yolo8 모델 성능 개선¶
In [ ]:
!pip install ultralytics
Collecting ultralytics Downloading ultralytics-8.3.17-py3-none-any.whl.metadata (34 kB) Requirement already satisfied: numpy>=1.23.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (1.26.4) Requirement already satisfied: matplotlib>=3.3.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (3.7.1) Requirement already satisfied: opencv-python>=4.6.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (4.10.0.84) Requirement already satisfied: pillow>=7.1.2 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (10.4.0) Requirement already satisfied: pyyaml>=5.3.1 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (6.0.2) Requirement already satisfied: requests>=2.23.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (2.32.3) Requirement already satisfied: scipy>=1.4.1 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (1.13.1) Requirement already satisfied: torch>=1.8.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (2.4.1+cu121) Requirement already satisfied: torchvision>=0.9.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (0.19.1+cu121) Requirement already satisfied: tqdm>=4.64.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (4.66.5) Requirement already satisfied: psutil in /usr/local/lib/python3.10/dist-packages (from ultralytics) (5.9.5) Requirement already satisfied: py-cpuinfo in /usr/local/lib/python3.10/dist-packages (from ultralytics) (9.0.0) Requirement already satisfied: pandas>=1.1.4 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (2.2.2) Requirement already satisfied: seaborn>=0.11.0 in /usr/local/lib/python3.10/dist-packages (from ultralytics) (0.13.2) Collecting ultralytics-thop>=2.0.0 (from ultralytics) Downloading ultralytics_thop-2.0.9-py3-none-any.whl.metadata (9.3 kB) Requirement already satisfied: contourpy>=1.0.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (1.3.0) Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (0.12.1) Requirement already satisfied: fonttools>=4.22.0 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (4.54.1) Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (1.4.7) Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (24.1) Requirement already satisfied: pyparsing>=2.3.1 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (3.2.0) Requirement already satisfied: python-dateutil>=2.7 in /usr/local/lib/python3.10/dist-packages (from matplotlib>=3.3.0->ultralytics) (2.8.2) Requirement already satisfied: pytz>=2020.1 in /usr/local/lib/python3.10/dist-packages (from pandas>=1.1.4->ultralytics) (2024.2) Requirement already satisfied: tzdata>=2022.7 in /usr/local/lib/python3.10/dist-packages (from pandas>=1.1.4->ultralytics) (2024.2) Requirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python3.10/dist-packages (from requests>=2.23.0->ultralytics) (3.4.0) Requirement already satisfied: idna<4,>=2.5 in /usr/local/lib/python3.10/dist-packages (from requests>=2.23.0->ultralytics) (3.10) Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.10/dist-packages (from requests>=2.23.0->ultralytics) (2.2.3) Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.10/dist-packages (from requests>=2.23.0->ultralytics) (2024.8.30) Requirement already satisfied: filelock in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (3.16.1) Requirement already satisfied: typing-extensions>=4.8.0 in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (4.12.2) Requirement already satisfied: sympy in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (1.13.3) Requirement already satisfied: networkx in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (3.4.1) Requirement already satisfied: jinja2 in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (3.1.4) Requirement already satisfied: fsspec in /usr/local/lib/python3.10/dist-packages (from torch>=1.8.0->ultralytics) (2024.6.1) Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.10/dist-packages (from python-dateutil>=2.7->matplotlib>=3.3.0->ultralytics) (1.16.0) Requirement already satisfied: MarkupSafe>=2.0 in /usr/local/lib/python3.10/dist-packages (from jinja2->torch>=1.8.0->ultralytics) (3.0.1) Requirement already satisfied: mpmath<1.4,>=1.1.0 in /usr/local/lib/python3.10/dist-packages (from sympy->torch>=1.8.0->ultralytics) (1.3.0) Downloading ultralytics-8.3.17-py3-none-any.whl (876 kB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 876.6/876.6 kB 40.7 MB/s eta 0:00:00 Downloading ultralytics_thop-2.0.9-py3-none-any.whl (26 kB) Installing collected packages: ultralytics-thop, ultralytics Successfully installed ultralytics-8.3.17 ultralytics-thop-2.0.9
In [ ]:
from ultralytics import YOLO
import cv2 # OpenCV 라이브러리를 불러오기
from google.colab.patches import cv2_imshow # from google.colab.patches import cv2_imshow: Google Colab 환경에서 OpenCV 이미지를 표시하기 위한 함수
# YOLOv8 모델 로드
model = YOLO('yolov8n.pt') # YOLOv8의 경량 모델
# 이미지 로드
img = cv2.imread('Dog_rawPixel01.jpg')
# 객체 탐지
# 로드된 YOLOv8 모델을 사용하여 이미지에서 객체를 탐지합니다. results 변수에는 탐지 결과가 저장
results = model(img)
# 결과 시각화
# plot() 함수는 탐지된 객체 주변에 경계 상자를 그려 이미지를 반환
img_with_detections = results[0].plot() # 첫 번째 결과를 시각화
# 결과 이미지 표시
cv2_imshow(img_with_detections)
Output hidden; open in https://colab.research.google.com to view.
다른 이미지 확인¶
In [ ]:
# 이미지 로드
img = cv2.imread('life_unsplash.jpeg')
# 객체 탐지
# 로드된 YOLOv8 모델을 사용하여 이미지에서 객체를 탐지합니다. results 변수에는 탐지 결과가 저장
results = model(img)
# 결과 시각화
# plot() 함수는 탐지된 객체 주변에 경계 상자를 그려 이미지를 반환
img_with_detections = results[0].plot() # 첫 번째 결과를 시각화
# 결과 이미지 표시
cv2_imshow(img_with_detections)
0: 640x448 1 umbrella, 2 chairs, 1 couch, 1 bed, 1 dining table, 1 book, 39.6ms Speed: 2.9ms preprocess, 39.6ms inference, 1.4ms postprocess per image at shape (1, 3, 640, 448)
성능 개선 - 더 좋은 모델 사용¶
In [ ]:
# YOLO8의 주요 모델 크기는 다음과 같습니다 (작은 것부터 큰 순서로):
# YOLOv8n (nano)
# YOLOv8s (small)
# YOLOv8m (medium)
# YOLOv8l (large)
# YOLOv8x (extra large)
In [ ]:
from ultralytics import YOLO
import cv2 # OpenCV 라이브러리를 불러오기
from google.colab.patches import cv2_imshow # from google.colab.patches import cv2_imshow: Google Colab 환경에서 OpenCV 이미지를 표시하기 위한 함수
# YOLOv8 모델 로드
model = YOLO('yolov8l.pt') # YOLOv8의 경량 모델
# 이미지 로드
img = cv2.imread('life_unsplash.jpeg')
# 객체 탐지
# 로드된 YOLOv8 모델을 사용하여 이미지에서 객체를 탐지합니다. results 변수에는 탐지 결과가 저장
results = model(img)
# 결과 시각화
# plot() 함수는 탐지된 객체 주변에 경계 상자를 그려 이미지를 반환
img_with_detections = results[0].plot() # 첫 번째 결과를 시각화
# 결과 이미지 표시
cv2_imshow(img_with_detections)
0: 640x448 1 chair, 1 bed, 1 dining table, 1 book, 1 scissors, 48.9ms Speed: 2.2ms preprocess, 48.9ms inference, 1.9ms postprocess per image at shape (1, 3, 640, 448)
COCO128 데이터 셋을 이용한 모델 성능 평가¶
데이터 준비¶
In [ ]:
!rm -rf /content/datasets
!rm -rf /content/coco128
!rm -rf /content/runs
데이터 다운로드 및 압축 풀기¶
In [ ]:
# 필요한 모듈 임포트
from ultralytics import YOLO
from ultralytics.utils.downloads import download
import os
import zipfile
# COCO128 데이터셋 다운로드
download(url='https://ultralytics.com/assets/coco128.zip')
# 다운로드된 파일 확인
print("Downloaded files:", os.listdir())
# ZIP 파일 압축 해제
with zipfile.ZipFile('coco128.zip', 'r') as zip_ref:
zip_ref.extractall('.')
