Spaces:
Paused
Paused
File size: 13,905 Bytes
5c32cd0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 |
# Openpose
# Original from CMU https://github.com/CMU-Perceptual-Computing-Lab/openpose
# 2nd Edited by https://github.com/Hzzone/pytorch-openpose
# 3rd Edited by ControlNet
# 4th Edited by ControlNet (added face and correct hands)
# 5th Edited by ControlNet (Improved JSON serialization/deserialization, and lots of bug fixs)
# This preprocessor is licensed by CMU for non-commercial use only.
import os
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
import json
import torch
import numpy as np
from . import util
from .body import Body, BodyResult, Keypoint
from .hand import Hand
from .face import Face
from .types import PoseResult, HandResult, FaceResult
from modules import devices
from annotator.annotator_path import models_path
from typing import Tuple, List, Callable, Union, Optional
body_model_path = "https://huggingface.co/lllyasviel/Annotators/resolve/main/body_pose_model.pth"
hand_model_path = "https://huggingface.co/lllyasviel/Annotators/resolve/main/hand_pose_model.pth"
face_model_path = "https://huggingface.co/lllyasviel/Annotators/resolve/main/facenet.pth"
remote_onnx_det = "https://huggingface.co/yzd-v/DWPose/resolve/main/yolox_l.onnx"
remote_onnx_pose = "https://huggingface.co/yzd-v/DWPose/resolve/main/dw-ll_ucoco_384.onnx"
def draw_poses(poses: List[PoseResult], H, W, draw_body=True, draw_hand=True, draw_face=True):
"""
Draw the detected poses on an empty canvas.
Args:
poses (List[PoseResult]): A list of PoseResult objects containing the detected poses.
H (int): The height of the canvas.
W (int): The width of the canvas.
draw_body (bool, optional): Whether to draw body keypoints. Defaults to True.
draw_hand (bool, optional): Whether to draw hand keypoints. Defaults to True.
draw_face (bool, optional): Whether to draw face keypoints. Defaults to True.
Returns:
numpy.ndarray: A 3D numpy array representing the canvas with the drawn poses.
"""
canvas = np.zeros(shape=(H, W, 3), dtype=np.uint8)
for pose in poses:
if draw_body:
canvas = util.draw_bodypose(canvas, pose.body.keypoints)
if draw_hand:
canvas = util.draw_handpose(canvas, pose.left_hand)
canvas = util.draw_handpose(canvas, pose.right_hand)
if draw_face:
canvas = util.draw_facepose(canvas, pose.face)
return canvas
def decode_json_as_poses(json_string: str, normalize_coords: bool = False) -> Tuple[List[PoseResult], int, int]:
""" Decode the json_string complying with the openpose JSON output format
to poses that controlnet recognizes.
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/02_output.md
Args:
json_string: The json string to decode.
normalize_coords: Whether to normalize coordinates of each keypoint by canvas height/width.
`draw_pose` only accepts normalized keypoints. Set this param to True if
the input coords are not normalized.
Returns:
poses
canvas_height
canvas_width
"""
pose_json = json.loads(json_string)
height = pose_json['canvas_height']
width = pose_json['canvas_width']
def chunks(lst, n):
"""Yield successive n-sized chunks from lst."""
