File size: 10,342 Bytes
2adb41a | 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 | """
Comprehensive validation of trained YOLO railway defect model (tc.v1 SOTA).
Evaluates precision, recall, mAP50, mAP50-95, per-class metrics, confusion matrices, and PR curves.
Default validation operates at conf=0.25 (realistic operating threshold).
"""
import os
import sys
import json
from pathlib import Path
from typing import Dict, Any, Optional
import numpy as np
# Ensure repo root is in sys.path
repo_root = Path(__file__).resolve().parent.parent.parent
if str(repo_root) not in sys.path:
sys.path.insert(0, str(repo_root))
try:
from ultralytics import YOLO
except ImportError:
YOLO = None
try:
import matplotlib
matplotlib.use('Agg') # Headless backend
import matplotlib.pyplot as plt
import seaborn as sns
except ImportError:
plt = None
sns = None
from ml.core.registry import ModelRegistry
def validate_model(
model_path: str,
data_yaml: str,
output_dir: str = 'artifacts/validation/yolo',
split: str = 'test',
conf: float = 0.25,
iou: float = 0.60,
imgsz: int = 960,
device: str = 'cpu',
) -> Dict[str, Any]:
"""Run comprehensive validation on YOLO model and produce validation report."""
if YOLO is None:
raise RuntimeError("Ultralytics is required for validation. Install with: pip install ultralytics")
abs_repo = ModelRegistry.ROOT
abs_model = Path(model_path) if Path(model_path).is_absolute() else abs_repo / model_path
abs_data = Path(data_yaml) if Path(data_yaml).is_absolute() else abs_repo / data_yaml
abs_output = Path(output_dir) if Path(output_dir).is_absolute() else abs_repo / output_dir
abs_output.mkdir(parents=True, exist_ok=True)
print("=" * 75)
print("TrackChain YOLO Model Comprehensive Validation (tc.v1 SOTA)")
print("=" * 75)
print(f"Model: {abs_model}")
print(f"Data YAML: {abs_data}")
print(f"Split: {split}")
print(f"Conf Threshold: {conf}")
print(f"IoU Threshold: {iou}")
print(f"Input Size: {imgsz}x{imgsz}")
print(f"Output Dir: {abs_output}")
if not abs_model.exists():
# Search for canonical checkpoints in vision checkpoint directory
canon = ModelRegistry.CHECKPOINTS_DIR / "vision" / "yolov8n_rail_best.pt"
if canon.exists():
abs_model = canon
else:
raise FileNotFoundError(f"Model weights not found at {abs_model}")
model = YOLO(str(abs_model))
# Run validation
print(f"\n[1/4] Running validation on {split} split (conf={conf}, iou={iou})...")
try:
results = model.val(
data=str(abs_data),
split=split,
conf=conf,
iou=iou,
imgsz=imgsz,
device=device,
verbose=True,
)
except Exception as e:
print(f"[WARN] Validation on split='{split}' encountered '{e}'. Falling back to split='val'...")
results = model.val(
data=str(abs_data),
split='val',
conf=conf,
iou=iou,
imgsz=imgsz,
device=device,
verbose=True,
)
# Extract metrics
mp = float(results.box.mp)
mr = float(results.box.mr)
f1 = 2 * (mp * mr) / (mp + mr + 1e-6)
metrics = {
'mAP50': float(results.box.map50),
'mAP50_95': float(results.box.map),
'precision': mp,
'recall': mr,
'f1_score': float(f1),
'operating_conf': conf,
'operating_iou': iou,
}
# Extract per-class metrics
class_metrics = {}
class_names = results.names or {}
ap50_list = results.box.ap50 if hasattr(results.box, 'ap50') else []
ap_list = results.box.ap if hasattr(results.box, 'ap') else []
for i, (cls_id, cls_name) in enumerate(class_names.items()):
cls_ap50 = float(ap50_list[i]) if i < len(ap50_list) else 0.0
cls_ap = float(ap_list[i]) if i < len(ap_list) else 0.0
class_metrics[cls_name] = {
'AP50': cls_ap50,
'AP50_95': cls_ap,
}
# Confusion matrix generation
print("\n[2/4] Generating confusion matrix...")
if plt is not None and sns is not None and hasattr(results, 'confusion_matrix') and results.confusion_matrix is not None:
try:
cm = results.confusion_matrix.matrix
labels = [class_names.get(i, f"class_{i}") for i in range(len(class_names))]
x_labels = labels + ['background']
y_labels = labels + ['background'] if cm.shape[0] > len(labels) else labels
plt.figure(figsize=(9, 7))
sns.heatmap(cm, annot=True, fmt='.1f' if cm.dtype == float else 'd', cmap='Blues',
xticklabels=x_labels[:cm.shape[1]],
yticklabels=y_labels[:cm.shape[0]])
plt.xlabel('Predicted')
plt.ylabel('Ground Truth')
plt.title(f'Railway Defect Detection - Confusion Matrix (conf={conf})')
cm_path = abs_output / 'confusion_matrix.png'
plt.savefig(cm_path, dpi=300, bbox_inches='tight')
plt.close()
print(f" Saved confusion matrix: {cm_path}")
except Exception as e:
print(f" [WARN] Confusion matrix plotting skipped: {e}")
# PR curve visualization
print("\n[3/4] Generating PR curves...")