# 압축 해제 후 파일 확인
print("Files after extraction:", os.listdir())
print("Contents of coco128 folder:", os.listdir('coco128'))
Unzipping /content/coco128.zip to /content/coco128...: 100%|██████████| 263/263 [00:00<00:00, 3458.06file/s]
Downloaded files: ['.config', 'coco128.zip', 'yolov8_coco128_trained_custom_improved.pt', 'yolov8s.pt', 'coco128_custom.yaml', 'yolov8_transfer_learning_50epoch.pt', 'Dog_rawPixel01.jpg', 'yolov8x.pt', 'life_unsplash.jpeg', 'yolov8l.pt', 'coco128', 'yolov8_coco128_trained_custom.pt', 'yolo11n.pt', 'yolov8n.pt', 'ferrari.jpg', 'sample_data'] Files after extraction: ['.config', 'coco128.zip', 'yolov8_coco128_trained_custom_improved.pt', 'yolov8s.pt', 'coco128_custom.yaml', 'yolov8_transfer_learning_50epoch.pt', 'Dog_rawPixel01.jpg', 'yolov8x.pt', 'life_unsplash.jpeg', 'yolov8l.pt', 'coco128', 'yolov8_coco128_trained_custom.pt', 'yolo11n.pt', 'yolov8n.pt', 'ferrari.jpg', 'sample_data'] Contents of coco128 folder: ['LICENSE', 'labels', 'README.txt', 'images']
In [ ]:
import cv2 # OpenCV 라이브러리를 불러오기
from google.colab.patches import cv2_imshow # from google.colab.patches import cv2_imshow: Google Colab 환경에서 OpenCV 이미지를 표시하기 위한 함수
# YOLOv8 모델 로드
model = YOLO('yolov8l.pt') # YOLOv8의 경량 모델
# 이미지 로드
img = cv2.imread('coco128/images/train2017/000000000081.jpg')
# 객체 탐지
# 로드된 YOLOv8 모델을 사용하여 이미지에서 객체를 탐지합니다. results 변수에는 탐지 결과가 저장
results = model(img)
# 결과 시각화
# plot() 함수는 탐지된 객체 주변에 경계 상자를 그려 이미지를 반환
img_with_detections = results[0].plot() # 첫 번째 결과를 시각화
# 결과 이미지 표시
cv2_imshow(img_with_detections)
0: 448x640 1 airplane, 65.1ms Speed: 1.7ms preprocess, 65.1ms inference, 2.7ms postprocess per image at shape (1, 3, 448, 640)
yaml 파일 생성¶
- YAML 파일은 YAML(YAML Ain't Markup Language)의 줄임말로, 데이터를 구조화해서 표현할 때 사용하는 사람이 읽기 쉬운 데이터 형식입니다. 주로 설정 파일로 많이 사용됩니다.
- YOLOv8에서 YAML 파일은 주로 모델 설정, 데이터셋 정보, 학습 파라미터 등을 정의하는 데 사용됩니다. 예를 들어, 데이터셋 경로나 클래스 정보, 학습 시 사용될 매개변수 등을 YAML 파일에서 설정합니다.
# 데이터셋의 루트 디렉토리 경로
path: /path/to/dataset
# 훈련 이미지의 상대 경로
train: images/train
# 검증 이미지의 상대 경로
val: images/val
# 테스트 이미지의 상대 경로 (선택사항)
test: images/test
# 클래스 이름
names:
0: class1
1: class2
2: class3
# ... 추가 클래스들
# 또는 클래스 이름을 리스트로 표현할 수 있습니다
# names: ['class1', 'class2', 'class3', ...]
# 추가 설정 (선택사항)
nc: 3 # 클래스의 수
path: ./coco128 # 데이터셋의 경로
train: images/train # 학습 이미지 경로
val: images/val # 검증 이미지 경로
test: images/test # 테스트 이미지 경로
nc: 80 # 클래스 개수
names: ['person', 'bicycle', 'car', ...] # 클래스 이름 리스트
yaml 파일 작성해 보기¶
In [ ]:
import yaml
from ultralytics import YOLO
import os
# COCO128 데이터셋 경로 확인
dataset_path = 'coco128'
if not os.path.exists(dataset_path):
raise FileNotFoundError(f"Dataset directory '{dataset_path}' not found.")
