File size: 13,229 Bytes
efb1801 |
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 |
#!/usr/bin/env python3
"""
Benchmark YOLO models for performance on Raspberry Pi 4B (or current machine).
Measures inference time, FPS, and memory usage for different model formats.
"""
import argparse
import time
import os
import sys
from pathlib import Path
import numpy as np
import cv2
import yaml
from ultralytics import YOLO
import psutil
import platform
def get_system_info():
"""Get system information for benchmarking context."""
info = {
'system': platform.system(),
'processor': platform.processor(),
'architecture': platform.architecture()[0],
'python_version': platform.python_version(),
'cpu_count': psutil.cpu_count(logical=False),
'memory_gb': psutil.virtual_memory().total / (1024**3),
}
return info
def load_test_images(dataset_path, max_images=50):
"""Load test images from dataset for benchmarking."""
test_images = []
# Try multiple possible locations
possible_paths = [
Path(dataset_path) / "test" / "images",
Path(dataset_path) / "valid" / "images",
Path(dataset_path) / "val" / "images",
Path(dataset_path) / "train" / "images",
]
for path in possible_paths:
if path.exists():
image_files = list(path.glob("*.jpg")) + list(path.glob("*.png"))
if image_files:
test_images = [str(p) for p in image_files[:max_images]]
print(f"Found {len(test_images)} images in {path}")
break
if not test_images:
# Create dummy images if no dataset found
print("No test images found. Creating dummy images for benchmarking.")
test_images = []
for i in range(10):
# Create a dummy image
dummy_img = np.random.randint(0, 255, (640, 640, 3), dtype=np.uint8)
dummy_path = f"/tmp/dummy_{i}.jpg"
cv2.imwrite(dummy_path, dummy_img)
test_images.append(dummy_path)
return test_images
def benchmark_model(model_path, test_images, img_size=640, warmup=10, runs=100):
"""
Benchmark a single model.
Args:
model_path: Path to model file (.pt, .onnx, .tflite)
test_images: List of image paths for testing
img_size: Input image size
warmup: Number of warmup runs
runs: Number of benchmark runs
Returns:
Dictionary with benchmark results
"""
print(f"\n{'='*60}")
print(f"Benchmarking: {model_path}")
print(f"{'='*60}")
results = {
'model': os.path.basename(model_path),
'format': Path(model_path).suffix[1:],
'size_mb': os.path.getsize(model_path) / (1024 * 1024) if os.path.exists(model_path) else 0,
'inference_times': [],
'memory_usage_mb': [],
'success': False
}
# Check if model exists
if not os.path.exists(model_path):
print(f" β Model not found: {model_path}")
return results
try:
# Load model
print(f" Loading model...")
start_load = time.time()
model = YOLO(model_path)
load_time = time.time() - start_load
results['load_time'] = load_time
# Warmup
print(f" Warming up ({warmup} runs)...")
for i in range(warmup):
if i >= len(test_images):
img_path = test_images[0]
else:
img_path = test_images[i]
img = cv2.imread(img_path)
if img is None:
# Create dummy image
img = np.random.randint(0, 255, (img_size, img_size, 3), dtype=np.uint8)
_ = model(img, verbose=False)
# Benchmark runs
print(f" Running benchmark ({runs} runs)...")
for i in range(runs):
# Cycle through test images
img_idx = i % len(test_images)
img_path = test_images[img_idx]
img = cv2.imread(img_path)
if img is None:
img = np.random.randint(0, 255, (img_size, img_size, 3), dtype=np.uint8)
# Measure memory before
process = psutil.Process(os.getpid())
mem_before = process.memory_info().rss / 1024 / 1024 # MB
# Inference
start_time = time.perf_counter()
results_inference = model(img, verbose=False)
inference_time = time.perf_counter() - start_time
# Measure memory after
mem_after = process.memory_info().rss / 1024 / 1024 # MB
mem_used = mem_after - mem_before
results['inference_times'].append(inference_time)
results['memory_usage_mb'].append(mem_used)
# Print progress
if (i + 1) % 20 == 0:
print(f" Completed {i+1}/{runs} runs...")
