File size: 11,377 Bytes
3a6b652 |
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 |
"""
학습된 모델 성능 평가 스크립트
- Validation 데이터로 상세 평가
- 클래스별 성능 분석
- 혼동 행렬, PR 곡선 등 시각화
"""
from ultralytics import YOLO
from pathlib import Path
import json
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from sklearn.metrics import classification_report, confusion_matrix
def evaluate_model(model_path, data_yaml='dataset_split/data.yaml', save_dir='evaluation_results'):
"""
모델 상세 평가
Args:
model_path: 학습된 모델 경로 (.pt 파일)
data_yaml: 데이터셋 설정 파일
save_dir: 결과 저장 디렉토리
"""
print("=" * 70)
print("모델 성능 평가")
print("=" * 70)
# 모델 로드
print(f"\n모델 로드: {model_path}")
model = YOLO(model_path)
# 저장 디렉토리 생성
save_path = Path(save_dir)
save_path.mkdir(exist_ok=True)
# ========================================
# 1. 전체 Validation 평가
# ========================================
print("\n" + "=" * 70)
print("1. Validation 데이터셋 평가")
print("=" * 70)
metrics = model.val(
data=data_yaml,
split='val',
save_json=True,
save_hybrid=True,
conf=0.001,
iou=0.6,
max_det=300,
plots=True,
)
# 전체 성능 출력
print("\n📊 전체 성능 지표:")
print(f" mAP50 : {metrics.box.map50:.4f} (50% IoU에서 정확도)")
print(f" mAP50-95 : {metrics.box.map:.4f} (50-95% IoU 평균)")
print(f" Precision : {metrics.box.mp:.4f} (정밀도)")
print(f" Recall : {metrics.box.mr:.4f} (재현율)")
# 클래스별 성능
print("\n📋 클래스별 성능:")
print(f"{'Class':<15} {'mAP50':>8} {'mAP50-95':>10} {'Precision':>10} {'Recall':>8}")
print("-" * 65)
class_names = ['Plastic', 'Vinyl', 'Can', 'Glass', 'Paper']
for i, name in enumerate(class_names):
if i < len(metrics.box.ap50):
map50 = metrics.box.ap50[i]
map50_95 = metrics.box.ap[i]
precision = metrics.box.p[i] if i < len(metrics.box.p) else 0
recall = metrics.box.r[i] if i < len(metrics.box.r) else 0
print(f"{name:<15} {map50:>8.4f} {map50_95:>10.4f} {precision:>10.4f} {recall:>8.4f}")
# ========================================
# 2. 상세 분석 - 예측 결과 수집
# ========================================
print("\n" + "=" * 70)
print("2. 상세 분석 - Validation 이미지 예측")
print("=" * 70)
# Validation 이미지 경로 읽기
val_txt = Path('dataset_split/val.txt')
if val_txt.exists():
with open(val_txt, 'r') as f:
val_images = [line.strip() for line in f.readlines()]
else:
# val.txt가 없으면 직접 찾기
val_images_dir = Path('dataset_split/images/val')
val_images = list(val_images_dir.glob('**/*.[jJ][pP][gG]'))
print(f"Validation 이미지 수: {len(val_images)}")
# 예측 결과 수집
all_true_labels = []
all_pred_labels = []
all_confidences = []
print("예측 진행 중...")
for img_path in val_images[:100]: # 일부만 샘플링 (시간 절약)
# 예측
results = model.predict(img_path, verbose=False, conf=0.25)
# Ground Truth 레이블 읽기
label_path = str(img_path).replace('/images/', '/labels/').replace('\\images\\', '\\labels\\')
label_path = label_path.replace('.jpg', '.txt').replace('.JPG', '.txt')
if Path(label_path).exists():
with open(label_path, 'r') as f:
for line in f:
parts = line.strip().split()
if len(parts) >= 5:
true_class = int(parts[0])
all_true_labels.append(true_class)
# 예측 결과
for r in results:
for box in r.boxes:
pred_class = int(box.cls[0])
conf = float(box.conf[0])
all_pred_labels.append(pred_class)
all_confidences.append(conf)
# ========================================
# 3. 혼동 행렬 (Confusion Matrix)
# ========================================
print("\n" + "=" * 70)
print("3. 혼동 행렬 생성")
print("=" * 70)
if len(all_true_labels) > 0 and len(all_pred_labels) > 0:
# 길이 맞추기
min_len = min(len(all_true_labels), len(all_pred_labels))
all_true_labels = all_true_labels[:min_len]
all_pred_labels = all_pred_labels[:min_len]
# Confusion Matrix
cm = confusion_matrix(all_true_labels, all_pred_labels, labels=range(5))
# 시각화
plt.figure(figsize=(10, 8))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',
xticklabels=class_names,
yticklabels=class_names)
plt.title('Confusion Matrix', fontsize=16, fontweight='bold')
plt.ylabel('True Label', fontsize=12)
plt.xlabel('Predicted Label', fontsize=12)
plt.tight_layout()
cm_path = save_path / 'confusion_matrix_detailed.png'
plt.savefig(cm_path, dpi=300, bbox_inches='tight')
print(f"✅ 혼동 행렬 저장: {cm_path}")
plt.close()
# Classification Report
print("\n📊 Classification Report:")
report = classification_report(
all_true_labels,
all_pred_labels,
target_names=class_names,
digits=4
)
print(report)
# 리포트 저장
with open(save_path / 'classification_report.txt', 'w') as f:
f.write(report)
# ========================================
# 4. 성능 분석 그래프
# ========================================
print("\n" + "=" * 70)