# YAML 파일 생성
yaml_content = {
'path': os.path.abspath(dataset_path), # 절대 경로 사용
'train': 'images/train2017',
'val': 'images/train2017', # COCO128은 검증 세트가 없으므로 훈련 세트를 사용
'names': {0: 'person', 1: 'bicycle', 2: 'car', 3: 'motorcycle', 4: 'airplane', 5: 'bus', 6: 'train', 7: 'truck', 8: 'boat', 9: 'traffic light', 10: 'fire hydrant',
11: 'stop sign', 12: 'parking meter', 13: 'bench', 14: 'bird', 15: 'cat', 16: 'dog', 17: 'horse', 18: 'sheep', 19: 'cow', 20: 'elephant',
21: 'bear', 22: 'zebra', 23: 'giraffe', 24: 'backpack', 25: 'umbrella', 26: 'handbag', 27: 'tie', 28: 'suitcase', 29: 'frisbee', 30: 'skis',
31: 'snowboard', 32: 'sports ball', 33: 'kite', 34: 'baseball bat', 35: 'baseball glove', 36: 'skateboard', 37: 'surfboard', 38: 'tennis racket', 39: 'bottle', 40: 'wine glass',
41: 'cup', 42: 'fork', 43: 'knife', 44: 'spoon', 45: 'bowl', 46: 'banana', 47: 'apple', 48: 'sandwich', 49: 'orange', 50: 'broccoli',
51: 'carrot', 52: 'hot dog', 53: 'pizza', 54: 'donut', 55: 'cake', 56: 'chair', 57: 'couch', 58: 'potted plant', 59: 'bed', 60: 'dining table',
61: 'toilet', 62: 'tv', 63: 'laptop', 64: 'mouse', 65: 'remote', 66: 'keyboard', 67: 'cell phone', 68: 'microwave', 69: 'oven', 70: 'toaster',
71: 'sink', 72: 'refrigerator', 73: 'book', 74: 'clock', 75: 'vase', 76: 'scissors', 77: 'teddy bear', 78: 'hair drier', 79: 'toothbrush'}
}
yaml_path = 'coco128_custom.yaml'
with open(yaml_path, 'w') as file:
yaml.dump(yaml_content, file, default_flow_style=False)
print(f"Created custom YAML file: {yaml_path}")
Created custom YAML file: coco128_custom.yaml
모델 원본 성능¶
In [ ]:
from ultralytics import YOLO
import cv2
import matplotlib.pyplot as plt
# 원본 YOLOv8 모델 로드
model = YOLO('yolov8n.pt')
# 테스트 데이터셋 경로 (COCO 형식의 데이터셋 사용)
# test_data = 'coco128_custom.yaml'
# 모델 평가
results = model.val(data=test_data, conf=0.25, iou=0.5)
# 결과 출력
print("원본 YOLOv8 모델 성능:")
print(f"mAP50: {results.box.map50:.4f}")
print(f"mAP50-95: {results.box.map:.4f}")
# 샘플 이미지에 대한 예측
sample_image = cv2.imread('Dog_rawPixel01.jpg')
sample_results = model(sample_image)
# 결과 시각화
img_with_boxes = sample_results[0].plot()
plt.figure(figsize=(12, 8))
plt.imshow(cv2.cvtColor(img_with_boxes, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.title("원본 YOLOv8 모델 예측 결과")
plt.show()
# 예측 결과 분석
for r in sample_results:
for box in r.boxes:
class_id = int(box.cls[0])
confidence = float(box.conf[0])
x1, y1, x2, y2 = box.xyxy[0].tolist()
print(f"클래스: {model.names[class_id]}, 신뢰도: {confidence:.2f}, 바운딩 박스: [{x1:.2f}, {y1:.2f}, {x2:.2f}, {y2:.2f}]")
# 클래스 수 출력
print(f"클래스 수: {len(model.names)}")
print("클래스 목록:")
for i, name in model.names.items():
print(f"{i}: {name}")
Ultralytics 8.3.16 🚀 Python-3.10.12 torch-2.4.1+cu121 CUDA:0 (Tesla T4, 15102MiB) YOLOv8n summary (fused): 168 layers, 3,151,904 parameters, 0 gradients, 8.7 GFLOPs
val: Scanning /content/coco128/labels/train2017.cache... 126 images, 2 backgrounds, 0 corrupt: 100%|██████████| 128/128 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 8/8 [00:03<00:00, 2.61it/s]
all 128 929 0.699 0.493 0.626 0.494
person 61 254 0.84 0.661 0.78 0.609
bicycle 3 6 0.667 0.333 0.499 0.43
car 12 46 0.909 0.217 0.571 0.406
motorcycle 4 5 0.667 0.8 0.825 0.671
airplane 5 6 0.8 0.667 0.8 0.612
bus 5 7 0.556 0.714 0.794 0.742
train 3 3 1 0.667 0.833 0.733
truck 5 12 1 0.25 0.625 0.444
boat 2 6 0.5 0.167 0.374 0.0748
traffic light 4 14 0.667 0.143 0.429 0.386
stop sign 2 2 1 0.5 0.75 0.6
bench 5 9 1 0.222 0.611 0.412
bird 2 16 0.917 0.688 0.825 0.493
cat 4 4 1 1 0.995 0.821
dog 9 9 0.615 0.889 0.766 0.627
horse 1 2 0.667 1 0.995 0.514
elephant 4 17 0.929 0.765 0.874 0.706
bear 1 1 1 1 0.995 0.995
zebra 2 4 1 1 0.995 0.973
giraffe 4 9 0.889 0.889 0.926 0.712
backpack 4 6 0.667 0.333 0.556 0.364
umbrella 4 18 1 0.444 0.722 0.527
handbag 9 19 0 0 0 0
tie 6 7 0.833 0.714 0.785 0.583
suitcase 2 4 0.5 0.5 0.539 0.479
frisbee 5 5 0.8 0.8 0.839 0.775
skis 1 1 1 1 0.995 0.497
snowboard 2 7 0.8 0.571 0.744 0.555
sports ball 6 6 0.667 0.333 0.556 0.369
kite 2 10 0.8 0.4 0.619 0.271
baseball bat 4 4 1 0.25 0.625 0.375
baseball glove 4 7 0.75 0.429 0.642 0.465
skateboard 3 5 1 0.6 0.8 0.563
tennis racket 5 7 0.667 0.286 0.524 0.424
bottle 6 18 0.545 0.333 0.382 0.27
wine glass 5 16 0.714 0.312 0.559 0.45
cup 10 36 0.667 0.278 0.503 0.419
fork 6 6 1 0.167 0.583 0.525
knife 7 16 0.778 0.438 0.656 0.481
spoon 5 22 0.667 0.182 0.438 0.275
bowl 9 28 0.692 0.643 0.689 0.55
banana 1 1 0 0 0 0
sandwich 2 2 0.333 0.5 0.25 0.25
orange 1 4 1 0.25 0.625 0.312
broccoli 4 11 0.5 0.182 0.388 0.365
carrot 3 24 0.786 0.458 0.63 0.434
hot dog 1 2 0.5 0.5 0.622 0.622
pizza 5 5 0.833 1 0.995 0.865
donut 2 14 0.667 1 0.94 0.873
cake 4 4 0.8 1 0.995 0.904
chair 9 35 0.6 0.514 0.503 0.291
couch 5 6 0.6 0.5 0.648 0.561
