1.YOLOv8-seg简介
YOLOv8-seg是YOLO系列模型的其中一个版本。YOLOv8-seg在继承YOLO系列模型高效性和准确性的基础上，增加了实例分割的能力。
2.数据集
使用的数据集较简单，主要以下目录:
images:存放原始图片(1500张)，大小为128x128。部分如下:
images_json:存放labelme标注的json文件与原图。部分图如下：
masks:存放单通道掩码
mask_txt:存放masks中每个标签掩码图对应的每个像素值
palette_mask:存放标签掩码图的调色板图或伪彩色图。
事实上，本次训练任务只需要images与images_json。
3.下载安装包
需要下载ultralytics，github下载或者pip安装(pip安装只有ultralytics)，建议github下载，里面内容更全，包括例子与说明。
官网github地址:https://github.com/ultralytics/ultralytics
官网文档:https://docs.ultralytics.com/
下载后，主要关注examples与ultralytics
3.获取YOLOV8-seg训练的数据集格式及文件
YOLOV8-seg模型在进行实例分割时，首先会执行目标检测以识别图像中的物体，然后再对这些物体进行分割。故训练时需要分割预训练权重yolov8n-seg.pt的同时，也需要对应的目标检测yolov8n.pt权重。如果网络良好可以不用下载，当程序检测到没有这些文件时，会自动下载。关于这两个文件直接去官网下载或者网上下载(如图)，这里也给个百度盘的链接: 链接：https://pan.baidu.com/s/1Tkzi8bflpIuGTIqR18AFOg提取码：hniz
3.1划分数据集与生成yaml文件
# -*- coding: utf-8 -*-
from tqdm import tqdm
import shutil
import random
import os
import argparse
from collections import Counter
import yaml
import json
# 检查文件夹是否存在
def mkdir(path):
if not os.path.exists(path):
os.makedirs(path)
def convert_to_polygon(point1,point2):
x1, y1 = point1
x2, y2 = point2
return [[x1,y1],[x2,y1],[x2,y2],[x1,y2]]
def convert_label_json(json_dir, save_dir, classes):
json_paths = os.listdir(json_dir)
classes = classes.split(',')
mkdir(save_dir)
for json_path in tqdm(json_paths):
# for json_path in json_paths:
path = os.path.join(json_dir, json_path)
with open(path, 'r') as load_f:
json_dict = json.load(load_f)
h, w = json_dict['imageHeight'], json_dict['imageWidth']
# save txt path
txt_path = os.path.join(save_dir, json_path.replace('json', 'txt'))
txt_file = open(txt_path, 'w')
for shape_dict in json_dict['shapes']:
shape_type = shape_dict.get('shape_type',None)
label = shape_dict['label']
label_index = classes.index(label)
points = shape_dict['points']
if shape_type == "rectangle":
point1=points[0]
point2=points[1]
points=convert_to_polygon(point1,point2)
points_nor_list = []
for point in points:
points_nor_list.append(point[0] / w)
points_nor_list.append(point[1] / h)
points_nor_list = list(map(lambda x: str(x), points_nor_list))
points_nor_str = ' '.join(points_nor_list)
label_str = str(label_index) + ' ' + points_nor_str + '\n'
txt_file.writelines(label_str)
def get_classes(json_dir):
'''
统计路径下 JSON 文件里的各类别标签数量
'''
names = []
json_files = [os.path.join(json_dir, f) for f in os.listdir(json_dir) if f.endswith('.json')]
for json_path in json_files:
with open(json_path, 'r') as f:
data = json.load(f)
for shape in data['shapes']:
name = shape['label']
names.append(name)
result = Counter(names)
return result
def main(image_dir, json_dir, txt_dir, save_dir):
# 创建文件夹
mkdir(save_dir)
images_dir = os.path.join(save_dir, 'images')
labels_dir = os.path.join(save_dir, 'labels')
img_train_path = os.path.join(images_dir, 'train')
img_val_path = os.path.join(images_dir, 'val')
label_train_path = os.path.join(labels_dir, 'train')
label_val_path = os.path.join(labels_dir, 'val')
mkdir(images_dir)
mkdir(labels_dir)
mkdir(img_train_path)
mkdir(img_val_path)
mkdir(label_train_path)
mkdir(label_val_path)
# 数据集划分比例，训练集75%，验证集15%，测试集15%，按需修改
train_percent = 0.90
val_percent = 0.10
total_txt = os.listdir(txt_dir)
num_txt = len(total_txt)
list_all_txt = range(num_txt) # 范围 range(0, num)
num_train = int(num_txt * train_percent)
num_val = int(num_txt * val_percent)
train = random.sample(list_all_txt, num_train)