print("4. 성능 분석 그래프 생성")
print("=" * 70)
# 클래스별 mAP 비교
fig, axes = plt.subplots(2, 2, figsize=(14, 10))
# (1) 클래스별 mAP50
ax1 = axes[0, 0]
map50_values = [metrics.box.ap50[i] if i < len(metrics.box.ap50) else 0
for i in range(5)]
colors = ['#FF6B6B', '#4ECDC4', '#45B7D1', '#FFA07A', '#98D8C8']
bars1 = ax1.bar(class_names, map50_values, color=colors, alpha=0.7, edgecolor='black')
ax1.set_ylabel('mAP50', fontsize=12, fontweight='bold')
ax1.set_title('클래스별 mAP50', fontsize=14, fontweight='bold')
ax1.set_ylim(0, 1)
ax1.grid(axis='y', alpha=0.3)
# 값 표시
for bar in bars1:
height = bar.get_height()
ax1.text(bar.get_x() + bar.get_width()/2., height,
f'{height:.3f}', ha='center', va='bottom', fontsize=10)
# (2) 클래스별 mAP50-95
ax2 = axes[0, 1]
map50_95_values = [metrics.box.ap[i] if i < len(metrics.box.ap) else 0
for i in range(5)]
bars2 = ax2.bar(class_names, map50_95_values, color=colors, alpha=0.7, edgecolor='black')
ax2.set_ylabel('mAP50-95', fontsize=12, fontweight='bold')
ax2.set_title('클래스별 mAP50-95', fontsize=14, fontweight='bold')
ax2.set_ylim(0, 1)
ax2.grid(axis='y', alpha=0.3)
for bar in bars2:
height = bar.get_height()
ax2.text(bar.get_x() + bar.get_width()/2., height,
f'{height:.3f}', ha='center', va='bottom', fontsize=10)
# (3) Precision vs Recall
ax3 = axes[1, 0]
precision_values = [metrics.box.p[i] if i < len(metrics.box.p) else 0
for i in range(5)]
recall_values = [metrics.box.r[i] if i < len(metrics.box.r) else 0
for i in range(5)]
x = np.arange(len(class_names))
width = 0.35
ax3.bar(x - width/2, precision_values, width, label='Precision',
color='skyblue', alpha=0.8, edgecolor='black')
ax3.bar(x + width/2, recall_values, width, label='Recall',
color='lightcoral', alpha=0.8, edgecolor='black')
ax3.set_ylabel('Score', fontsize=12, fontweight='bold')
ax3.set_title('Precision vs Recall', fontsize=14, fontweight='bold')
ax3.set_xticks(x)
ax3.set_xticklabels(class_names)
ax3.legend()
ax3.set_ylim(0, 1)
ax3.grid(axis='y', alpha=0.3)
# (4) F1-Score
ax4 = axes[1, 1]
f1_scores = [2 * (p * r) / (p + r) if (p + r) > 0 else 0
for p, r in zip(precision_values, recall_values)]
bars4 = ax4.bar(class_names, f1_scores, color=colors, alpha=0.7, edgecolor='black')
ax4.set_ylabel('F1-Score', fontsize=12, fontweight='bold')
ax4.set_title('클래스별 F1-Score', fontsize=14, fontweight='bold')
ax4.set_ylim(0, 1)
ax4.grid(axis='y', alpha=0.3)
for bar in bars4:
height = bar.get_height()
ax4.text(bar.get_x() + bar.get_width()/2., height,
f'{height:.3f}', ha='center', va='bottom', fontsize=10)
plt.tight_layout()
performance_path = save_path / 'performance_analysis.png'
plt.savefig(performance_path, dpi=300, bbox_inches='tight')
print(f"✅ 성능 분석 그래프 저장: {performance_path}")
plt.close()
# ========================================
# 5. 결과 JSON 저장
# ========================================
results_dict = {
'overall': {
'mAP50': float(metrics.box.map50),
'mAP50_95': float(metrics.box.map),
'precision': float(metrics.box.mp),
'recall': float(metrics.box.mr),
},
'per_class': {}
}
for i, name in enumerate(class_names):
if i < len(metrics.box.ap50):
results_dict['per_class'][name] = {
'mAP50': float(metrics.box.ap50[i]),
'mAP50_95': float(metrics.box.ap[i]),
'precision': float(metrics.box.p[i]) if i < len(metrics.box.p) else 0,
'recall': float(metrics.box.r[i]) if i < len(metrics.box.r) else 0,
'f1_score': f1_scores[i],
}
json_path = save_path / 'evaluation_results.json'
with open(json_path, 'w', encoding='utf-8') as f:
json.dump(results_dict, f, indent=2, ensure_ascii=False)
print(f"✅ 결과 JSON 저장: {json_path}")
# ========================================
# 최종 요약
# ========================================
print("\n" + "=" * 70)
print("평가 완료!")
print("=" * 70)
print(f"\n저장 위치: {save_path.absolute()}")
print("\n생성된 파일:")
print(f" - confusion_matrix_detailed.png")
print(f" - performance_analysis.png")
print(f" - classification_report.txt")
print(f" - evaluation_results.json")
return metrics, results_dict
if __name__ == '__main__':
import sys
# 모델 경로 입력
if len(sys.argv) > 1:
model_path = sys.argv[1]
else:
# 기본값: 최신 학습 모델
default_path = 'waste_classification/yolov8n_5class/weights/best.pt'
print(f"모델 경로를 입력하세요 (엔터: {default_path}):")
user_input = input().strip()
model_path = user_input if user_input else default_path
# 모델 존재 확인
if not Path(model_path).exists():
print(f"❌ 모델 파일을 찾을 수 없습니다: {model_path}")
sys.exit(1)
# 평가 실행
evaluate_model(model_path)
|