Yolov5 practice: teach object detection by hand

Abstract ：​YOLOv5 It's not a single model , It's a family of models , It includes YOLOv5s、YOLOv5m、YOLO...

This article is shared from Huawei cloud community 《YoloV5 actual combat ： Teach object detection by hand ——YoloV5》, author ： AI Ho .

Abstract

YOLOV5 Strictly speaking, it's not YOLO The fifth version of , Because it didn't get YOLO The father of Joe Redmon Recognition , But the overall performance of the test data is good . The details are as follows

YOLOv5 It's not a single model , It's a family of models , It includes YOLOv5s、YOLOv5m、YOLOv5l、YOLOv5x、YOLOv5x+TTA, It's a bit like this EfficientDet. Because I couldn't find it V5 The paper of , We can only learn it from the code . In general and YOLOV4 almost , Think of it as YOLOV5 The enhanced .

Project address ：GitHub - ultralytics/yolov5: YOLOv5 in PyTorch > ONNX > CoreML > TFLite

Training

1、 Download code

Project address ：https://github.com/ultralytics/YOLOv5, Recently, the author has updated some code .

2、 Configuration environment

matplotlib>=3.2.2numpy>=1.18.5opencv-python>=4.1.2pillowPyYAML>=5.3scipy>=1.4.1tensorboard>=2.2torch>=1.6.0torchvision>=0.7.0tqdm>=4.41.0

3、 Prepare the dataset

The dataset uses Labelme The data format of the annotation , Dataset from RSOD Two kinds of data sets, aircraft and oil tank, are obtained from the data set , And turn it into Labelme Annotated datasets .

The address of the dataset ： https://pan.baidu.com/s/1iTUpvA9_cwx1qiH8zbRmDg

Extraction code ：gr6g

perhaps ：LabelmeData.zip_yolov5 actual combat - Deep learning document resources -CSDN download

Decompress the downloaded data set and put it in the root directory of the project . Prepare for the next step of generating test data sets . Here's the picture ：

4、 Generate data set

YoloV5 The data set of is not the same as that of previous versions , Let's take a look at the converted dataset first .

The data structure is as follows ：

images Folder storage train and val Pictures of the

labels Inside the store train and val Object data for , Each of them txt Document and images The pictures inside are one-to-one .

txt The contents of the document are as follows ：

Format ： Object category x y w h

Are coordinates real coordinates , It's calculated by dividing the coordinates by the width and height , It's the ratio of width to height .

Let's write the code to generate the data set , newly build LabelmeToYoloV5.py, Then write the following code .

import osimport numpy as npimport jsonfrom glob import globimport cv2from sklearn.model_selection import train_test_splitfrom os import getcwdclasses = ["aircraft", "oiltank"]# 1. Tag path labelme_path = "LabelmeData/"isUseTest = True  #  Whether to create test Set # 3. Get the pending file files = glob(labelme_path + "*.json")files = [i.replace("\\", "/").split("/")[-1].split(".json")[0] for i in files]print(files)if isUseTest:    trainval_files, test_files = train_test_split(files, test_size=0.1, random_state=55)else:    trainval_files = files# splittrain_files, val_files = train_test_split(trainval_files, test_size=0.1, random_state=55)def convert(size, box):    dw = 1. / (size[0])    dh = 1. / (size[1])    x = (box[0] + box[1]) / 2.0 - 1    y = (box[2] + box[3]) / 2.0 - 1    w = box[1] - box[0]    h = box[3] - box[2]    x = x * dw    w = w * dw    y = y * dh    h = h * dh    return (x, y, w, h)wd = getcwd()print(wd)def ChangeToYolo5(files, txt_Name):    if not os.path.exists('tmp/'):        os.makedirs('tmp/')    list_file = open('tmp/%s.txt' % (txt_Name), 'w')    for json_file_ in files:        json_filename = labelme_path + json_file_ + ".json"        imagePath = labelme_path + json_file_ + ".jpg"        list_file.write('%s/%s\n' % (wd, imagePath))        out_file = open('%s/%s.txt' % (labelme_path, json_file_), 'w')        json_file = json.load(open(json_filename, "r", encoding="utf-8"))        height, width, channels = cv2.imread(labelme_path + json_file_ + ".jpg").shape        for multi in json_file["shapes"]:            points = np.array(multi["points"])            xmin = min(points[:, 0]) if min(points[:, 0]) > 0 else 0            xmax = max(points[:, 0]) if max(points[:, 0]) > 0 else 0            ymin = min(points[:, 1]) if min(points[:, 1]) > 0 else 0            ymax = max(points[:, 1]) if max(points[:, 1]) > 0 else 0            label = multi["label"]            if xmax <= xmin:                pass            elif ymax <= ymin:                pass            else:                cls_id = classes.index(label)                b = (float(xmin), float(xmax), float(ymin), float(ymax))                bb = convert((width, height), b)                out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')                print(json_filename, xmin, ymin, xmax, ymax, cls_id)ChangeToYolo5(train_files, "train")ChangeToYolo5(val_files, "val")ChangeToYolo5(test_files, "test")