# 在全部数据集中取出train
val = [i for i in list_all_txt if not i in train]
# 再从val_test取出num_val个元素，val_test剩下的元素就是test
# val = random.sample(list_all_txt, num_val)
print("训练集数目：{}, 验证集数目：{}".format(len(train), len(val)))
for i in list_all_txt:
name = total_txt[i][:-4]
srcImage = os.path.join(image_dir, name + '.png')#如果图片是jpg就改为.jpg
srcLabel = os.path.join(txt_dir, name + '.txt')
if i in train:
dst_train_Image = os.path.join(img_train_path, name + '.png')#如果图片是jpg就改为.jpg
dst_train_Label = os.path.join(label_train_path, name + '.txt')
shutil.copyfile(srcImage, dst_train_Image)
shutil.copyfile(srcLabel, dst_train_Label)
elif i in val:
dst_val_Image = os.path.join(img_val_path, name + '.png')#如果图片是jpg就改为.jpg
dst_val_Label = os.path.join(label_val_path, name + '.txt')
shutil.copyfile(srcImage, dst_val_Image)
shutil.copyfile(srcLabel, dst_val_Label)
obj_classes = get_classes(json_dir)
classes = list(obj_classes.keys())
# 编写yaml文件
classes_txt = {i: classes[i] for i in range(len(classes))} # 标签类别
data = {
'path': os.path.join(os.getcwd(), save_dir),
'train': "images/train",
'val': "images/val",
'names': classes_txt,
'nc': len(classes)
}
with open(save_dir + '/segment.yaml', 'w', encoding="utf-8") as file:
yaml.dump(data, file, allow_unicode=True)
print("标签：", dict(obj_classes))
if __name__ == "__main__":
classes_list = 'circle,rect' # 类名
parser = argparse.ArgumentParser(description='json convert to txt params')
parser.add_argument('--image-dir', type=str, default=r'D:\software\pythonworksapce\yolo8_seg_train\data\images', help='图片地址') #图片文件夹路径
parser.add_argument('--json-dir', type=str, default=r'D:\software\pythonworksapce\yolo8_seg_train\data\json_out', help='json地址')#labelme标注的纯json文件夹路径
parser.add_argument('--txt-dir', type=str, default=r'D:\software\pythonworksapce\yolo8_seg_train\train_data\save_txt', help='保存txt文件地址')#标注的坐标的txt文件存放的路径
parser.add_argument('--save-dir', default=r'D:\software\pythonworksapce\yolo8_seg_train\train_data', type=str, help='保存最终分割好的数据集地址')#segment.yaml存放的路径
parser.add_argument('--classes', type=str, default=classes_list, help='classes')
args = parser.parse_args()
json_dir = args.json_dir
txt_dir = args.txt_dir
image_dir = args.image_dir
save_dir = args.save_dir
classes = args.classes
# json格式转txt格式
convert_label_json(json_dir, txt_dir, classes)
# 划分数据集，生成yaml训练文件
main(image_dir, json_dir, txt_dir, save_dir)
上述代码中，生成的数据集，只支持多边形标注与矩形标注。
划分完后，train_data目录下将会生成如下文件:
images中有tain,val两个文件夹，每个文件夹包含原始图片
labels中有tain,val两个文件夹，每个文件夹包含每个图对应的txt文件，文件中每行的最前面为数字类别索引，后面为x1 y1 x2 y2 x3 y3 ......组成的坐标点归一化后的数据。如图：
save_txt为中间生成的，用于划分labels的
segment.yaml为训练时需要配置的文件,nc表示类别数,具体内容如下:
4.训练
from ultralytics import YOLO
if __name__ == '__main__':
model = YOLO(r"D:\software\pythonworksapce\yolo8_seg_train\yolov8n-seg.yaml",task="segment").load(r"./yolov8n-seg.pt") # build from YAML and transfer weights
results = model.train(data=r"D:\software\pythonworksapce\yolo8_seg_train\train_data\segment.yaml", epochs=200,imgsz=128, device=[0])
注意:我们写的yolov8n-seg.yaml，其实有yolov8-seg.yaml这个文件就可以了，后面的n程序会自动适配到最小的模型，可以参考yolov8-seg.yaml源文件注释。
5.转onnx模型
from ultralytics import YOLO
# Load a model
# model = YOLO("yolo11n.pt") # load an official model
model = YOLO(r"D:\software\pythonworksapce\yolo8_seg_train\runs\segment\train\weights\best.pt") # load a custom trained model
# Export the model
model.export(format="onnx")
5.onnx推理
可以使用ultralytics自带的onnx推理程序。如图：
