PCKAccuracy¶
- class mmeval.metrics.PCKAccuracy(thr: float = 0.2, norm_item: Union[str, Sequence[str]] = 'bbox', **kwargs)[源代码]¶
PCK accuracy evaluation metric, which is widely used in pose estimation.
Calculate the pose accuracy of Percentage of Correct Keypoints (PCK) for each individual keypoint and the averaged accuracy across all keypoints. PCK metric measures the accuracy of the localization of the body joints. The distances between predicted positions and the ground-truth ones are typically normalized by the person bounding box size. The threshold (thr) of the normalized distance is commonly set as 0.05, 0.1 or 0.2 etc.
注解
length of dataset: N
num_keypoints: K
number of keypoint dimensions: D (typically D = 2)
- 参数
thr (float) – Threshold of PCK calculation. Defaults to 0.2.
norm_item (str | Sequence[str]) – The item used for normalization. Valid items include ‘bbox’, ‘head’, ‘torso’, which correspond to ‘PCK’, ‘PCKh’ and ‘tPCK’ respectively. Defaults to
'bbox'
.**kwargs – Keyword parameters passed to
BaseMetric
.
实际案例
>>> from mmeval import PCKAccuracy >>> import numpy as np >>> num_keypoints = 15 >>> keypoints = np.random.random((1, num_keypoints, 2)) * 10 >>> predictions = [{'coords': keypoints}] >>> keypoints_visible = np.ones((1, num_keypoints)).astype(bool) >>> bbox_size = np.random.random((1, 2)) * 10 >>> groundtruths = [{ ... 'coords': keypoints, ... 'mask': keypoints_visible, ... 'bbox_size': bbox_size, ... }] >>> pck_metric = PCKAccuracy(thr=0.5, norm_item='bbox') >>> pck_metric(predictions, groundtruths) OrderedDict([('PCK@0.5', 1.0)])
- add(predictions: List[Dict], groundtruths: List[Dict]) → None[源代码]¶
Process one batch of predictions and groundtruths and add the intermediate results to self._results.
- 参数
predictions (Sequence[dict]) –
Predictions from the model. Each prediction dict has the following keys:
coords (np.ndarray, [1, K, D]): predicted keypoints coordinates
groundtruths (Sequence[dict]) –
The ground truth labels. Each groundtruth dict has the following keys:
coords (np.ndarray, [1, K, D]): ground truth keypoints coordinates
mask (np.ndarray, [1, K]): ground truth keypoints_visible
bbox_size (np.ndarray, optional, [1, 2]): ground truth bbox size
head_size (np.ndarray, optional, [1, 2]): ground truth head size
torso_size (np.ndarray, optional, [1, 2]): ground truth torso size