potted plant 9 14 0.692 0.643 0.746 0.541
bed 3 3 1 0.667 0.833 0.733
dining table 10 13 0.4 0.462 0.514 0.422
toilet 2 2 1 0.5 0.75 0.75
tv 2 2 0.5 0.5 0.622 0.622
laptop 2 3 0 0 0 0
mouse 2 2 0 0 0 0
remote 5 8 1 0.5 0.75 0.692
cell phone 5 8 0 0 0 0
microwave 3 3 0.667 0.667 0.777 0.744
oven 5 5 0.5 0.4 0.481 0.389
sink 4 6 0.5 0.167 0.292 0.233
refrigerator 5 5 0.667 0.4 0.598 0.489
book 6 29 0.75 0.103 0.423 0.311
clock 8 9 0.778 0.778 0.85 0.743
vase 2 2 0.4 1 0.828 0.8
scissors 1 1 0 0 0 0
teddy bear 6 21 1 0.333 0.667 0.487
toothbrush 2 5 1 0.4 0.7 0.469
Speed: 0.3ms preprocess, 5.2ms inference, 0.0ms loss, 4.8ms postprocess per image
Results saved to runs/detect/val2
원본 YOLOv8 모델 성능:
mAP50: 0.6260
mAP50-95: 0.4942
0: 480x640 2 dogs, 67.3ms
Speed: 4.3ms preprocess, 67.3ms inference, 2.1ms postprocess per image at shape (1, 3, 480, 640)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50896 (\N{HANGUL SYLLABLE WEON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 48376 (\N{HANGUL SYLLABLE BON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 47784 (\N{HANGUL SYLLABLE MO}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 45944 (\N{HANGUL SYLLABLE DEL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50696 (\N{HANGUL SYLLABLE YE}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 52769 (\N{HANGUL SYLLABLE CEUG}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44208 (\N{HANGUL SYLLABLE GYEOL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44284 (\N{HANGUL SYLLABLE GWA}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
클래스: dog, 신뢰도: 0.87, 바운딩 박스: [602.20, 193.65, 908.10, 742.98] 클래스: dog, 신뢰도: 0.75, 바운딩 박스: [309.01, 220.69, 621.53, 732.43] 클래스 수: 80 클래스 목록: 0: person 1: bicycle 2: car 3: motorcycle 4: airplane 5: bus 6: train 7: truck 8: boat 9: traffic light 10: fire hydrant 11: stop sign 12: parking meter 13: bench 14: bird 15: cat 16: dog 17: horse 18: sheep 19: cow 20: elephant 21: bear 22: zebra 23: giraffe 24: backpack 25: umbrella 26: handbag 27: tie 28: suitcase 29: frisbee 30: skis 31: snowboard 32: sports ball 33: kite 34: baseball bat 35: baseball glove 36: skateboard 37: surfboard 38: tennis racket 39: bottle 40: wine glass 41: cup 42: fork 43: knife 44: spoon 45: bowl 46: banana 47: apple 48: sandwich 49: orange 50: broccoli 51: carrot 52: hot dog 53: pizza 54: donut 55: cake 56: chair 57: couch 58: potted plant 59: bed 60: dining table 61: toilet 62: tv 63: laptop 64: mouse 65: remote 66: keyboard 67: cell phone 68: microwave 69: oven 70: toaster 71: sink 72: refrigerator 73: book 74: clock 75: vase 76: scissors 77: teddy bear 78: hair drier 79: toothbrush
YOLOv8 모델 정보¶
- 버전: Ultralytics 8.3.16
- Python 버전: Python-3.10.12
- PyTorch 버전: torch-2.4.1+cu121
- CUDA: 사용되는 GPU (Tesla T4, 15102MiB)
- 모델 요약:
- 총 168 layers
- 3,151,904 개의 파라미터
- 0 gradients
- 8.7 GFLOPs (FLOPs는 부동소수점 연산 횟수를 나타냄)
val: Scanning /content/coco128/labels/train2017.cache... 126 images, 2 backgrounds, 0 corrupt: 100%|██████████| 128/128 [00:00<?, ?it/s]
검증 단계¶
- 검증 데이터셋: 총 126장의 이미지(배경 포함)에서 데이터를 스캔했습니다.
- 검증 결과 요약:
- 각 클래스에 대한 이미지 수, 인스턴스 수, 박스 정확도(P), 재현율(R), 평균 정밀도(mAP50), mAP50-95의 값을 요약합니다.
Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 8/8 [00:03<00:00, 2.61it/s]
all 128 929 0.699 0.493 0.626 0.494
person 61 254 0.84 0.661 0.78 0.609
bicycle 3 6 0.667 0.333 0.499 0.43
car 12 46 0.909 0.217 0.571 0.406
motorcycle 4 5 0.667 0.8 0.825 0.671
airplane 5 6 0.8 0.667 0.8 0.612
bus 5 7 0.556 0.714 0.794 0.742
train 3 3 1 0.667 0.833 0.733
truck 5 12 1 0.25 0.625 0.444
...
클래스별 성능¶
"all" 클래스(모든 클래스 통합):
- 이미지 수: 128
- 인스턴스 수: 929
- Box P: 0.699 (정확도)
- R: 0.493 (재현율)
- mAP50: 0.626 (평균 정밀도에 대한 값)
- mAP50-95: 0.494
특정 클래스 예시:
- 사람(person): P: 0.84, R: 0.661, mAP50: 0.78
- 자동차(car): P: 0.909, R: 0.217, mAP50: 0.571
- 개(dog): P: 0.615, R: 0.889, mAP50: 0.766
- 과일(banana): P: 0.0, R: 0.0, mAP50: 0.0 (모델이 과일 인식에 실패)
Speed: 0.3ms preprocess, 5.2ms inference, 0.0ms loss, 4.8ms postprocess per image
Results saved to runs/detect/val2
원본 YOLOv8 모델 성능:
mAP50: 0.6260
mAP50-95: 0.4942
성능 데이터¶
속도:
- 이미지 전처리: 0.3ms
- 추론(inference): 5.2ms
- 손실(loss): 0.0ms
- 후처리(postprocess): 4.8ms
최종 성능 결과
- YOLOv8 모델의 성능 지표:
- mAP50: 0.6260 (0.626)
- mAP50-95: 0.4942 (0.4942)
주요 지표인 mAP50과 mAP50-95를 분석
mAP50 (Mean Average Precision at IoU=0.50): 0.6260
이 값은 IoU(Intersection over Union) 임계값이 0.5일 때의 평균 정밀도를 나타냅니다.
0.6260은 62.60%의 정확도를 의미합니다.
이는 모델이 객체의 위치를 대략적으로 잘 찾아내고 있음을 나타냅니다.
mAP50-95 (Mean Average Precision at IoU=0.50:0.95): 0.4942
이 값은 IoU 임계값을 0.5에서 0.95까지 변화시키면서 계산한 평균 정밀도의 평균입니다.
0.4942는 49.42%의 정확도를 의미합니다.
이 값이 mAP50보다 낮은 것은 정상적이며, 더 엄격한 기준으로 평가했기 때문입니다.
성능 평가:
일반적인 기준:
mAP50이 0.6 이상이면 괜찮은 성능으로 간주됩니다.
mAP50-95가 0.4 이상이면 준수한 성능으로 볼 수 있습니다.