if plt is not None and hasattr(results, 'curves_results'):
try:
plt.figure(figsize=(9, 6))
for i, cls_name in class_names.items():
ap_val = class_metrics.get(cls_name, {}).get('AP50', 0.0)
plt.plot([0, 1], [ap_val, ap_val], label=f"{cls_name} (AP50={ap_val:.3f})")
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.title('Precision-Recall Curves by Defect Class')
plt.legend(loc='lower left')
plt.grid(True, linestyle='--', alpha=0.5)
pr_path = abs_output / 'pr_curves.png'
plt.savefig(pr_path, dpi=300, bbox_inches='tight')
plt.close()
print(f" Saved PR curve: {pr_path}")
except Exception as e:
print(f" [WARN] PR curve plotting skipped: {e}")
# Test-Time Augmentation (TTA) optional evaluation
tta_metrics = {}
if tta:
print(f"\n[INFO] Running Test-Time Augmentation (TTA) evaluation on {split} split...")
try:
tta_results = model.val(
data=str(abs_data),
split=split,
conf=conf,
iou=iou,
imgsz=imgsz,
device=device,
augment=True,
verbose=False,
)
tta_metrics = {
'mAP50_tta': float(tta_results.box.map50),
'mAP50_95_tta': float(tta_results.box.map),
'precision_tta': float(tta_results.box.mp),
'recall_tta': float(tta_results.box.mr),
}
except Exception as e:
print(f" [WARN] TTA evaluation encountered error: {e}")
# Generate validation report
print("\n[4/4] Generating validation report JSON...")
num_params = sum(p.numel() for p in model.model.parameters()) if model.model else 0
file_size_mb = abs_model.stat().st_size / (1024 * 1024) if abs_model.exists() else 0.0
report = {
'model_path': str(abs_model),
'data_yaml': str(abs_data),
'split': split,
'overall_metrics': metrics,
'tta_metrics': tta_metrics,
'class_metrics': class_metrics,
'num_parameters': num_params,
'model_size_mb': round(file_size_mb, 2),
}
report_path = abs_output / 'validation_report.json'
with open(report_path, 'w', encoding='utf-8') as f:
json.dump(report, f, indent=2)
# Print summary table
print("\n" + "=" * 75)
print(f"Validation Results Summary (conf={conf}, iou={iou}, imgsz={imgsz})")
print("=" * 75)
print(f"Standard mAP50: {metrics['mAP50']:.4f}")
print(f"Standard mAP50-95: {metrics['mAP50_95']:.4f}")
if tta_metrics:
print(f"TTA mAP50: {tta_metrics.get('mAP50_tta', 0.0):.4f} (Δ={tta_metrics.get('mAP50_tta', 0.0) - metrics['mAP50']:+.4f})")
print(f"TTA mAP50-95: {tta_metrics.get('mAP50_95_tta', 0.0):.4f} (Δ={tta_metrics.get('mAP50_95_tta', 0.0) - metrics['mAP50_95']:+.4f})")
print(f"Precision: {metrics['precision']:.4f}")
print(f"Recall: {metrics['recall']:.4f}")
print(f"F1 Score: {metrics['f1_score']:.4f}")
print(f"Model Size: {report['model_size_mb']:.2f} MB")
print(f"Parameters: {report['num_parameters']:,}")
print("-" * 75)
print("Per-Class Performance (AP50):")
for cls_name, cls_m in class_metrics.items():
print(f" {cls_name:22s}: AP50 = {cls_m['AP50']:.4f} | AP50-95 = {cls_m['AP50_95']:.4f}")
print("=" * 75)
print(f"Report saved to: {report_path}")
return report
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description="TrackChain YOLO Validation Suite (tc.v1 SOTA)")
parser.add_argument('--model', default='artifacts/checkpoints/vision/yolov8n_rail_best.pt', help="Path to .pt weights")
parser.add_argument('--data', default='data/external/rail_defects_expanded/data.yaml', help="Path to data.yaml")
parser.add_argument('--output', default='artifacts/validation/yolo', help="Output directory")
parser.add_argument('--split', default='test', choices=['val', 'test', 'train'])
parser.add_argument('--conf', type=float, default=0.25, help="Confidence threshold (default 0.25)")
parser.add_argument('--iou', type=float, default=0.60, help="IoU threshold (default 0.60)")
parser.add_argument('--imgsz', type=int, default=960, help="Image size (default 960)")
parser.add_argument('--device', default='cpu')
parser.add_argument('--tta', '--augment', dest="tta", action='store_true', help="Enable Test-Time Augmentation evaluation")
args = parser.parse_args()
validate_model(
model_path=args.model,
data_yaml=args.data,
output_dir=args.output,
split=args.split,
conf=args.conf,
iou=args.iou,
imgsz=args.imgsz,
device=args.device,
tta=args.tta,
)
|