import os
import os.path as osp
import shutil
from itertools import chain
from xml.dom import minidom
import torch
import torch.nn.functional as F
import numpy as np
from torch_geometric.data import (InMemoryDataset, Data, download_url,
extract_tar)
try:
import torchvision.models as models
import torchvision.transforms as T
from PIL import Image
except ImportError:
models = None
T = None
Image = None
[docs]class PascalVOCKeypoints(InMemoryDataset):
r"""The Pascal VOC 2011 dataset with Berkely annotations of keypoints from
the `"Poselets: Body Part Detectors Trained Using 3D Human Pose
Annotations" <https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/
human/ poselets_iccv09.pdf>`_ paper, containing 3 to 23 keypoints per
example over 20 categories.
The dataset is pre-filtered to exclude difficult, occluded and truncated
objects.
The keypoints contain interpolated features from a pre-trained VGG16 model
on ImageNet (:obj:`relu4_2` and :obj:`relu5_1`).
Args:
root (string): Root directory where the dataset should be saved.
category (string): The category of the images (one of
:obj:`"Aeroplane"`, :obj:`"Bicycle"`, :obj:`"Bird"`,
:obj:`"Boat"`, :obj:`"Bottle"`, :obj:`"Bus"`, :obj:`"Car"`,
:obj:`"Cat"`, :obj:`"Chair"`, :obj:`"Diningtable"`, :obj:`"Dog"`,
:obj:`"Horse"`, :obj:`"Motorbike"`, :obj:`"Person"`,
:obj:`"Pottedplant"`, :obj:`"Sheep"`, :obj:`"Sofa"`,
:obj:`"Train"`, :obj:`"TVMonitor"`)
train (bool, optional): If :obj:`True`, loads the training dataset,
otherwise the test dataset. (default: :obj:`True`)
transform (callable, optional): A function/transform that takes in an
:obj:`torch_geometric.data.Data` object and returns a transformed
version. The data object will be transformed before every access.
(default: :obj:`None`)
pre_transform (callable, optional): A function/transform that takes in
an :obj:`torch_geometric.data.Data` object and returns a
transformed version. The data object will be transformed before
being saved to disk. (default: :obj:`None`)
pre_filter (callable, optional): A function that takes in an
:obj:`torch_geometric.data.Data` object and returns a boolean
value, indicating whether the data object should be included in the
final dataset. (default: :obj:`None`)
"""
image_url = ('http://host.robots.ox.ac.uk/pascal/VOC/voc2011/'
'VOCtrainval_25-May-2011.tar')
annotation_url = ('https://www2.eecs.berkeley.edu/Research/Projects/CS/'
'vision/shape/poselets/voc2011_keypoints_Feb2012.tgz')
split_url = 'http://cvgl.stanford.edu/projects/ucn/voc2011_pairs.npz'
categories = [
'aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat',
'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike', 'person',
'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor'
]
def __init__(self, root, category, train=True, transform=None,
pre_transform=None, pre_filter=None):
self.category = category.lower()
assert self.category in self.categories
super(PascalVOCKeypoints, self).__init__(root, transform,
pre_transform, pre_filter)
path = self.processed_paths[0] if train else self.processed_paths[1]
self.data, self.slices = torch.load(path)
@property
def raw_file_names(self):
return ['images', 'annotations', 'splits.npz']
@property
def processed_file_names(self):
category = self.category
return [x.format(category) for x in ['training_{}.pt', 'test_{}.pt']]
def download(self):
path = download_url(self.image_url, self.raw_dir)
extract_tar(path, self.raw_dir, mode='r')
os.unlink(path)
image_path = osp.join(self.raw_dir, 'TrainVal', 'VOCdevkit', 'VOC2011')
os.rename(image_path, osp.join(self.raw_dir, 'images'))
shutil.rmtree(osp.join(self.raw_dir, 'TrainVal'))
path = download_url(self.annotation_url, self.raw_dir)
extract_tar(path, self.raw_dir)
os.unlink(path)
path = download_url(self.split_url, self.raw_dir)
os.rename(path, osp.join(self.raw_dir, 'splits.npz'))
def process(self):
if models is None or T is None or Image is None:
raise ImportError('Package `torchvision` could not be found.')