이 모델의 경우:
mAP50이 0.6260으로, 기본적인 객체 탐지 능력이 양호합니다.
mAP50-95가 0.4942로, 더 엄격한 기준에서도 준수한 성능을 보여줍니다.
해석:
이 모델은 객체의 대략적인 위치를 잘 찾아내며, 정확한 경계 상자 예측에서도 준수한 성능을 보입니다.
일반적인 객체 탐지 작업에 적합한 수준의 성능을 가지고 있습니다.
개선 가능성:
더 높은 성능이 필요하다면, 모델 아키텍처 개선, 데이터 증강, 더 긴 훈련 시간 등을 통해 성능을 향상시킬 수 있습니다.
용도에 따른 평가:
일반적인 객체 탐지 작업에는 충분한 성능입니다.
매우 정밀한 경계 상자가 필요한 특수한 응용 분야에서는 추가적인 개선이 필요할 수 있습니다.
결론적으로, 이 YOLOv8 모델은 전반적으로 양호한 성능을 보여주고 있으며, 대부분의 일반적인 객체 탐지 작업에 적합할 것으로 판단됩니다. 특정 응용 분야나 더 높은 정확도가 요구되는 경우에는 추가적인 최적화나 훈련이 필요할 수 있습니다.
In [ ]:
from ultralytics import YOLO
import cv2
import matplotlib.pyplot as plt
# 원본 YOLOv8 모델 로드
model = YOLO('yolov8n.pt')
# 테스트 데이터셋 경로 (COCO 형식의 데이터셋 사용)
test_data = 'coco128_custom.yaml'
# 모델 평가
results = model.val(data=test_data, conf=0.25, iou=0.5)
# 결과 출력
print("원본 YOLOv8 모델 성능:")
print(f"mAP50: {results.box.map50:.4f}")
print(f"mAP50-95: {results.box.map:.4f}")
# 샘플 이미지에 대한 예측
sample_image = cv2.imread('life_unsplash.jpeg')
sample_results = model(sample_image)
# 결과 시각화
img_with_boxes = sample_results[0].plot()
plt.figure(figsize=(12, 8))
plt.imshow(cv2.cvtColor(img_with_boxes, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.title("원본 YOLOv8 모델 예측 결과")
plt.show()
# 예측 결과 분석
for r in sample_results:
for box in r.boxes:
class_id = int(box.cls[0])
confidence = float(box.conf[0])
x1, y1, x2, y2 = box.xyxy[0].tolist()
print(f"클래스: {model.names[class_id]}, 신뢰도: {confidence:.2f}, 바운딩 박스: [{x1:.2f}, {y1:.2f}, {x2:.2f}, {y2:.2f}]")
# 클래스 수 출력
print(f"클래스 수: {len(model.names)}")
print("클래스 목록:")
for i, name in model.names.items():
print(f"{i}: {name}")
Ultralytics 8.3.16 🚀 Python-3.10.12 torch-2.4.1+cu121 CUDA:0 (Tesla T4, 15102MiB) YOLOv8n summary (fused): 168 layers, 3,151,904 parameters, 0 gradients, 8.7 GFLOPs
val: Scanning /content/coco128/labels/train2017.cache... 126 images, 2 backgrounds, 0 corrupt: 100%|██████████| 128/128 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 8/8 [00:04<00:00, 1.90it/s]
all 128 929 0.699 0.493 0.626 0.494
person 61 254 0.84 0.661 0.78 0.609
bicycle 3 6 0.667 0.333 0.499 0.43
car 12 46 0.909 0.217 0.571 0.406
motorcycle 4 5 0.667 0.8 0.825 0.671
airplane 5 6 0.8 0.667 0.8 0.612
bus 5 7 0.556 0.714 0.794 0.742
train 3 3 1 0.667 0.833 0.733
truck 5 12 1 0.25 0.625 0.444
boat 2 6 0.5 0.167 0.374 0.0748
traffic light 4 14 0.667 0.143 0.429 0.386
stop sign 2 2 1 0.5 0.75 0.6
bench 5 9 1 0.222 0.611 0.412
bird 2 16 0.917 0.688 0.825 0.493
cat 4 4 1 1 0.995 0.821
dog 9 9 0.615 0.889 0.766 0.627
horse 1 2 0.667 1 0.995 0.514
elephant 4 17 0.929 0.765 0.874 0.706
bear 1 1 1 1 0.995 0.995
zebra 2 4 1 1 0.995 0.973
giraffe 4 9 0.889 0.889 0.926 0.712
backpack 4 6 0.667 0.333 0.556 0.364
umbrella 4 18 1 0.444 0.722 0.527
handbag 9 19 0 0 0 0
tie 6 7 0.833 0.714 0.785 0.583
suitcase 2 4 0.5 0.5 0.539 0.479
frisbee 5 5 0.8 0.8 0.839 0.775
skis 1 1 1 1 0.995 0.497
snowboard 2 7 0.8 0.571 0.744 0.555
sports ball 6 6 0.667 0.333 0.556 0.369
kite 2 10 0.8 0.4 0.619 0.271
baseball bat 4 4 1 0.25 0.625 0.375
baseball glove 4 7 0.75 0.429 0.642 0.465
skateboard 3 5 1 0.6 0.8 0.563
tennis racket 5 7 0.667 0.286 0.524 0.424
bottle 6 18 0.545 0.333 0.382 0.27
wine glass 5 16 0.714 0.312 0.559 0.45
cup 10 36 0.667 0.278 0.503 0.419
fork 6 6 1 0.167 0.583 0.525
knife 7 16 0.778 0.438 0.656 0.481
spoon 5 22 0.667 0.182 0.438 0.275
bowl 9 28 0.692 0.643 0.689 0.55
banana 1 1 0 0 0 0
sandwich 2 2 0.333 0.5 0.25 0.25
orange 1 4 1 0.25 0.625 0.312
broccoli 4 11 0.5 0.182 0.388 0.365
carrot 3 24 0.786 0.458 0.63 0.434
hot dog 1 2 0.5 0.5 0.622 0.622
pizza 5 5 0.833 1 0.995 0.865
donut 2 14 0.667 1 0.94 0.873
cake 4 4 0.8 1 0.995 0.904
chair 9 35 0.6 0.514 0.503 0.291
couch 5 6 0.6 0.5 0.648 0.561
potted plant 9 14 0.692 0.643 0.746 0.541
bed 3 3 1 0.667 0.833 0.733
dining table 10 13 0.4 0.462 0.514 0.422
