feat: update Python processors and add utility scripts
- Update ASR, face, OCR, pose processors - Add release pre-flight check script - Add synonym generation, chunk processing scripts - Add face recognition, stamp search utilities
This commit is contained in:
Warren
376
scripts/pose_processor_mps.py
Normal file
376
scripts/pose_processor_mps.py
Normal file
@@ -0,0 +1,376 @@
|
||||
#!/opt/homebrew/bin/python3.11
|
||||
"""
|
||||
Pose Processor - Apple MPS Optimized Version
|
||||
Uses YOLOv8 Pose with Apple Silicon MPS acceleration
|
||||
|
||||
Features:
|
||||
- Automatic MPS/CPU fallback
|
||||
- Metal GPU acceleration for inference
|
||||
- YOLOv8 Pose model support
|
||||
- Memory-optimized for unified memory architecture
|
||||
"""
|
||||
|
||||
import sys
|
||||
import json
|
||||
import argparse
|
||||
import os
|
||||
import signal
|
||||
import time
|
||||
from datetime import datetime
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
|
||||
import cv2
|
||||
import numpy as np
|
||||
import torch
|
||||
from ultralytics import YOLO
|
||||
|
||||
|
||||
# COCO keypoint names (17 keypoints)
|
||||
KEYPOINT_NAMES = [
|
||||
"nose",
|
||||
"left_eye",
|
||||
"right_eye",
|
||||
"left_ear",
|
||||
"right_ear",
|
||||
"left_shoulder",
|
||||
"right_shoulder",
|
||||
"left_elbow",
|
||||
"right_elbow",
|
||||
"left_wrist",
|
||||
"right_wrist",
|
||||
"left_hip",
|
||||
"right_hip",
|
||||
"left_knee",
|
||||
"right_knee",
|
||||
"left_ankle",
|
||||
"right_ankle",
|
||||
]
|
||||
|
||||
# Keypoint connections for skeleton visualization
|
||||
KEYPOINT_CONNECTIONS = [
|
||||
("left_shoulder", "right_shoulder"),
|
||||
("left_shoulder", "left_elbow"),
|
||||
("left_elbow", "left_wrist"),
|
||||
("right_shoulder", "right_elbow"),
|
||||
("right_elbow", "right_wrist"),
|
||||
("left_shoulder", "left_hip"),
|
||||
("right_shoulder", "right_hip"),
|
||||
("left_hip", "right_hip"),
|
||||
("left_hip", "left_knee"),
|
||||
("left_knee", "left_ankle"),
|
||||
("right_hip", "right_knee"),
|
||||
("right_knee", "right_ankle"),
|
||||
]
|
||||
|
||||
|
||||
def get_device() -> str:
|
||||
"""Determine the best available device for inference"""
|
||||
if torch.backends.mps.is_available():
|
||||
return "mps"
|
||||
elif torch.cuda.is_available():
|
||||
return "cuda"
|
||||
else:
|
||||
return "cpu"
|
||||
|
||||
|
||||
def signal_handler(signum, frame):
|
||||
"""Handle interrupt signals gracefully"""
|
||||
print(f"\n[Pose] Received signal {signum}, saving results and exiting...")
|
||||
sys.exit(0)
|
||||
|
||||
|
||||
def process_video_pose(
|
||||
video_path: str,
|
||||
output_path: str,
|
||||
model_name: str = "yolov8n-pose",
|
||||
confidence: float = 0.5,
|
||||
device: str = "auto",
|
||||
sample_interval: int = 30,
|
||||
resume: bool = True,
|
||||
save_interval: int = 30,
|
||||
) -> Dict:
|
||||
"""
|
||||
Process video for pose estimation with MPS acceleration
|
||||
|
||||
Args:
|
||||
video_path: Path to input video file
|
||||
output_path: Path to output JSON file
|
||||
model_name: YOLO Pose model name (yolov8n-pose/s/m/l/x)
|
||||
confidence: Confidence threshold for keypoints
|
||||
device: Device to use ('auto', 'mps', 'cuda', 'cpu')
|
||||
sample_interval: Process every N frames
|
||||
resume: Whether to resume from existing results
|
||||
save_interval: Auto-save interval in seconds
|
||||
|
||||
Returns:
|
||||
Dictionary with pose estimation results and metadata
|
||||
"""
|
||||
# Set up signal handlers
|
||||
signal.signal(signal.SIGTERM, signal_handler)
|
||||
signal.signal(signal.SIGINT, signal_handler)
|
||||
|
||||
# Determine device
|
||||
if device == "auto":
|
||||
device = get_device()
|
||||
|
||||
print(f"[Pose] Starting pose estimation with device: {device}")
|
||||
print(f"[Pose] Model: {model_name}, Confidence: {confidence}")
|
||||
|
||||
# Load model
|