这里我稍微添加了几个自定义的函数,推理代码及结果如下:
import argparse
import os
from datetime import datetime
import cv2
import numpy as np
import onnxruntime as ort
from ultralytics.utils import ASSETS, yaml_load
from ultralytics.utils.checks import check_yaml
from ultralytics.utils.plotting import Colors
class YOLOv8Seg:
"""YOLOv8 segmentation model."""
def __init__(self, onnx_model, yaml_path="coco128.yaml"):
"""
Initialization.
Args:
onnx_model (str): Path to the ONNX model.
"""
# Build Ort session
self.session = ort.InferenceSession(onnx_model,
providers=['CUDAExecutionProvider', 'CPUExecutionProvider']
if ort.get_device() == 'GPU' else ['CPUExecutionProvider'])
# Numpy dtype: support both FP32 and FP16 onnx model
self.ndtype = np.half if self.session.get_inputs()[0].type == 'tensor(float16)' else np.single
# Get model width and height(YOLOv8-seg only has one input)
self.model_height, self.model_width = [x.shape for x in self.session.get_inputs()][0][-2:]
# Load COCO class names
self.classes = yaml_load(check_yaml(yaml_path))['names']
# Create color palette
self.color_palette = Colors()
def __call__(self, im0, conf_threshold=0.4, iou_threshold=0.45, nm=32):
"""
The whole pipeline: pre-process -> inference -> post-process.
Args:
im0 (Numpy.ndarray): original input image.
conf_threshold (float): confidence threshold for filtering predictions.
iou_threshold (float): iou threshold for NMS.
nm (int): the number of masks.
Returns:
boxes (List): list of bounding boxes.
segments (List): list of segments.
masks (np.ndarray): [N, H, W], output masks.
"""
# Pre-process
im, ratio, (pad_w, pad_h) = self.preprocess(im0)
print("im.shape", im.shape)
# Ort inference
preds = self.session.run(None, {self.session.get_inputs()[0].name: im})
# Post-process
boxes, segments, masks = self.postprocess(preds,
im0=im0,
ratio=ratio,
pad_w=pad_w,
pad_h=pad_h,
conf_threshold=conf_threshold,
iou_threshold=iou_threshold,
nm=nm)
return boxes, segments, masks
def preprocess(self, img):
"""
Pre-processes the input image.
Args:
img (Numpy.ndarray): image about to be processed.
Returns:
img_process (Numpy.ndarray): image preprocessed for inference.
ratio (tuple): width, height ratios in letterbox.
pad_w (float): width padding in letterbox.
pad_h (float): height padding in letterbox.
"""
# Resize and pad input image using letterbox() (Borrowed from Ultralytics)
shape = img.shape[:2] # original image shape
new_shape = (self.model_height, self.model_width)
r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
ratio = r, r
new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
pad_w, pad_h = (new_shape[1] - new_unpad[0]) / 2, (new_shape[0] - new_unpad[1]) / 2 # wh padding
if shape[::-1] != new_unpad: # resize
img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
top, bottom = int(round(pad_h - 0.1)), int(round(pad_h + 0.1))
left, right = int(round(pad_w - 0.1)), int(round(pad_w + 0.1))
img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=(114, 114, 114))
# Transforms: HWC to CHW -> BGR to RGB -> div(255) -> contiguous -> add axis(optional)
img = np.ascontiguousarray(np.einsum('HWC->CHW', img)[::-1], dtype=self.ndtype) / 255.0
img_process = img[None] if len(img.shape) == 3 else img
return img_process, ratio, (pad_w, pad_h)
def postprocess(self, preds, im0, ratio, pad_w, pad_h, conf_threshold, iou_threshold, nm=32):
"""
Post-process the prediction.
