YOLOv8实例分割获取每个输出的面积

 官方文档传送门：

片段 -Ultralytics YOLOv8 文档https://docs.ultralytics.com/zh/tasks/segment/seg-predict官方文档示例：

from ultralytics import YOLO

# Load a model
model = YOLO('yolov8n-seg.pt')  # load an official model
model = YOLO('path/to/best.pt')  # load a custom model

# Predict with the model
results = model('https://ultralytics.com/images/bus.jpg')  # predict on an image

 results是一个列表，其中每个元素包含boxes（类别、边界框、置信度等基本信息）、masks（分割的掩码）、keypoints（关于姿态估计，分割任务时为None）、probs（关于分类，分割任务时为None）

从boxes中获取每个对象的cls（类别索引），需从张量中获取值，再手动根据yaml文件改成标签（待优化）；再从mask中获取对应对象的xy（边缘点坐标），计算面积使用cv2.contourArea()函数。

运行代码示例：

from ultralytics import YOLO
import cv2

# Load a model
model = YOLO("模型路径")  # load an official model

# Define path to the image file
source = "图像路径"

# Predict with the model
results = model(source=source, save=False)  # predict on an image

for result in results:
    boxes = result.boxes  # Boxes 对象，用于边界框输出
    masks = result.masks  # Masks 对象，用于分割掩码输出
    for box, mask in zip(boxes, masks):
        for cls, contour in zip(box.cls, mask.xy):
            class_id = cls.item() # 获取张量的值
            if class_id == 0.0: # 根据yaml文件换成自己的
                print("embryo")
            elif class_id == 1.0:
                print("cell")
            print(cv2.contourArea(contour)) # 计算轮廓面积

 其中模型对应的.yaml文件：

path: ../datasets/WSY-seg  # dataset root dir
train: images/train  # train images (relative to 'path') 128 images
val: images/train  # val images (relative to 'path') 128 images
test:  # test images (optional)

# Classes
names:
  0: embryo
  1: cell

输出结果：

image 1/1 图像路径: 640x640 1 embryo, 1 cell, 27.9ms       
Speed: 4.7ms preprocess, 27.9ms inference, 8.0ms postprocess per image at shape (1, 3, 640, 640)        
cell
34007.87353515625
embryo
76430.6640625

注意此处的embryo是包含了cell的，是不能标中空的。

批量处理图片方法：

这里是用另外一个模型，分割图像中的cell和segment。

source是文件夹路径。该代码生成一个df，包括图片名称和对应的面积。这样可以导出excel或csv。

 

import pandas as pd
from ultralytics import YOLO
import cv2

def get_last_part_of_string(string):
    last_backslash_index = string.rfind('\\')
    if last_backslash_index != -1:
        return string[last_backslash_index + 1:]
    else:
        return string

# Load a model
model = YOLO(r"E:\202306\segment\train_happy\weights\best.pt")

# Define path to the image file
source = r"E:\202306\人工评价＋问题(1)\人工评价＋问题\embryo_dataset_F25数据集\JE577_EMB7_DISCARD"

# Predict with the model
results = model(source=source, save=False, conf=0.7)

data = []  # Initialize an empty list to store the data



for result in results:
    record = {}
    boxes = result.boxes
    masks = result.masks
    path = get_last_part_of_string(result.path)
    record["Path"] = path
    cell_areas = []
    cell_areas_box = []
    segment_areas = []

    for box, mask in zip(boxes, masks):
        for cls, boxxyxy, maskContour in zip(box.cls, box.xyxy, mask.xy):
            class_id = cls.item()
            x1, y1, x2, y2 = [item.item() for item in boxxyxy]
            if class_id == 1.0:  # cell
                boxArea = (x2 - x1) * (y2 - y1)
                cell_areas_box.append(boxArea)
                cell_areas.append(cv2.contourArea(maskContour))
            elif class_id == 2.0:  # segment
                segmentArea = cv2.contourArea(maskContour)
                segment_areas.append(segmentArea)



    record["len_cell_areas"] = len(cell_areas)
    record["cell_areas"] = cell_areas
    data.append(record)

# Create DataFrame
df = pd.DataFrame(data)


print(df)

运行结果示例：

       Path  len_cell_areas                                         cell_areas
0   图片01.png               1                                [5982.572664737556]
1   图片02.png               1                                 [5917.78414849074]
2   图片03.png               1                                [5838.779726962515]
3   图片04.png               1                                [5920.048454688906]
4   图片05.png               1                                [5870.545019034616]
5   图片06.png               1                                [6221.850625191393]
6   图片07.png               1                                [6003.519054066506]
7   图片08.png               2             [4716.300056756678, 6220.655430036932]
8   图片09.png               2              [2925.302606760233, 3053.55887752166]
9   图片10.png               2            [3267.8019318568986, 2970.969230435032]
10  图片11.png               2                [11576.5380859375, 12864.990234375]
11  图片12.png               2             [2546.697752431646, 2434.669943689063]
12  图片13.png               2           [3835.7404975981335, 2332.3920037686767]
13  图片14.png               1                                [6029.749145191818]
14  图片15.png               3  [10395.81298828125, 9761.3525390625, 6403.8085...
15  图片16.png               3  [1893.9676402151235, 2125.634363901001, 996.73...
16  图片17.png               3  [1779.0463765645109, 2163.7527813625347, 940.5...
17  图片18.png               2            [2098.2091324186476, 2094.938419651953]
18  图片19.png               4  [1714.886724891694, 1636.82584444643, 762.3676...
19  图片20.png               4  [895.5303360021499, 907.3558746517228, 878.043...
20  图片21.png               5  [797.3409886407608, 1180.4745866096928, 742.86...