# Calculate statistics
if results['inference_times']:
times = np.array(results['inference_times'])
results['avg_inference_ms'] = np.mean(times) * 1000
results['std_inference_ms'] = np.std(times) * 1000
results['min_inference_ms'] = np.min(times) * 1000
results['max_inference_ms'] = np.max(times) * 1000
results['fps'] = 1.0 / np.mean(times)
results['avg_memory_mb'] = np.mean(results['memory_usage_mb'])
results['success'] = True
print(f" β
Benchmark completed:")
print(f" Model size: {results['size_mb']:.2f} MB")
print(f" Avg inference: {results['avg_inference_ms']:.2f} ms")
print(f" FPS: {results['fps']:.2f}")
print(f" Memory usage: {results['avg_memory_mb']:.2f} MB")
else:
print(f" β No inference times recorded")
except Exception as e:
print(f" β Error benchmarking {model_path}: {e}")
import traceback
traceback.print_exc()
return results
def benchmark_all_models(models_to_test, test_images, img_size=640):
"""Benchmark multiple models and return results."""
all_results = []
for model_info in models_to_test:
model_path = model_info['path']
if not os.path.exists(model_path):
print(f"Skipping {model_path} - not found")
continue
results = benchmark_model(
model_path=model_path,
test_images=test_images,
img_size=img_size,
warmup=10,
runs=50 # Reduced for faster benchmarking
)
results.update({
'name': model_info['name'],
'description': model_info.get('description', '')
})
all_results.append(results)
return all_results
def print_results_table(results):
"""Print benchmark results in a formatted table."""
print("\n" + "="*100)
print("BENCHMARK RESULTS")
print("="*100)
print(f"{'Model':<30} {'Format':<8} {'Size (MB)':<10} {'Inference (ms)':<15} {'FPS':<10} {'Memory (MB)':<12} {'Status':<10}")
print("-"*100)
for r in results:
if r['success']:
print(f"{r['name'][:28]:<30} {r['format']:<8} {r['size_mb']:>9.2f} "
f"{r['avg_inference_ms']:>14.2f} {r['fps']:>9.2f} {r['avg_memory_mb']:>11.2f} {'β
':<10}")
else:
print(f"{r['name'][:28]:<30} {r['format']:<8} {r['size_mb']:>9.2f} "
f"{'N/A':>14} {'N/A':>9} {'N/A':>11} {'β':<10}")
print("="*100)
# Find best model by FPS
successful = [r for r in results if r['success']]
if successful:
best_by_fps = max(successful, key=lambda x: x['fps'])
best_by_size = min(successful, key=lambda x: x['size_mb'])
best_by_memory = min(successful, key=lambda x: x['avg_memory_mb'])
print(f"\nπ Best by FPS: {best_by_fps['name']} ({best_by_fps['fps']:.2f} FPS)")
print(f"π Best by size: {best_by_size['name']} ({best_by_size['size_mb']:.2f} MB)")
print(f"π Best by memory: {best_by_memory['name']} ({best_by_memory['avg_memory_mb']:.2f} MB)")
def save_results_to_csv(results, output_path="benchmark_results.csv"):
"""Save benchmark results to CSV file."""