splits = np.load(osp.join(self.raw_dir, 'splits.npz'),
allow_pickle=True)
category_idx = self.categories.index(self.category)
train_split = list(splits['train'])[category_idx]
test_split = list(splits['test'])[category_idx]
image_path = osp.join(self.raw_dir, 'images', 'JPEGImages')
info_path = osp.join(self.raw_dir, 'images', 'Annotations')
annotation_path = osp.join(self.raw_dir, 'annotations')
labels = {}
vgg16_outputs = []
def hook(module, x, y):
vgg16_outputs.append(y)
vgg16 = models.vgg16(pretrained=True)
vgg16.eval()
vgg16.features[20].register_forward_hook(hook) # relu4_2
vgg16.features[25].register_forward_hook(hook) # relu5_1
transform = T.Compose([
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
train_data_list, test_data_list = [], []
for i, name in enumerate(chain(train_split, test_split)):
filename = '_'.join(name.split('/')[1].split('_')[:-1])
idx = int(name.split('_')[-1].split('.')[0]) - 1
path = osp.join(info_path, '{}.xml'.format(filename))
obj = minidom.parse(path).getElementsByTagName('object')[idx]
trunc = obj.getElementsByTagName('truncated')[0].firstChild.data
occ = obj.getElementsByTagName('occluded')
occ = '0' if len(occ) == 0 else occ[0].firstChild.data
diff = obj.getElementsByTagName('difficult')[0].firstChild.data
if bool(int(trunc)) or bool(int(occ)) or bool(int(diff)):
continue
xmin = float(obj.getElementsByTagName('xmin')[0].firstChild.data)
xmax = float(obj.getElementsByTagName('xmax')[0].firstChild.data)
ymin = float(obj.getElementsByTagName('ymin')[0].firstChild.data)
ymax = float(obj.getElementsByTagName('ymax')[0].firstChild.data)
box = (xmin, ymin, xmax, ymax)
dom = minidom.parse(osp.join(annotation_path, name))
keypoints = dom.getElementsByTagName('keypoint')
if len(keypoints) == 0:
continue
poss, ys = [], []
for keypoint in keypoints:
label = keypoint.attributes['name'].value
if label not in labels:
labels[label] = len(labels)
ys.append(labels[label])
poss.append(float(keypoint.attributes['x'].value))
poss.append(float(keypoint.attributes['y'].value))
y = torch.tensor(ys, dtype=torch.long)
pos = torch.tensor(poss, dtype=torch.float).view(-1, 2)
# Adjust bounding box (and positions) + add a small offset because
# some keypoints lay outside the bounding boxes.
box = (min(pos[:, 0].min().item(), box[0]) - 20,
min(pos[:, 1].min().item(), box[1]) - 20,
max(pos[:, 0].max().item(), box[2]) + 20,
max(pos[:, 1].max().item(), box[3]) + 20)
pos[:, 0] = pos[:, 0] - box[0]
pos[:, 1] = pos[:, 1] - box[1]
path = osp.join(image_path, '{}.jpg'.format(filename))
with open(path, 'rb') as f:
img = Image.open(f).convert('RGB').crop(box)
img = transform(img)
size = img.size()[-2:]
vgg16_outputs.clear()
with torch.no_grad():
vgg16(img.unsqueeze(0))
xs = []
for out in vgg16_outputs:
out = F.interpolate(out, size, mode='bilinear',
align_corners=False)
out = out.squeeze(0).permute(1, 2, 0)
out = out[pos[:, 1].round().long(), pos[:, 0].round().long()]
xs.append(out)
# Reset position.
pos[:, 0] = pos[:, 0] + box[0] - xmin
pos[:, 1] = pos[:, 1] + box[1] - ymin
data = Data(x=torch.cat(xs, dim=-1), pos=pos, y=y)
if self.pre_filter is not None and not self.pre_filter(data):
continue
if self.pre_transform is not None:
data = self.pre_transform(data)
if i < len(train_split):
train_data_list.append(data)
else:
test_data_list.append(data)
torch.save(self.collate(train_data_list), self.processed_paths[0])
torch.save(self.collate(test_data_list), self.processed_paths[1])
def __repr__(self):
return '{}({}, category={})'.format(self.__class__.__name__, len(self),
self.category)