Args:
preds (Numpy.ndarray): predictions come from ort.session.run().
im0 (Numpy.ndarray): [h, w, c] original input image.
ratio (tuple): width, height ratios in letterbox.
pad_w (float): width padding in letterbox.
pad_h (float): height padding in letterbox.
conf_threshold (float): conf threshold.
iou_threshold (float): iou threshold.
nm (int): the number of masks.
Returns:
boxes (List): list of bounding boxes.
segments (List): list of segments.
masks (np.ndarray): [N, H, W], output masks.
"""
x, protos = preds[0], preds[1] # Two outputs: predictions and protos
# Transpose the first output: (Batch_size, xywh_conf_cls_nm, Num_anchors) -> (Batch_size, Num_anchors, xywh_conf_cls_nm)
x = np.einsum('bcn->bnc', x)
# Predictions filtering by conf-threshold
x = x[np.amax(x[..., 4:-nm], axis=-1) > conf_threshold]
# Create a new matrix which merge these(box, score, cls, nm) into one
# For more details about `numpy.c_()`: https://numpy.org/doc/1.26/reference/generated/numpy.c_.html
x = np.c_[x[..., :4], np.amax(x[..., 4:-nm], axis=-1), np.argmax(x[..., 4:-nm], axis=-1), x[..., -nm:]]
# NMS filtering
x = x[cv2.dnn.NMSBoxes(x[:, :4], x[:, 4], conf_threshold, iou_threshold)]
# print("x",x)
# Decode and return
if len(x) > 0:
# Bounding boxes format change: cxcywh -> xyxy
x[..., [0, 1]] -= x[..., [2, 3]] / 2
x[..., [2, 3]] += x[..., [0, 1]]
# Rescales bounding boxes from model shape(model_height, model_width) to the shape of original image
x[..., :4] -= [pad_w, pad_h, pad_w, pad_h]
x[..., :4] /= min(ratio)
# Bounding boxes boundary clamp
x[..., [0, 2]] = x[:, [0, 2]].clip(0, im0.shape[1])
x[..., [1, 3]] = x[:, [1, 3]].clip(0, im0.shape[0])
# Process masks
masks = self.process_mask(protos[0], x[:, 6:], x[:, :4], im0.shape)
# Masks -> Segments(contours)
segments = self.masks2segments(masks)
return x[..., :6], segments, masks # boxes, segments, masks
else:
return [], [], []
@staticmethod
def masks2segments(masks):
"""
It takes a list of masks(n,h,w) and returns a list of segments(n,xy) (Borrowed from
https://github.com/ultralytics/ultralytics/blob/465df3024f44fa97d4fad9986530d5a13cdabdca/ultralytics/utils/ops.py#L750)
Args:
masks (numpy.ndarray): the output of the model, which is a tensor of shape (batch_size, 160, 160).
Returns:
segments (List): list of segment masks.
"""
segments = []
for x in masks.astype('uint8'):
c = cv2.findContours(x, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)[0] # CHAIN_APPROX_SIMPLE
if c:
c = np.array(c[np.array([len(x) for x in c]).argmax()]).reshape(-1, 2)
else:
c = np.zeros((0, 2)) # no segments found
segments.append(c.astype('float32'))
return segments
@staticmethod
def crop_mask(masks, boxes):
"""
It takes a mask and a bounding box, and returns a mask that is cropped to the bounding box. (Borrowed from
https://github.com/ultralytics/ultralytics/blob/465df3024f44fa97d4fad9986530d5a13cdabdca/ultralytics/utils/ops.py#L599)
Args:
masks (Numpy.ndarray): [n, h, w] tensor of masks.
boxes (Numpy.ndarray): [n, 4] tensor of bbox coordinates in relative point form.
Returns:
(Numpy.ndarray): The masks are being cropped to the bounding box.