import csv
with open(output_path, 'w', newline='') as csvfile:
fieldnames = ['name', 'format', 'size_mb', 'avg_inference_ms',
'std_inference_ms', 'min_inference_ms', 'max_inference_ms',
'fps', 'avg_memory_mb', 'load_time', 'success']
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
for r in results:
writer.writerow({
'name': r['name'],
'format': r['format'],
'size_mb': r.get('size_mb', 0),
'avg_inference_ms': r.get('avg_inference_ms', 0),
'std_inference_ms': r.get('std_inference_ms', 0),
'min_inference_ms': r.get('min_inference_ms', 0),
'max_inference_ms': r.get('max_inference_ms', 0),
'fps': r.get('fps', 0),
'avg_memory_mb': r.get('avg_memory_mb', 0),
'load_time': r.get('load_time', 0),
'success': r['success']
})
print(f"\nπ Results saved to {output_path}")
def main():
parser = argparse.ArgumentParser(description='Benchmark YOLO models for performance')
parser.add_argument('--dataset', type=str, default='model/dataset_strawberry_detect_v3',
help='Path to dataset for test images')
parser.add_argument('--img-size', type=int, default=640,
help='Input image size for inference')
parser.add_argument('--output', type=str, default='benchmark_results.csv',
help='Output CSV file for results')
parser.add_argument('--config', type=str, default='config.yaml',
help='Path to config file')
args = parser.parse_args()
# Load config
config = {}
if os.path.exists(args.config):
with open(args.config, 'r') as f:
config = yaml.safe_load(f)
# Get system info
system_info = get_system_info()
print("="*60)
print("SYSTEM INFORMATION")
print("="*60)
for key, value in system_info.items():
print(f"{key.replace('_', ' ').title():<20}: {value}")
# Define models to test
models_to_test = [
# Base YOLO models
{'name': 'YOLOv8n', 'path': 'yolov8n.pt', 'description': 'Ultralytics YOLOv8n'},
{'name': 'YOLOv8s', 'path': 'yolov8s.pt', 'description': 'Ultralytics YOLOv8s'},
{'name': 'YOLOv8m', 'path': 'yolov8m.pt', 'description': 'Ultralytics YOLOv8m'},
# Custom trained models
{'name': 'Strawberry YOLOv11n', 'path': 'model/weights/strawberry_yolov11n.pt', 'description': 'Custom trained on strawberry dataset'},
{'name': 'Strawberry YOLOv11n ONNX', 'path': 'model/weights/strawberry_yolov11n.onnx', 'description': 'ONNX export'},
# Ripeness detection models
{'name': 'Ripeness YOLOv11n', 'path': 'model/weights/ripeness_detection_yolov11n.pt', 'description': 'Ripeness detection model'},
{'name': 'Ripeness YOLOv11n ONNX', 'path': 'model/weights/ripeness_detection_yolov11n.onnx', 'description': 'ONNX export'},
]
# Check which models exist
existing_models = []
for model in models_to_test:
if os.path.exists(model['path']):
existing_models.append(model)
else:
print(f"β οΈ Model not found: {model['path']}")
if not existing_models:
print("β No models found for benchmarking.")
print("Please train a model first or download pretrained weights.")
sys.exit(1)
# Load test images
print(f"\nπ· Loading test images from {args.dataset}...")
test_images = load_test_images(args.dataset, max_images=50)
print(f" Loaded {len(test_images)} test images")
# Run benchmarks
print(f"\nπ Starting benchmarks...")
results = benchmark_all_models(existing_models, test_images, img_size=args.img_size)
# Print results
print_results_table(results)
# Save results
save_results_to_csv(results, args.output)
# Generate recommendations
print(f"\nπ‘ RECOMMENDATIONS FOR RASPBERRY PI 4B:")
print(f" 1. For fastest inference: Choose model with highest FPS")
print(f" 2. For memory-constrained environments: Choose smallest model")
print(f" 3. For best accuracy/speed tradeoff: Consider YOLOv8s")
print(f" 4. For edge deployment: Convert to TFLite INT8 for ~2-3x speedup")
# Check if we're on Raspberry Pi
if 'arm' in platform.machine().lower() or 'raspberry' in platform.system().lower():
print(f"\nπ― Running on Raspberry Pi - results are accurate for deployment.")
else:
print(f"\nβ οΈ Not running on Raspberry Pi - results are for reference only.")
print(f" Actual Raspberry Pi performance may be 2-5x slower.")
if __name__ == '__main__':
main() |