mmpose/projects/skps/custom_codecs/skps_heatmap.py

165 lines
5.5 KiB
Python

# Copyright (c) OpenMMLab. All rights reserved.
from typing import Optional, Tuple
import numpy as np
from mmpose.codecs.base import BaseKeypointCodec
from mmpose.codecs.utils.gaussian_heatmap import \
generate_unbiased_gaussian_heatmaps
from mmpose.codecs.utils.post_processing import get_heatmap_maximum
from mmpose.registry import KEYPOINT_CODECS
@KEYPOINT_CODECS.register_module()
class SKPSHeatmap(BaseKeypointCodec):
"""Generate heatmap the same with MSRAHeatmap, and produce offset map
within x and y directions.
Note:
- instance number: N
- keypoint number: K
- keypoint dimension: D
- image size: [w, h]
- heatmap size: [W, H]
- offset_map size: [W, H]
Encoded:
- heatmaps (np.ndarray): The generated heatmap in shape (K, H, W)
where [W, H] is the `heatmap_size`
- offset_maps (np.ndarray): The generated offset map in x and y
direction in shape (2K, H, W) where [W, H] is the
`offset_map_size`
- keypoint_weights (np.ndarray): The target weights in shape (N, K)
Args:
input_size (tuple): Image size in [w, h]
heatmap_size (tuple): Heatmap size in [W, H]
sigma (float): The sigma value of the Gaussian heatmap
"""
def __init__(self, input_size: Tuple[int, int],
heatmap_size: Tuple[int, int], sigma: float) -> None:
super().__init__()
self.input_size = input_size
self.heatmap_size = heatmap_size
self.sigma = sigma
self.scale_factor = (np.array(input_size) /
heatmap_size).astype(np.float32)
self.y_range, self.x_range = np.meshgrid(
np.arange(0, self.heatmap_size[1]),
np.arange(0, self.heatmap_size[0]),
indexing='ij')
def encode(self,
keypoints: np.ndarray,
keypoints_visible: Optional[np.ndarray] = None) -> dict:
"""Encode keypoints into heatmaps. Note that the original keypoint
coordinates should be in the input image space.
Args:
keypoints (np.ndarray): Keypoint coordinates in shape (N, K, D)
keypoints_visible (np.ndarray): Keypoint visibilities in shape
(N, K)
Returns:
dict:
- heatmaps (np.ndarray): The generated heatmap in shape
(K, H, W) where [W, H] is the `heatmap_size`
- offset_maps (np.ndarray): The generated offset maps in x and y
directions in shape (2*K, H, W) where [W, H] is the
`offset_map_size`
- keypoint_weights (np.ndarray): The target weights in shape
(N, K)
"""
assert keypoints.shape[0] == 1, (
f'{self.__class__.__name__} only support single-instance '
'keypoint encoding')
if keypoints_visible is None:
keypoints_visible = np.ones(keypoints.shape[:2], dtype=np.float32)
heatmaps, keypoint_weights = generate_unbiased_gaussian_heatmaps(
heatmap_size=self.heatmap_size,
keypoints=keypoints / self.scale_factor,
keypoints_visible=keypoints_visible,
sigma=self.sigma)
offset_maps = self.generate_offset_map(
heatmap_size=self.heatmap_size,
keypoints=keypoints / self.scale_factor,
)
encoded = dict(
heatmaps=heatmaps,
keypoint_weights=keypoint_weights[0],
displacements=offset_maps)
return encoded
def generate_offset_map(self, heatmap_size: Tuple[int, int],
keypoints: np.ndarray):
N, K, _ = keypoints.shape
# batchsize 1
keypoints = keypoints[0]
# caution: there will be a broadcast which produce
# offside_x and offside_y with shape 64x64x98
offset_x = keypoints[:, 0] - np.expand_dims(self.x_range, axis=-1)
offset_y = keypoints[:, 1] - np.expand_dims(self.y_range, axis=-1)
offset_map = np.concatenate([offset_x, offset_y], axis=-1)
offset_map = np.transpose(offset_map, axes=[2, 0, 1])
return offset_map
def decode(self, encoded: np.ndarray,
offset_maps: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
"""Decode keypoint coordinates from heatmaps. The decoded keypoint
coordinates are in the input image space.
Args:
encoded (np.ndarray): Heatmaps in shape (K, H, W)
Returns:
tuple:
- keypoints (np.ndarray): Decoded keypoint coordinates in shape
(N, K, D)
- scores (np.ndarray): The keypoint scores in shape (N, K). It
usually represents the confidence of the keypoint prediction
"""
heatmaps = encoded.copy()
offset_maps = offset_maps.copy()
K, H, W = heatmaps.shape
keypoints, scores = get_heatmap_maximum(heatmaps)
offset_x = offset_maps[:K, ...]
offset_y = offset_maps[K:, ...]
keypoints_interger = keypoints.astype(np.int32)
keypoints_decimal = np.zeros_like(keypoints)
for i in range(K):
[x, y] = keypoints_interger[i]
if x < 0 or y < 0:
x = y = 0
# caution: torch tensor shape is nchw, so index should be i,y,x
keypoints_decimal[i][0] = x + offset_x[i, y, x]
keypoints_decimal[i][1] = y + offset_y[i, y, x]
# Restore the keypoint scale
keypoints_decimal = keypoints_decimal * self.scale_factor
return keypoints_decimal[None], scores[None]