toilet 2 2 1 0.5 0.75 0.75
tv 2 2 0.5 0.5 0.622 0.622
laptop 2 3 0 0 0 0
mouse 2 2 0 0 0 0
remote 5 8 1 0.5 0.75 0.692
cell phone 5 8 0 0 0 0
microwave 3 3 0.667 0.667 0.777 0.744
oven 5 5 0.5 0.4 0.481 0.389
sink 4 6 0.5 0.167 0.292 0.233
refrigerator 5 5 0.667 0.4 0.598 0.489
book 6 29 0.75 0.103 0.423 0.311
clock 8 9 0.778 0.778 0.85 0.743
vase 2 2 0.4 1 0.828 0.8
scissors 1 1 0 0 0 0
teddy bear 6 21 1 0.333 0.667 0.487
toothbrush 2 5 1 0.4 0.7 0.469
Speed: 0.5ms preprocess, 5.1ms inference, 0.0ms loss, 9.1ms postprocess per image
Results saved to runs/detect/val3
원본 YOLOv8 모델 성능:
mAP50: 0.6260
mAP50-95: 0.4942
0: 640x448 1 umbrella, 2 chairs, 1 couch, 1 bed, 1 dining table, 1 book, 48.3ms
Speed: 2.3ms preprocess, 48.3ms inference, 1.7ms postprocess per image at shape (1, 3, 640, 448)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50896 (\N{HANGUL SYLLABLE WEON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 48376 (\N{HANGUL SYLLABLE BON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 47784 (\N{HANGUL SYLLABLE MO}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 45944 (\N{HANGUL SYLLABLE DEL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50696 (\N{HANGUL SYLLABLE YE}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 52769 (\N{HANGUL SYLLABLE CEUG}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44208 (\N{HANGUL SYLLABLE GYEOL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44284 (\N{HANGUL SYLLABLE GWA}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
클래스: chair, 신뢰도: 0.60, 바운딩 박스: [197.60, 334.23, 313.21, 528.42] 클래스: chair, 신뢰도: 0.57, 바운딩 박스: [0.00, 274.12, 50.74, 378.62] 클래스: couch, 신뢰도: 0.50, 바운딩 박스: [260.07, 142.78, 386.80, 390.63] 클래스: book, 신뢰도: 0.46, 바운딩 박스: [172.28, 279.28, 250.58, 328.04] 클래스: bed, 신뢰도: 0.44, 바운딩 박스: [260.65, 142.66, 386.50, 392.55] 클래스: umbrella, 신뢰도: 0.40, 바운딩 박스: [15.75, 0.23, 384.39, 132.08] 클래스: dining table, 신뢰도: 0.30, 바운딩 박스: [69.72, 274.91, 305.50, 511.90] 클래스 수: 80 클래스 목록: 0: person 1: bicycle 2: car 3: motorcycle 4: airplane 5: bus 6: train 7: truck 8: boat 9: traffic light 10: fire hydrant 11: stop sign 12: parking meter 13: bench 14: bird 15: cat 16: dog 17: horse 18: sheep 19: cow 20: elephant 21: bear 22: zebra 23: giraffe 24: backpack 25: umbrella 26: handbag 27: tie 28: suitcase 29: frisbee 30: skis 31: snowboard 32: sports ball 33: kite 34: baseball bat 35: baseball glove 36: skateboard 37: surfboard 38: tennis racket 39: bottle 40: wine glass 41: cup 42: fork 43: knife 44: spoon 45: bowl 46: banana 47: apple 48: sandwich 49: orange 50: broccoli 51: carrot 52: hot dog 53: pizza 54: donut 55: cake 56: chair 57: couch 58: potted plant 59: bed 60: dining table 61: toilet 62: tv 63: laptop 64: mouse 65: remote 66: keyboard 67: cell phone 68: microwave 69: oven 70: toaster 71: sink 72: refrigerator 73: book 74: clock 75: vase 76: scissors 77: teddy bear 78: hair drier 79: toothbrush
클래스: chair, 신뢰도: 0.60, 바운딩 박스: [197.60, 334.23, 313.21, 528.42]
클래스: chair, 신뢰도: 0.57, 바운딩 박스: [0.00, 274.12, 50.74, 378.62]
클래스: couch, 신뢰도: 0.50, 바운딩 박스: [260.07, 142.78, 386.80, 390.63]
클래스: book, 신뢰도: 0.46, 바운딩 박스: [172.28, 279.28, 250.58, 328.04]
클래스: bed, 신뢰도: 0.44, 바운딩 박스: [260.65, 142.66, 386.50, 392.55]
클래스: umbrella, 신뢰도: 0.40, 바운딩 박스: [15.75, 0.23, 384.39, 132.08]
클래스: dining table, 신뢰도: 0.30, 바운딩 박스: [69.72, 274.91, 305.50, 511.90]
이 결과에서 주목할 점:
신뢰도: 대부분의 탐지가 중간 정도의 신뢰도(0.4~0.6)를 가지고 있어, 모델이 완전히 확신하지는 않지만 어느 정도 신뢰할 만한 탐지를 했음을 나타냅니다.
중복 탐지: 의자가 두 번 탐지되었는데, 이는 실제로 두 개의 의자가 있거나 같은 의자를 다르게 인식했을 가능성이 있습니다.
혼동: 소파와 침대가 거의 같은 위치에 탐지되었는데, 이는 모델이 이 두 객체를 구분하는 데 어려움을 겪고 있음을 나타냅니다.
낮은 신뢰도 탐지: 식탁의 신뢰도가 30%로 가장 낮아, 이 탐지는 다소 불확실할 수 있습니다.
다양한 객체: 모델이 가구(의자, 소파, 침대, 식탁)부터 작은 물건(책)과 우산까지 다양한 종류의 객체를 탐지할 수 있음을 보여줍니다.