for i in range(0, len(lst), n):
yield lst[i:i + n]
def decompress_keypoints(numbers: Optional[List[float]]) -> Optional[List[Optional[Keypoint]]]:
if not numbers:
return None
assert len(numbers) % 3 == 0
def create_keypoint(x, y, c):
if c < 1.0:
return None
keypoint = Keypoint(x, y)
return keypoint
return [
create_keypoint(x, y, c)
for x, y, c in chunks(numbers, n=3)
]
return (
[
PoseResult(
body=BodyResult(keypoints=decompress_keypoints(pose.get('pose_keypoints_2d'))),
left_hand=decompress_keypoints(pose.get('hand_left_keypoints_2d')),
right_hand=decompress_keypoints(pose.get('hand_right_keypoints_2d')),
face=decompress_keypoints(pose.get('face_keypoints_2d'))
)
for pose in pose_json['people']
],
height,
width,
)
def encode_poses_as_json(poses: List[PoseResult], canvas_height: int, canvas_width: int) -> dict:
""" Encode the pose as a JSON compatible dict following openpose JSON output format:
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/02_output.md
"""
def compress_keypoints(keypoints: Union[List[Keypoint], None]) -> Union[List[float], None]:
if not keypoints:
return None
return [
value
for keypoint in keypoints
for value in (
[float(keypoint.x), float(keypoint.y), 1.0]
if keypoint is not None
else [0.0, 0.0, 0.0]
)
]
return {
'people': [
{
'pose_keypoints_2d': compress_keypoints(pose.body.keypoints),
"face_keypoints_2d": compress_keypoints(pose.face),
"hand_left_keypoints_2d": compress_keypoints(pose.left_hand),
"hand_right_keypoints_2d":compress_keypoints(pose.right_hand),
}
for pose in poses
],
'canvas_height': canvas_height,
'canvas_width': canvas_width,
}
class OpenposeDetector:
"""
A class for detecting human poses in images using the Openpose model.
Attributes:
model_dir (str): Path to the directory where the pose models are stored.
"""
model_dir = os.path.join(models_path, "openpose")
def __init__(self):
self.device = devices.get_device_for("controlnet")
self.body_estimation = None
self.hand_estimation = None
self.face_estimation = None
self.dw_pose_estimation = None
def load_model(self):
"""
Load the Openpose body, hand, and face models.
"""
body_modelpath = os.path.join(self.model_dir, "body_pose_model.pth")
hand_modelpath = os.path.join(self.model_dir, "hand_pose_model.pth")
face_modelpath = os.path.join(self.model_dir, "facenet.pth")
if not os.path.exists(body_modelpath):
from basicsr.utils.download_util import load_file_from_url
load_file_from_url(body_model_path, model_dir=self.model_dir)
if not os.path.exists(hand_modelpath):
from basicsr.utils.download_util import load_file_from_url
load_file_from_url(hand_model_path, model_dir=self.model_dir)
if not os.path.exists(face_modelpath):
from basicsr.utils.download_util import load_file_from_url
load_file_from_url(face_model_path, model_dir=self.model_dir)
self.body_estimation = Body(body_modelpath)
self.hand_estimation = Hand(hand_modelpath)
self.face_estimation = Face(face_modelpath)
def load_dw_model(self):
from .wholebody import Wholebody # DW Pose
def load_model(filename: str, remote_url: str):
local_path = os.path.join(self.model_dir, filename)
if not os.path.exists(local_path):
from basicsr.utils.download_util import load_file_from_url
load_file_from_url(remote_url, model_dir=self.model_dir)
return local_path
onnx_det = load_model("yolox_l.onnx", remote_onnx_det)
onnx_pose = load_model("dw-ll_ucoco_384.onnx", remote_onnx_pose)
self.dw_pose_estimation = Wholebody(onnx_det, onnx_pose)
def unload_model(self):
"""
Unload the Openpose models by moving them to the CPU.
Note: DW Pose models always run on CPU, so no need to `unload` them.
"""
if self.body_estimation is not None:
self.body_estimation.model.to("cpu")
self.hand_estimation.model.to("cpu")
self.face_estimation.model.to("cpu")
def detect_hands(self, body: BodyResult, oriImg) -> Tuple[Union[HandResult, None], Union[HandResult, None]]:
left_hand = None
right_hand = None
H, W, _ = oriImg.shape
for x, y, w, is_left in util.handDetect(body, oriImg):
peaks = self.hand_estimation(oriImg[y:y+w, x:x+w, :]).astype(np.float32)
if peaks.ndim == 2 and peaks.shape[1] == 2:
peaks[:, 0] = np.where(peaks[:, 0] < 1e-6, -1, peaks[:, 0] + x) / float(W)
peaks[:, 1] = np.where(peaks[:, 1] < 1e-6, -1, peaks[:, 1] + y) / float(H)
hand_result = [
Keypoint(x=peak[0], y=peak[1])
for peak in peaks
]
if is_left:
left_hand = hand_result
else:
right_hand = hand_result
return left_hand, right_hand
def detect_face(self, body: BodyResult, oriImg) -> Union[FaceResult, None]:
face = util.faceDetect(body, oriImg)
if face is None:
return None
x, y, w = face
H, W, _ = oriImg.shape
heatmaps = self.face_estimation(oriImg[y:y+w, x:x+w, :])
peaks = self.face_estimation.compute_peaks_from_heatmaps(heatmaps).astype(np.float32)
if peaks.ndim == 2 and peaks.shape[1] == 2:
peaks[:, 0] = np.where(peaks[:, 0] < 1e-6, -1, peaks[:, 0] + x) / float(W)
peaks[:, 1] = np.where(peaks[:, 1] < 1e-6, -1, peaks[:, 1] + y) / float(H)
return [
Keypoint(x=peak[0], y=peak[1])
for peak in peaks
]
return None
def detect_poses(self, oriImg, include_hand=False, include_face=False) -> List[PoseResult]:
"""
Detect poses in the given image.
Args:
oriImg (numpy.ndarray): The input image for pose detection.
include_hand (bool, optional): Whether to include hand detection. Defaults to False.
include_face (bool, optional): Whether to include face detection. Defaults to False.
Returns:
List[PoseResult]: A list of PoseResult objects containing the detected poses.
"""
if self.body_estimation is None:
self.load_model()
self.body_estimation.model.to(self.device)
self.hand_estimation.model.to(self.device)
self.face_estimation.model.to(self.device)
self.body_estimation.cn_device = self.device
self.hand_estimation.cn_device = self.device
self.face_estimation.cn_device = self.device
oriImg = oriImg[:, :, ::-1].copy()
H, W, C = oriImg.shape
with torch.no_grad():
candidate, subset = self.body_estimation(oriImg)
bodies = self.body_estimation.format_body_result(candidate, subset)
results = []
for body in bodies:
left_hand, right_hand, face = (None,) * 3
if include_hand:
left_hand, right_hand = self.detect_hands(body, oriImg)
if include_face:
face = self.detect_face(body, oriImg)
results.append(PoseResult(BodyResult(
keypoints=[
Keypoint(
x=keypoint.x / float(W),
y=keypoint.y / float(H)
) if keypoint is not None else None
for keypoint in body.keypoints
],
total_score=body.total_score,
total_parts=body.total_parts
), left_hand, right_hand, face))
return results
def detect_poses_dw(self, oriImg) -> List[PoseResult]:
"""
Detect poses in the given image using DW Pose:
https://github.com/IDEA-Research/DWPose
Args:
oriImg (numpy.ndarray): The input image for pose detection.
Returns:
List[PoseResult]: A list of PoseResult objects containing the detected poses.
"""
from .wholebody import Wholebody # DW Pose
self.load_dw_model()
with torch.no_grad():
keypoints_info = self.dw_pose_estimation(oriImg.copy())
return Wholebody.format_result(keypoints_info)
def __call__(
self, oriImg, include_body=True, include_hand=False, include_face=False,
use_dw_pose=False, json_pose_callback: Callable[[str], None] = None,
):
"""
Detect and draw poses in the given image.
Args:
oriImg (numpy.ndarray): The input image for pose detection and drawing.
include_body (bool, optional): Whether to include body keypoints. Defaults to True.
include_hand (bool, optional): Whether to include hand keypoints. Defaults to False.
include_face (bool, optional): Whether to include face keypoints. Defaults to False.
use_dw_pose (bool, optional): Whether to use DW pose detection algorithm. Defaults to False.
json_pose_callback (Callable, optional): A callback that accepts the pose JSON string.
Returns:
numpy.ndarray: The image with detected and drawn poses.
"""
H, W, _ = oriImg.shape
if use_dw_pose:
poses = self.detect_poses_dw(oriImg)
else:
poses = self.detect_poses(oriImg, include_hand, include_face)
if json_pose_callback:
json_pose_callback(encode_poses_as_json(poses, H, W))
return draw_poses(poses, H, W, draw_body=include_body, draw_hand=include_hand, draw_face=include_face)
|