"""
n, h, w = masks.shape
x1, y1, x2, y2 = np.split(boxes[:, :, None], 4, 1)
r = np.arange(w, dtype=x1.dtype)[None, None, :]
c = np.arange(h, dtype=x1.dtype)[None, :, None]
return masks * ((r >= x1) * (r < x2) * (c >= y1) * (c < y2))
def process_mask(self, protos, masks_in, bboxes, im0_shape):
"""
Takes the output of the mask head, and applies the mask to the bounding boxes. This produces masks of higher quality
but is slower. (Borrowed from https://github.com/ultralytics/ultralytics/blob/465df3024f44fa97d4fad9986530d5a13cdabdca/ultralytics/utils/ops.py#L618)
Args:
protos (numpy.ndarray): [mask_dim, mask_h, mask_w].
masks_in (numpy.ndarray): [n, mask_dim], n is number of masks after nms.
bboxes (numpy.ndarray): bboxes re-scaled to original image shape.
im0_shape (tuple): the size of the input image (h,w,c).
Returns:
(numpy.ndarray): The upsampled masks.
"""
c, mh, mw = protos.shape
masks = np.matmul(masks_in, protos.reshape((c, -1))).reshape((-1, mh, mw)).transpose(1, 2, 0) # HWN
masks = np.ascontiguousarray(masks)
masks = self.scale_mask(masks, im0_shape) # re-scale mask from P3 shape to original input image shape
masks = np.einsum('HWN -> NHW', masks) # HWN -> NHW
masks = self.crop_mask(masks, bboxes)
return np.greater(masks, 0.5)
@staticmethod
def scale_mask(masks, im0_shape, ratio_pad=None):
"""
Takes a mask, and resizes it to the original image size. (Borrowed from
https://github.com/ultralytics/ultralytics/blob/465df3024f44fa97d4fad9986530d5a13cdabdca/ultralytics/utils/ops.py#L305)
Args:
masks (np.ndarray): resized and padded masks/images, [h, w, num]/[h, w, 3].
im0_shape (tuple): the original image shape.
ratio_pad (tuple): the ratio of the padding to the original image.
Returns:
masks (np.ndarray): The masks that are being returned.
"""
im1_shape = masks.shape[:2]
if ratio_pad is None: # calculate from im0_shape
gain = min(im1_shape[0] / im0_shape[0], im1_shape[1] / im0_shape[1]) # gain = old / new
pad = (im1_shape[1] - im0_shape[1] * gain) / 2, (im1_shape[0] - im0_shape[0] * gain) / 2 # wh padding
else:
pad = ratio_pad[1]
# Calculate tlbr of mask
top, left = int(round(pad[1] - 0.1)), int(round(pad[0] - 0.1)) # y, x
bottom, right = int(round(im1_shape[0] - pad[1] + 0.1)), int(round(im1_shape[1] - pad[0] + 0.1))
if len(masks.shape) < 2:
raise ValueError(f'"len of masks shape" should be 2 or 3, but got {len(masks.shape)}')
masks = masks[top:bottom, left:right]
masks = cv2.resize(masks, (im0_shape[1], im0_shape[0]),
interpolation=cv2.INTER_LINEAR) # INTER_CUBIC would be better
if len(masks.shape) == 2:
masks = masks[:, :, None]
return masks
def draw_and_visualize(self, im, bboxes, segments, vis=False, save=True):
"""
Draw and visualize results.
Args:
im (np.ndarray): original image, shape [h, w, c].
bboxes (numpy.ndarray): [n, 4], n is number of bboxes.
segments (List): list of segment masks.
vis (bool): imshow using OpenCV.
save (bool): save image annotated.