In [ ]:
from ultralytics import YOLO
import cv2
import matplotlib.pyplot as plt
# 원본 YOLOv8 모델 로드
model = YOLO('yolov8l.pt')
# 테스트 데이터셋 경로 (COCO 형식의 데이터셋 사용)
test_data = 'coco128_custom.yaml'
# 모델 평가
results = model.val(data=test_data, conf=0.25, iou=0.5)
# 결과 출력
print("원본 YOLOv8 모델 성능:")
print(f"mAP50: {results.box.map50:.4f}")
print(f"mAP50-95: {results.box.map:.4f}")
# 샘플 이미지에 대한 예측
sample_image = cv2.imread('life_unsplash.jpeg')
sample_results = model(sample_image)
# 결과 시각화
img_with_boxes = sample_results[0].plot()
plt.figure(figsize=(12, 8))
plt.imshow(cv2.cvtColor(img_with_boxes, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.title("원본 YOLOv8 모델 예측 결과")
plt.show()
# 예측 결과 분석
for r in sample_results:
for box in r.boxes:
class_id = int(box.cls[0])
confidence = float(box.conf[0])
x1, y1, x2, y2 = box.xyxy[0].tolist()
print(f"클래스: {model.names[class_id]}, 신뢰도: {confidence:.2f}, 바운딩 박스: [{x1:.2f}, {y1:.2f}, {x2:.2f}, {y2:.2f}]")
# 클래스 수 출력
print(f"클래스 수: {len(model.names)}")
print("클래스 목록:")
for i, name in model.names.items():
print(f"{i}: {name}")
Ultralytics 8.3.16 🚀 Python-3.10.12 torch-2.4.1+cu121 CUDA:0 (Tesla T4, 15102MiB) YOLOv8l summary (fused): 268 layers, 43,668,288 parameters, 0 gradients, 165.2 GFLOPs
val: Scanning /content/coco128/labels/train2017.cache... 126 images, 2 backgrounds, 0 corrupt: 100%|██████████| 128/128 [00:00<?, ?it/s]
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 8/8 [00:05<00:00, 1.34it/s]
all 128 929 0.788 0.723 0.793 0.667
person 61 254 0.88 0.748 0.846 0.726
bicycle 3 6 1 0.333 0.667 0.55
car 12 46 0.955 0.457 0.711 0.473
motorcycle 4 5 0.833 1 0.962 0.853
airplane 5 6 1 1 0.995 0.986
bus 5 7 0.833 0.714 0.833 0.762
train 3 3 1 1 0.995 0.973
truck 5 12 0.6 0.5 0.566 0.393
boat 2 6 1 0.667 0.833 0.595
traffic light 4 14 1 0.214 0.607 0.423
stop sign 2 2 1 1 0.995 0.895
bench 5 9 0.857 0.667 0.809 0.715
bird 2 16 1 1 0.995 0.734
cat 4 4 1 1 0.995 0.926
dog 9 9 0.818 1 0.995 0.901
horse 1 2 1 1 0.995 0.846
elephant 4 17 1 0.941 0.971 0.887
bear 1 1 1 1 0.995 0.895
zebra 2 4 1 1 0.995 0.98
giraffe 4 9 0.889 0.889 0.938 0.793
backpack 4 6 1 0.667 0.833 0.627
umbrella 4 18 0.889 0.889 0.928 0.7
handbag 9 19 0.667 0.316 0.514 0.406
tie 6 7 0.833 0.714 0.833 0.705
suitcase 2 4 1 0.75 0.875 0.732
frisbee 5 5 0.8 0.8 0.879 0.793
skis 1 1 1 1 0.995 0.895
snowboard 2 7 1 0.857 0.928 0.834
sports ball 6 6 0.6 0.5 0.648 0.423
kite 2 10 0.6 0.3 0.508 0.146
baseball bat 4 4 0.8 1 0.995 0.513
baseball glove 4 7 0.667 0.571 0.685 0.465
skateboard 3 5 0.6 0.6 0.668 0.513
tennis racket 5 7 0.667 0.571 0.715 0.272
bottle 6 18 0.733 0.611 0.643 0.521
wine glass 5 16 0.917 0.688 0.815 0.602
cup 10 36 0.903 0.778 0.851 0.679
fork 6 6 0.75 0.5 0.604 0.457
knife 7 16 0.8 0.75 0.816 0.643
spoon 5 22 0.75 0.545 0.671 0.567
bowl 9 28 0.846 0.786 0.852 0.771
banana 1 1 1 1 0.995 0.995
sandwich 2 2 0.667 1 0.995 0.995
orange 1 4 0 0 0 0
broccoli 4 11 0.5 0.182 0.388 0.365
carrot 3 24 0.867 0.542 0.708 0.564
hot dog 1 2 0.667 1 0.995 0.995
pizza 5 5 0.833 1 0.995 0.861
donut 2 14 0.737 1 0.967 0.904
cake 4 4 0.8 1 0.995 0.911
chair 9 35 0.7 0.8 0.814 0.616
couch 5 6 0.714 0.833 0.811 0.692
potted plant 9 14 0.769 0.714 0.801 0.68
bed 3 3 0.75 1 0.995 0.723
dining table 10 13 0.588 0.769 0.786 0.652
toilet 2 2 1 1 0.995 0.958
tv 2 2 0.667 1 0.995 0.954
laptop 2 3 1 0.333 0.667 0.667
mouse 2 2 0 0 0 0
remote 5 8 1 0.625 0.812 0.757
cell phone 5 8 0.833 0.625 0.74 0.563
microwave 3 3 1 1 0.995 0.897
oven 5 5 0.333 0.4 0.365 0.285
sink 4 6 0.429 0.5 0.406 0.293
refrigerator 5 5 0.8 0.8 0.879 0.761
book 6 29 0.727 0.276 0.523 0.396
clock 8 9 0.8 0.889 0.911 0.772
vase 2 2 0.333 1 0.995 0.995
scissors 1 1 0 0 0 0
teddy bear 6 21 0.938 0.714 0.845 0.681
toothbrush 2 5 1 1 0.995 0.844
Speed: 0.2ms preprocess, 30.2ms inference, 0.0ms loss, 1.5ms postprocess per image
Results saved to runs/detect/val4
원본 YOLOv8 모델 성능:
mAP50: 0.7933
mAP50-95: 0.6669
0: 640x448 1 chair, 1 bed, 1 dining table, 1 book, 1 scissors, 59.4ms
Speed: 3.0ms preprocess, 59.4ms inference, 2.1ms postprocess per image at shape (1, 3, 640, 448)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50896 (\N{HANGUL SYLLABLE WEON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 48376 (\N{HANGUL SYLLABLE BON}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 47784 (\N{HANGUL SYLLABLE MO}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 45944 (\N{HANGUL SYLLABLE DEL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 50696 (\N{HANGUL SYLLABLE YE}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 52769 (\N{HANGUL SYLLABLE CEUG}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44208 (\N{HANGUL SYLLABLE GYEOL}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
/usr/local/lib/python3.10/dist-packages/IPython/core/pylabtools.py:151: UserWarning: Glyph 44284 (\N{HANGUL SYLLABLE GWA}) missing from current font.
