should be eighter \'train\' or \'test\'.')
+
+        if not _scipy_available:
+            raise ImportError(
+                'scipy is needed to extract labales from the .mat file.'
+                'Please install scipy:\n\n'
+                '\t$pip install scipy\n\n')
+
+        if data_dir == 'auto':
+            data_dir = flic_utils.get_flic()
+
+        img_paths = {os.path.basename(fn): fn for fn in glob.glob(
+            os.path.join(data_dir, 'images', '*.jpg'))}
+
+        label_annos = [
+            'poselet_hit_idx',
+            'moviename',
+            'coords',
+            'filepath',
+            'imgdims',
+            'currframe',
+            'torsobox',
+            'istrain',
+            'istest',
+            'isbad',
+            'isunchecked',
+        ]
+        annos = loadmat(os.path.join(data_dir, 'examples.mat'))
+
+        self.img_paths = list()
+        self.keypoints = list()
+        self.torsoboxes = list()
+        self.return_torsobox = return_torsobox
+
+        for label in annos['examples'][0]:
+            label = {label_annos[i]: val for i, val in enumerate(label)}
+            if not use_bad and int(label['isbad']) == 1:
+                continue
+            if not use_unchecked and int(label['isunchecked']) == 1:
+                continue
+            if ((split == 'train' and int(label['istrain']) == 0)
+                    or (split == 'test' and int(label['istest']) == 0)):
+                continue
+
+            self.img_paths.append(img_paths[label['filepath'][0]])
+            self.keypoints.append(label['coords'].T[:, ::-1])
+            if return_torsobox:
+                self.torsoboxes.append(label['torsobox'][0, [1, 0, 3, 2]])
+
+    def __len__(self):
+        return len(self.img_paths)
+
+    def get_example(self, i):
+        """Returns the i-th example.
+
+        Args:
+            i (int): The index of the example.
+
+        Returns:
+            tuple of an image and keypoints.
+            The image is in CHW format and its color channel is ordered in
+            RGB.
+            If :obj:`return_torsobox = True`,
+            a bounding box is appended to the returned value.
+
+        """
+        img = utils.read_image(self.img_paths[i])
+        keypoint = np.array(self.keypoints[i], dtype=np.float32)
+        if self.return_torsobox:
+            return img, keypoint, self.torsoboxes[i]
+        else:
+            return img, keypoint

chainercv/datasets/flic/flic_utils.py

@@ -0,0 +1,51 @@
+import os
+
+from chainer.dataset import download
+from chainercv import utils
+
+root = 'pfnet/chainercv/flic'
+
+url = 'http://vision.grasp.upenn.edu/video/FLIC-full.zip'
+
+flic_joint_names = [
+    'lsho',
+    'lelb',
+    'lwri',
+    'rsho',
+    'relb',
+    'rwri',
+    'lhip',
+    'lkne',
+    'lank',
+    'rhip',
+    'rkne',
+    'rank',
+    'leye',
+    'reye',
+    'lear',
+    'rear',
+    'nose',
+    'msho',
+    'mhip',
+    'mear',
+    'mtorso',
+    'mluarm',
+    'mruarm',
+    'mllarm',
+    'mrlarm',
+    'mluleg',
+    'mruleg',
+    'mllleg',
+    'mrlleg'
+]
+
+
+def get_flic():
+    data_root = download.get_dataset_directory(root)
+    dataset_dir = os.path.join(data_root, 'FLIC-full')
+    if not os.path.exists(dataset_dir):
+        download_file_path = utils.cached_download(url)
+        ext = os.path.splitext(url)[1]
+        utils.extractall(download_file_path, data_root, ext)
+
+    return dataset_dir

docs/source/reference/datasets.rst

@@ -62,6 +62,13 @@ CUBKeypointDataset
 ~~~~~~~~~~~~~~~~~~
 .. autoclass:: CUBKeypointDataset
 
+FLIC
+----
+
+FLICKeypointDataset
+~~~~~~~~~~~~~~~~~~~
+.. autoclass:: FLICKeypointDataset
+
 MS COCO
 -------
 
@@ -84,7 +91,6 @@ OnlineProductsDataset
 ~~~~~~~~~~~~~~~~~~~~~
 .. autoclass:: OnlineProductsDataset
 
-
 PASCAL VOC
 ----------