Returns:
None
"""
# Draw rectangles and polygons
im_canvas = im.copy()
for (*box, conf, cls_), segment in zip(bboxes, segments):
# draw contour and fill mask
cv2.polylines(im, np.int32([segment]), True, (255, 255, 255), 2) # white borderline
cv2.fillPoly(im_canvas, np.int32([segment]), self.color_palette(int(cls_), bgr=True))
# draw bbox rectangle
cv2.rectangle(im, (int(box[0]), int(box[1])), (int(box[2]), int(box[3])),
self.color_palette(int(cls_), bgr=True), 1, cv2.LINE_AA)
cv2.putText(im, f'{self.classes[cls_]}: {conf:.3f}', (int(box[0]), int(box[1] - 9)),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, self.color_palette(int(cls_), bgr=True), 2, cv2.LINE_AA)
# Mix image
im = cv2.addWeighted(im_canvas, 0.3, im, 0.7, 0)
# Show image
if vis:
cv2.imshow('demo', im)
cv2.waitKey(0)
cv2.destroyAllWindows()
# Save image
if save:
from datetime import datetime
# 获取当前时间
now = datetime.now()
# 格式化为 '年月日时分秒毫秒'
formatted_time = now.strftime('%Y%m%d%H%M%S') + str(now.microsecond // 1000).zfill(3)
cv2.imwrite(f'{formatted_time}.jpg', im)
####self def
def load_yolov8_seg_onnx_model(onnx_path, yaml_path):
yolov8_seg_model = YOLOv8Seg(onnx_path, yaml_path=yaml_path)
return yolov8_seg_model
####self def
def call_yolov8_seg_onnx_inference(img, yolov8_seg_model, conf=0.25, iou=0.45):
boxes, segments, _ = yolov8_seg_model(img, conf_threshold=conf, iou_threshold=iou)
return boxes, segments, _
####self def
def get_points_rect_class(boxes, segments):
for box, seg_points in zip(boxes, segments):
# print("type(seg_points)",type(seg_points))
class_index = int(box[-1])
confidence = box[-2]
# left_top
left_top_x = box[0]
left_top_y = box[1]
# right_botton
right_bottom_x = box[2]
right_bottom_y = box[3]
x = int(left_top_x)
y = int(left_top_y)
w = int(right_bottom_x - left_top_x)
h = int(right_bottom_y - left_top_y)
seg_points = seg_points.astype(int)
yield x, y, w, h, seg_points, class_index, confidence
####self def
def get_image_paths(folder_path, extension=".png", is_use_extension=False):
image_paths = []
# 遍历目录
for root, dirs, files in os.walk(folder_path):
for file in files:
# 检查文件扩展名
if file.endswith(extension) or is_use_extension == False:
# 构造完整的文件路径并添加到列表
image_path = os.path.join(root, file)
image_paths.append(image_path)
return image_paths
####self def
def get_boxes_contour(points):
contour = points.reshape((-1, 1, 2))
return contour
if __name__ == '__main__':
folder_path = r'D:\software\pythonworksapce\yolo8_seg_train\pre'
onnx_path = r'D:\software\pythonworksapce\yolo8_seg_train\runs\segment\train\weights\best.onnx'
yaml_path = r'D:\software\pythonworksapce\yolo8_seg_train\train_data\segment.yaml'
yolov8_seg_model = load_yolov8_seg_onnx_model(onnx_path, yaml_path)
images_paths = get_image_paths(folder_path, extension=".png")
for img_path in images_paths:
print("img_path", img_path)
img = cv2.imread(img_path, 1)
boxes, segments, _ = call_yolov8_seg_onnx_inference(img, yolov8_seg_model, conf=0.5, iou=0.4) #每个图的结果都在这里
if len(boxes) > 0:
yolov8_seg_model.draw_and_visualize(img, boxes, segments, vis=False, save=True)
测试的五张图效果如下:
原图
模型推理的效果图(与上图一一对应，这里使用时间命名了)
最后训了一个道路的数据集，看下效果。
数据集(几何图)链接:
通过网盘分享的文件：data.zip
链接: https://pan.baidu.com/s/1ZGnxNYz2pynRC1EtSAagjw 提取码: awie
小结:本文只是对yolov8-seg模型的训练进行了叙述，并未讲解模型结构，后续会再补充。另外本文再使用onnx推理图片时，使用了自带的ultralytics中自带的YOLOv8Seg这个类推理预测，但是也会导致程序冗余，比如会加载不需要使用的torch等包，读者可以研读代码，将核心代码提取出来，重新定义自己的前向预处理，后向结构处理函数。
若存在不足之处，欢迎评论与指正。
标签：json,masks,YOLOv8,seg,shape,path,推理,dir From： https://www.cnblogs.com/wancy/p/18442457