fig.canvas.print_figure(bytes_io, **kw)
클래스: chair, 신뢰도: 0.91, 바운딩 박스: [198.58, 331.67, 314.12, 531.07] 클래스: bed, 신뢰도: 0.87, 바운딩 박스: [268.38, 143.27, 386.54, 392.72] 클래스: book, 신뢰도: 0.51, 바운딩 박스: [172.49, 279.84, 249.48, 326.53] 클래스: scissors, 신뢰도: 0.41, 바운딩 박스: [151.41, 284.57, 172.81, 303.06] 클래스: dining table, 신뢰도: 0.39, 바운딩 박스: [70.52, 276.22, 309.80, 506.33] 클래스 수: 80 클래스 목록: 0: person 1: bicycle 2: car 3: motorcycle 4: airplane 5: bus 6: train 7: truck 8: boat 9: traffic light 10: fire hydrant 11: stop sign 12: parking meter 13: bench 14: bird 15: cat 16: dog 17: horse 18: sheep 19: cow 20: elephant 21: bear 22: zebra 23: giraffe 24: backpack 25: umbrella 26: handbag 27: tie 28: suitcase 29: frisbee 30: skis 31: snowboard 32: sports ball 33: kite 34: baseball bat 35: baseball glove 36: skateboard 37: surfboard 38: tennis racket 39: bottle 40: wine glass 41: cup 42: fork 43: knife 44: spoon 45: bowl 46: banana 47: apple 48: sandwich 49: orange 50: broccoli 51: carrot 52: hot dog 53: pizza 54: donut 55: cake 56: chair 57: couch 58: potted plant 59: bed 60: dining table 61: toilet 62: tv 63: laptop 64: mouse 65: remote 66: keyboard 67: cell phone 68: microwave 69: oven 70: toaster 71: sink 72: refrigerator 73: book 74: clock 75: vase 76: scissors 77: teddy bear 78: hair drier 79: toothbrush
전체적인 해석:
신뢰도 분포:
의자와 침대는 매우 높은 신뢰도(90% 이상)로 탐지되었습니다.
책은 중간 정도의 신뢰도로 탐지되었습니다.
가위와 식탁은 비교적 낮은 신뢰도로 탐지되었습니다.
객체 위치:
탐지된 객체들은 이미지의 다양한 부분에 분포되어 있습니다.
의자와 식탁이 이미지의 하단 부분에 큰 영역을 차지하고 있습니다.
가능한 시나리오:
이 이미지는 아마도 침실이나 거실의 일부를 보여주고 있을 것 같습니다.
큰 의자(또는 소파)와 침대가 주요 가구로 보입니다.
책과 가위 같은 작은 물건들이 테이블이나 다른 표면 위에 있을 가능성이 있습니다.
주의점:
가위와 식탁의 탐지는 신뢰도가 낮아 실제로 존재하지 않을 수 있습니다.
식탁으로 탐지된 것은 실제로는 다른 큰 가구(예: 커피 테이블)일 수 있습니다.
모델 성능:
모델은 큰 가구들(의자, 침대)을 매우 확실하게 인식하고 있습니다.
작은 물체(책, 가위)에 대해서는 덜 확신하고 있습니다.
일부 객체(식탁)에 대해서는 오탐지의 가능성이 있습니다.
모델 요약¶
In [ ]:
from ultralytics import YOLO
import os
# 모델 로드
all_model_list = ['yolov8n', 'yolov8s', 'yolov8m', 'yolov8l', 'yolov8x']
for one_model in all_model_list:
model = YOLO(f"{one_model}.pt")
# 모델 요약 정보 출력
print(f"\n{one_model} 모델 정보:")
# print(model)
# 모델 레이어 수 확인
num_layers = len(list(model.model.parameters()))
print(f"총 레이어 수: {num_layers}")
# 모델 파라미터 수 확인
num_params = sum(p.numel() for p in model.model.parameters())
print(f"총 파라미터 수: {num_params:,}")
# 모델 파일 용량 확인
model_path = f"{one_model}.pt"
file_size = os.path.getsize(model_path) / (1024 ** 2) # 파일 크기를 MB 단위로 변환
print(f"모델 파일 크기: {file_size:.2f} MB")
yolov8n 모델 정보: 총 레이어 수: 184 총 파라미터 수: 3,157,200 모델 파일 크기: 6.25 MB yolov8s 모델 정보: 총 레이어 수: 184 총 파라미터 수: 11,166,560 모델 파일 크기: 21.54 MB yolov8m 모델 정보: 총 레이어 수: 244 총 파라미터 수: 25,902,640 모델 파일 크기: 49.72 MB yolov8l 모델 정보: 총 레이어 수: 304 총 파라미터 수: 43,691,520 모델 파일 크기: 83.73 MB yolov8x 모델 정보: 총 레이어 수: 304 총 파라미터 수: 68,229,648 모델 파일 크기: 130.55 MB
In [ ]:
from ultralytics import YOLO
# 모델 로드
model = YOLO("yolov8n.pt") # 'yolov8n', 'yolov8s', 'yolov8m', 'yolov8l', 'yolov8x' 중 선택
# 모델 레이어 수 확인
num_layers = len(list(model.model.parameters()))
print(f"YOLOv8n 모델의 총 레이어 수: {num_layers}")
# 모델 요약
print(model.model)
YOLOv8n 모델의 총 레이어 수: 184
DetectionModel(
(model): Sequential(
(0): Conv(
(conv): Conv2d(3, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(16, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(16, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): C2f(
(cv1): Conv(
(conv): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(48, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(16, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(16, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(3): Conv(
(conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(4): C2f(
(cv1): Conv(
(conv): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0-1): 2 x Bottleneck(
(cv1): Conv(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(5): Conv(
(conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(6): C2f(
(cv1): Conv(
(conv): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0-1): 2 x Bottleneck(
(cv1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(7): Conv(
(conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(8): C2f(
(cv1): Conv(
(conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(384, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(9): SPPF(
(cv1): Conv(
(conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): MaxPool2d(kernel_size=5, stride=1, padding=2, dilation=1, ceil_mode=False)
)
(10): Upsample(scale_factor=2.0, mode='nearest')
(11): Concat()
(12): C2f(
(cv1): Conv(
(conv): Conv2d(384, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(192, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(13): Upsample(scale_factor=2.0, mode='nearest')
(14): Concat()
(15): C2f(
(cv1): Conv(
(conv): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(96, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(16): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(17): Concat()
(18): C2f(
(cv1): Conv(
(conv): Conv2d(192, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(192, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(19): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(20): Concat()
(21): C2f(
(cv1): Conv(
(conv): Conv2d(384, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(384, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(m): ModuleList(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(cv2): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
)
)
)
(22): Detect(
(cv2): ModuleList(
(0): Sequential(
(0): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
)
(1): Sequential(
(0): Conv(
(conv): Conv2d(128, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
)
(2): Sequential(
(0): Conv(
(conv): Conv2d(256, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
)
)
(cv3): ModuleList(
(0): Sequential(
(0): Conv(
(conv): Conv2d(64, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(80, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1))
)
(1): Sequential(
(0): Conv(
(conv): Conv2d(128, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(80, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1))
)
(2): Sequential(
(0): Conv(
(conv): Conv2d(256, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(1): Conv(
(conv): Conv2d(80, 80, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
(act): SiLU(inplace=True)
)
(2): Conv2d(80, 80, kernel_size=(1, 1), stride=(1, 1))
)
)
(dfl): DFL(
(conv): Conv2d(16, 1, kernel_size=(1, 1), stride=(1, 1), bias=False)
)
)
)
)
In [ ]: