Spaces:
Sleeping
Sleeping
File size: 9,710 Bytes
4853e68 | 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 | """Data preprocessing with MediaPipe"""
import cv2
import numpy as np
try:
import mediapipe as mp
MEDIAPIPE_AVAILABLE = True
except ImportError:
MEDIAPIPE_AVAILABLE = False
mp = None
from typing import Optional, Dict, List, Tuple
from pathlib import Path
import logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class MediaPipePreprocessor:
"""Extract landmarks from videos using MediaPipe Holistic"""
def __init__(self,
min_detection_confidence: float = 0.5,
min_tracking_confidence: float = 0.5):
"""
Initialize MediaPipe Holistic
Args:
min_detection_confidence: Minimum confidence for detection
min_tracking_confidence: Minimum confidence for tracking
"""
self.min_detection_confidence = min_detection_confidence
self.min_tracking_confidence = min_tracking_confidence
self.holistic = None
self.mp_holistic = None
# Landmark dimensions
self.pose_dim = 33 * 4 # 33 landmarks * (x, y, z, visibility)
self.hand_dim = 21 * 3 # 21 landmarks * (x, y, z)
self.face_dim = 468 * 3 # 468 landmarks * (x, y, z)
self.total_dim = self.pose_dim + 2 * self.hand_dim + self.face_dim
# Initialize MediaPipe
self._initialize_mediapipe()
def _initialize_mediapipe(self):
"""Initialize or reinitialize MediaPipe"""
if not MEDIAPIPE_AVAILABLE or mp is None:
logger.warning("MediaPipe not available - using dummy preprocessor")
return
try:
# Close existing instance if any
if self.holistic is not None:
try:
self.holistic.close()
except:
pass
# Create new instance
self.mp_holistic = mp.solutions.holistic
self.holistic = self.mp_holistic.Holistic(
min_detection_confidence=self.min_detection_confidence,
min_tracking_confidence=self.min_tracking_confidence,
model_complexity=1,
static_image_mode=False,
smooth_landmarks=True
)
logger.info("MediaPipe initialized successfully")
except Exception as e:
logger.error(f"Failed to initialize MediaPipe: {e}")
self.holistic = None
self.mp_holistic = None
raise RuntimeError(f"MediaPipe initialization failed: {e}")
def extract_landmarks_from_frame(self, frame: np.ndarray) -> Optional[np.ndarray]:
"""
Extract landmarks from a single frame
Args:
frame: RGB image frame
Returns:
Flattened landmark array of shape (total_dim,) or None if detection fails
"""
if self.holistic is None:
logger.error("MediaPipe not initialized! Call _initialize_mediapipe() first")
raise RuntimeError("MediaPipe not initialized. Please check MediaPipe installation.")
# Convert BGR to RGB if needed
if len(frame.shape) == 3 and frame.shape[2] == 3:
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
else:
frame_rgb = frame
# Process frame
results = self.holistic.process(frame_rgb)
# Extract landmarks
landmarks = []
# Pose landmarks (33 * 4 = 132)
if results.pose_landmarks:
pose = np.array([[lm.x, lm.y, lm.z, lm.visibility]
for lm in results.pose_landmarks.landmark]).flatten()
else:
pose = np.zeros(self.pose_dim)
landmarks.append(pose)
# Left hand landmarks (21 * 3 = 63)
if results.left_hand_landmarks:
left_hand = np.array([[lm.x, lm.y, lm.z]
for lm in results.left_hand_landmarks.landmark]).flatten()
else:
left_hand = np.zeros(self.hand_dim)
landmarks.append(left_hand)
# Right hand landmarks (21 * 3 = 63)
if results.right_hand_landmarks:
right_hand = np.array([[lm.x, lm.y, lm.z]
for lm in results.right_hand_landmarks.landmark]).flatten()
else:
right_hand = np.zeros(self.hand_dim)
landmarks.append(right_hand)
# Face landmarks (468 * 3 = 1404) - optional, can be excluded for efficiency
# Uncomment if you want to include face landmarks
# if results.face_landmarks:
# face = np.array([[lm.x, lm.y, lm.z]
# for lm in results.face_landmarks.landmark]).flatten()
# else:
# face = np.zeros(self.face_dim)
# landmarks.append(face)
# Concatenate all landmarks
landmarks_array = np.concatenate(landmarks)
return landmarks_array
def extract_landmarks_from_video(self,
video_path: str,
max_frames: int = 64,
target_fps: Optional[int] = None) -> Optional[np.ndarray]:
"""
Extract landmarks from video file
Args:
video_path: Path to video file
max_frames: Maximum number of frames to extract
target_fps: Target FPS for frame sampling (None = use all frames)
Returns:
Landmark sequence of shape (num_frames, total_dim) or None if failed
"""
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
logger.error(f"Failed to open video: {video_path}")
return None
# Get video properties
fps = cap.get(cv2.CAP_PROP_FPS)
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
# Calculate frame sampling rate
if target_fps and target_fps < fps:
frame_skip = int(fps / target_fps)
else:
frame_skip = 1
landmarks_sequence = []
frame_idx = 0
while len(landmarks_sequence) < max_frames:
ret, frame = cap.read()
if not ret:
break
# Sample frames
if frame_idx % frame_skip == 0:
landmarks = self.extract_landmarks_from_frame(frame)
if landmarks is not None:
landmarks_sequence.append(landmarks)
frame_idx += 1
cap.release()
if not landmarks_sequence:
logger.warning(f"No landmarks extracted from: {video_path}")
return None
# Convert to numpy array
landmarks_array = np.array(landmarks_sequence)
# Pad or truncate to max_frames
if len(landmarks_array) < max_frames:
# Pad with zeros
padding = np.zeros((max_frames - len(landmarks_array), landmarks_array.shape[1]))
landmarks_array = np.vstack([landmarks_array, padding])
else:
# Truncate
landmarks_array = landmarks_array[:max_frames]
return landmarks_array
def normalize_landmarks(self, landmarks: np.ndarray) -> np.ndarray:
"""
Normalize landmarks to zero mean and unit variance
Args:
landmarks: Landmark array of shape (num_frames, total_dim)
Returns:
Normalized landmarks
"""
# Calculate mean and std (excluding zero-padded frames)
non_zero_mask = np.any(landmarks != 0, axis=1)
if np.sum(non_zero_mask) > 0:
mean = landmarks[non_zero_mask].mean(axis=0)
std = landmarks[non_zero_mask].std(axis=0) + 1e-8
# Normalize
landmarks_normalized = landmarks.copy()
landmarks_normalized[non_zero_mask] = (landmarks[non_zero_mask] - mean) / std
return landmarks_normalized
return landmarks
def close(self):
"""Explicitly close MediaPipe resources"""
if self.holistic is not None:
try:
self.holistic.close()
logger.info("MediaPipe closed successfully")
except Exception as e:
logger.warning(f"Error closing MediaPipe: {e}")
finally:
self.holistic = None
def __del__(self):
"""Cleanup"""
self.close()
def preprocess_video(video_path: str,
max_frames: int = 64,
normalize: bool = True) -> Optional[np.ndarray]:
"""
Convenience function to preprocess a single video
Args:
video_path: Path to video file
max_frames: Maximum number of frames
normalize: Whether to normalize landmarks
Returns:
Preprocessed landmarks array
"""
preprocessor = MediaPipePreprocessor()
landmarks = preprocessor.extract_landmarks_from_video(video_path, max_frames)
if landmarks is not None and normalize:
landmarks = preprocessor.normalize_landmarks(landmarks)
return landmarks
|