||||
print(f"[Pose] Loading model: {model_name}")
|
||||
model = YOLO(f"{model_name}.pt")
|
||||
|
||||
# Move to device
|
||||
if device in ["mps", "cuda"]:
|
||||
model.to(device)
|
||||
|
||||
# Get video info
|
||||
cap = cv2.VideoCapture(video_path)
|
||||
fps = cap.get(cv2.CAP_PROP_FPS)
|
||||
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
||||
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
||||
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
||||
cap.release()
|
||||
|
||||
print(f"[Pose] Video: {width}x{height} @ {fps:.2f} FPS, {total_frames} frames")
|
||||
|
||||
# Load existing data if resuming
|
||||
existing_data = None
|
||||
last_processed_frame = 0
|
||||
|
||||
if resume and os.path.exists(output_path):
|
||||
try:
|
||||
with open(output_path, "r") as f:
|
||||
existing_data = json.load(f)
|
||||
frames = existing_data.get("frames", {})
|
||||
if frames:
|
||||
last_processed_frame = max(int(k) for k in frames.keys())
|
||||
print(f"[Pose] Resuming from frame {last_processed_frame}")
|
||||
except (json.JSONDecodeError, KeyError):
|
||||
pass
|
||||
|
||||
# Initialize result structure
|
||||
result = {
|
||||
"video_path": video_path,
|
||||
"model": model_name,
|
||||
"device": device,
|
||||
"confidence_threshold": confidence,
|
||||
"processed_at": datetime.now().isoformat(),
|
||||
"keypoint_names": KEYPOINT_NAMES,
|
||||
"connections": KEYPOINT_CONNECTIONS,
|
||||
"frames": {},
|
||||
}
|
||||
|
||||
if existing_data:
|
||||
result["frames"] = existing_data.get("frames", {})
|
||||
|
||||
# Process video
|
||||
print(f"[Pose] Processing video: {video_path}")
|
||||
start_time = time.time()
|
||||
|
||||
frame_count = 0
|
||||
pose_count = 0
|
||||
last_save_time = start_time
|
||||
|
||||
try:
|
||||
# Use stream mode for memory efficiency
|
||||
results = model(
|
||||
video_path,
|
||||
conf=confidence,
|
||||
device=device,
|
||||
stream=True,
|
||||
imgsz=640,
|
||||
pose=True,
|
||||
verbose=False,
|
||||
)
|
||||
|
||||
for idx, r in enumerate(results):
|
||||
# Skip frames based on sample_interval
|
||||
if idx % sample_interval != 0:
|
||||
continue
|
||||
|
||||
# Get pose results
|
||||
keypoints = r.keypoints
|
||||
|
||||
if keypoints is not None and len(keypoints) > 0:
|
||||
# Get keypoint data
|
||||
kp_data = keypoints.data.cpu().numpy()
|
||||
|
||||
frame_poses = []
|
||||
|
||||
for person_idx in range(len(keypoints)):
|
||||
person_keypoints = []
|
||||
|
||||
for kp_idx in range(min(17, len(kp_data[person_idx]))):
|
||||
kp = kp_data[person_idx][kp_idx]
|
||||
|
||||
# Keypoint: [x, y, confidence]
|
||||
if len(kp) >= 3 and kp[2] > confidence:
|
||||
person_keypoints.append(
|
||||
{
|
||||
"name": KEYPOINT_NAMES[kp_idx]
|
||||
if kp_idx < len(KEYPOINT_NAMES)
|
||||
else f"kp_{kp_idx}",
|
||||
"x": float(kp[0]),
|
||||
"y": float(kp[1]),
|
||||
"confidence": float(kp[2]),
|
||||
}
|
||||
)
|
||||
|
||||
if person_keypoints:
|
||||
frame_poses.append(
|
||||
{
|
||||
"keypoints": person_keypoints,
|
||||
"person_id": person_idx,
|
||||
}
|
||||
)
|
||||
pose_count += 1
|
||||
|
||||
if frame_poses:
|
||||
result["frames"][str(idx)] = {
|
||||
"timestamp": idx / fps if fps > 0 else 0,
|
||||
"poses": frame_poses,
|
||||
}
|
||||
|
||||
frame_count += 1
|
||||
|
||||
# Progress reporting
|
||||
if frame_count % 100 == 0:
|
||||
elapsed = time.time() - start_time
|
||||
fps_rate = frame_count / elapsed if elapsed > 0 else 0
|
||||
print(
|
||||
f"[Pose] Processed {frame_count} frames, {pose_count} poses, {fps_rate:.1f} FPS"
|
||||
)
|
||||
|
||||
# Periodic save
|
||||
if save_interval > 0 and time.time() - last_save_time > save_interval:
|
||||
with open(output_path, "w") as f:
|
||||
json.dump(result, f, indent=2)
|
||||
last_save_time = time.time()
|
||||
print(f"[Pose] Auto-saved at frame {frame_count}")
|
||||
|
||||
except Exception as e:
|
||||
print(f"[Pose] Error during processing: {e}")
|
||||
raise
|
||||
|
||||
# Final save
|
||||
elapsed_time = time.time() - start_time
|
||||
avg_fps = frame_count / elapsed_time if elapsed_time > 0 else 0
|
||||
|
||||
result["summary"] = {
|
||||
"total_frames": frame_count,
|
||||
"total_poses": pose_count,
|
||||
"processing_time": round(elapsed_time, 2),
|
||||
"average_fps": round(avg_fps, 2),
|
||||
"model": model_name,
|
||||
"device": device,
|
||||
}
|
||||
|
||||
# Save final results
|
||||
with open(output_path, "w") as f:
|
||||
json.dump(result, f, indent=2)
|
||||
|
||||
print(
|
||||
f"[Pose] Completed: {frame_count} frames, {pose_count} poses in {elapsed_time:.1f}s ({avg_fps:.1f} FPS)"
|
||||
)
|
||||
print(f"[Pose] Results saved to: {output_path}")
|
||||
|
||||
return result
|
||||
|
||||
|
||||
def benchmark_pose_models(video_path: str, num_frames: int = 100) -> Dict:
|
||||
"""Benchmark different YOLO Pose models and devices"""
|
||||
devices = ["cpu"]
|
||||
if torch.backends.mps.is_available():
|
||||
devices.append("mps")
|
||||
if torch.cuda.is_available():
|
||||
devices.append("cuda")
|
||||
|
||||
models = ["yolov8n-pose", "yolov8s-pose"]
|
||||
results = {}
|
||||
|
||||
for model_name in models:
|
||||
for device in devices:
|
||||
print(f"[Pose] Benchmarking {model_name} on {device}...")
|
||||
|
||||
model = YOLO(f"{model_name}.pt")
|
||||
if device != "cpu":
|
||||
model.to(device)
|
||||
|
||||
start_time = time.time()
|
||||
count = 0
|
||||
|
||||
try:
|
||||
for idx, r in enumerate(
|
||||
model(video_path, device=device, stream=True, imgsz=320, pose=True)
|
||||
):
|
||||
if idx >= num_frames:
|
||||
break
|
||||
count += 1
|
||||
except Exception as e:
|
||||
print(f"[Pose] Error: {e}")
|
||||
continue
|
||||
|
||||
elapsed = time.time() - start_time
|
||||
fps = count / elapsed if elapsed > 0 else 0
|
||||
|
||||
key = f"{model_name}_{device}"
|
||||
results[key] = {
|
||||
"frames": count,
|
||||
"time": round(elapsed, 2),
|
||||
"fps": round(fps, 2),
|
||||
}
|
||||
|
||||
return results
|
||||
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser(description="Pose Processor with MPS Support")
|
||||
parser.add_argument("--video", required=True, help="Input video path")
|
||||
parser.add_argument("--output", required=True, help="Output JSON path")
|
||||
parser.add_argument(
|
||||
"--model", default="yolov8n-pose", help="YOLO Pose model (yolov8n-pose/s/m/l/x)"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--confidence", type=float, default=0.5, help="Confidence threshold"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--device",
|
||||
default="auto",
|
||||
choices=["auto", "mps", "cuda", "cpu"],
|
||||
help="Device to use",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--sample-interval", type=int, default=30, help="Process every N frames"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--no-resume", action="store_true", help="Do not resume from existing results"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--save-interval", type=int, default=30, help="Auto-save interval in seconds"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--benchmark", action="store_true", help="Run benchmark instead of processing"
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
if args.benchmark:
|
||||
results = benchmark_pose_models(args.video)
|
||||
print("\n[Benchmark Results]")
|
||||
print(json.dumps(results, indent=2))
|
||||
else:
|
||||
process_video_pose(
|
||||
video_path=args.video,
|
||||
output_path=args.output,
|
||||
model_name=args.model,
|
||||
confidence=args.confidence,
|
||||
device=args.device,
|
||||
sample_interval=args.sample_interval,
|
||||
resume=not args.no_resume,
|
||||
save_interval=args.save_interval,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
Reference in New Issue
Block a user