Datasets¶
This section describes how to install common datasets used for training new models and for computing evaluation scores for entire datasets. In general, these datasets are large and require a computer with a good GPU to train and evaluate in reasonable times. Additional datasets are availble as plugins (for example CrowdPose).
Note
These datasets are not required to do pose predictions on your own images. Even for training, you are unlikely to need all the datasets for your use case.
Download COCO data¶
COCO is a great datasets containing many types of annotations, including bounding boxes, 2D poses, etc.
mkdir data-mscoco
cd data-mscoco
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
wget http://images.cocodataset.org/annotations/image_info_test2017.zip
unzip annotations_trainval2017.zip
unzip image_info_test2017.zip
mkdir images
cd images
wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/zips/train2017.zip
wget http://images.cocodataset.org/zips/test2017.zip
unzip val2017.zip
unzip train2017.zip
unzip test2017.zip
COCO Person Skeletons¶
COCO / kinematic tree / dense:
# HIDE CODE
# first make an annotation
ann_coco = openpifpaf.Annotation.from_cif_meta(
openpifpaf.plugins.coco.CocoKp().head_metas[0])
ann_kin = openpifpaf.Annotation.from_cif_meta(
openpifpaf.plugins.coco.CocoKp(skeleton=openpifpaf.plugins.coco.constants.KINEMATIC_TREE_SKELETON).head_metas[0])
ann_dense = openpifpaf.Annotation.from_cif_meta(
openpifpaf.plugins.coco.CocoKp(skeleton=openpifpaf.plugins.coco.constants.DENSER_COCO_PERSON_SKELETON).head_metas[0])
# visualize the annotation
openpifpaf.show.KeypointPainter.show_joint_scales = True
keypoint_painter = openpifpaf.show.KeypointPainter()
with openpifpaf.show.Canvas.annotation(ann_coco, ncols=3) as (ax1, ax2, ax3):
keypoint_painter.annotation(ax1, ann_coco)
keypoint_painter.annotation(ax2, ann_kin)
keypoint_painter.annotation(ax3, ann_dense)
COCO Person Keypoints¶
for i, name in enumerate(openpifpaf.plugins.coco.constants.COCO_KEYPOINTS):
print(i, name)
0 nose
1 left_eye
2 right_eye
3 left_ear
4 right_ear
5 left_shoulder
6 right_shoulder
7 left_elbow
8 right_elbow
9 left_wrist
10 right_wrist
11 left_hip
12 right_hip
13 left_knee
14 right_knee
15 left_ankle
16 right_ankle
print('associations')
kp_names = openpifpaf.plugins.coco.constants.COCO_KEYPOINTS
for i, (joint1, joint2) in enumerate(openpifpaf.plugins.coco.constants.COCO_PERSON_SKELETON):
print('{:2d}: {:15s} --> {}'.format(i, kp_names[joint1 - 1], kp_names[joint2 - 1]))
associations
0: left_ankle --> left_knee
1: left_knee --> left_hip
2: right_ankle --> right_knee
3: right_knee --> right_hip
4: left_hip --> right_hip
5: left_shoulder --> left_hip
6: right_shoulder --> right_hip
7: left_shoulder --> right_shoulder
8: left_shoulder --> left_elbow
9: right_shoulder --> right_elbow
10: left_elbow --> left_wrist
11: right_elbow --> right_wrist
12: left_eye --> right_eye
13: nose --> left_eye
14: nose --> right_eye
15: left_eye --> left_ear
16: right_eye --> right_ear
17: left_ear --> left_shoulder
18: right_ear --> right_shoulder
Download MPII data¶
This MPII data is currently not used anywhere.
mkdir data-mpii
cd data-mpii
wget https://datasets.d2.mpi-inf.mpg.de/andriluka14cvpr/mpii_human_pose_v1.tar.gz
wget https://datasets.d2.mpi-inf.mpg.de/andriluka14cvpr/mpii_human_pose_v1_u12_2.zip
Download NightOwls data¶
mkdir data-nightowls
cd data-nightowls
wget http://www.robots.ox.ac.uk/\~vgg/data/nightowls/python/nightowls_validation.json
wget http://www.robots.ox.ac.uk/\~vgg/data/nightowls/python/nightowls_validation.zip
unzip nightowls_validation.zip