cuiaiyu
commented
2 years ago Option 1:
Follow Get Started -> Data Preparation -> Optionally, you can also generate these files by yourself. in PATN Repository to get the key points label.
Option 2:
For a test imagexxx.jpg, run openpose to get the keypoints file xxx_keypoints.json (Body_25 label), and then you can load the key points as pytorch Tensor for the single test image by function
def load_pose_from_json(pose_json, target_size=(256,256), orig_size=(256,256)):
'''
This function converts the OpenPose detected key points (in .json file) to the desired heatmap.
input:
- pose_json (str): the file_path of the OpenPose detection in .json.
- target_size (tuple): the size of output heatmap
- orig_size (tuple): the size of original image that is used for OpenPose to detect the key points.
Output:
- heatmap (torch.Tensor) : the heatmap in size 18xHxW as specified by target_size
'''
with open(pose_json, 'r') as f:
anno = json.load(f)
if len(anno['people']) < 1:
a,b = target_size
return torch.zeros((18,a,b))
anno = list(anno['people'][0]['pose_keypoints_2d'])
x = np.array(anno[1::3])
y = np.array(anno[::3])
x[8:-1] = x[9:]
y = np.array(anno[::3])
y[8:-1] = y[9:]
x[x==0] = -1
y[y==0] = -1
coord = np.concatenate([x[:,None], y[:,None]], -1)
pose = pose_utils.cords_to_map(coord, target_size, orig_size)
pose = np.transpose(pose,(2, 0, 1))
pose = torch.Tensor(pose)
return pose[:18]
Djvnit
can you provide the datails of getting your dataset's pose keypoints?