This code will be executed in LabelmeData Generate... For each image txt Annotation data , At the same time tmp Create the training set under the folder 、 Verification set and test set txt,txt It records the path of the image , Generate for next step YoloV5 Data sets for training and testing . stay tmp New under the folder MakeData.py file , Generate the final result , The directory structure is shown in the figure below ：

open MakeData.py, Write the following code .

import shutilimport osfile_List = ["train", "val", "test"]for file in file_List:    if not os.path.exists('../VOC/images/%s' % file):        os.makedirs('../VOC/images/%s' % file)    if not os.path.exists('../VOC/labels/%s' % file):        os.makedirs('../VOC/labels/%s' % file)    print(os.path.exists('../tmp/%s.txt' % file))    f = open('../tmp/%s.txt' % file, 'r')    lines = f.readlines()    for line in lines:        print(line)        line = "/".join(line.split('/')[-5:]).strip()        shutil.copy(line, "../VOC/images/%s" % file)        line = line.replace('JPEGImages', 'labels')        line = line.replace('jpg', 'txt')        shutil.copy(line, "../VOC/labels/%s/" % file)

After execution, it can generate YoloV5 The data set used for training . give the result as follows ：

5、 Modify configuration parameters

 open voc.yaml file , Modify the configuration parameters inside train: VOC/images/train/  #  The path of the training set picture val: VOC/images/val/  #  Verify the path of the image set # number of classesnc: 2 # Category of detection , This dataset has two categories, so write 2# class namesnames: ["aircraft", "oiltank"]# The name of the category , And the transformation of data sets list Corresponding 

6、 modify train.py Parameters of

cfg Parameter is YoloV5  The configuration file for the model , The model files are stored in models Under the folder , Fill in different documents as required .weights Parameter is YoloV5 Pre training model of , and cfg Corresponding , example ：cfg The configuration is yolov5s.yaml,weights We need to configure yolov5s.ptdata Is the configuration file for the configuration dataset , We chose voc.yaml, So configuration data/voc.yaml Modify the above three parameters to start training , Other parameters are modified according to their own needs . The modified parameter configuration is as follows ：parser.add_argument('--weights', type=str, default='yolov5s.pt', help='initial weights path')parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml path')parser.add_argument('--data', type=str, default='data/voc.yaml', help='data.yaml path')

After the modification is completed , You can start training . As shown in the figure below ：

7、 See the training results

After going through 300epoch After training , We will be in runs Find the training weight file and some files of the training process under the folder . Pictured ：

test

The first thing you need to do is voc.yaml Add the path of test set in , open voc.yaml, stay val Add... After the field test: tmp/test.txt This line of code , Pictured ：

modify test.py Parameters in , The following parameters need to be modified .

parser = argparse.ArgumentParser(prog='test.py')
parser.add_argument('--weights', nargs='+', type=str, default='runs/exp7/weights/best.pt', help='model.pt path(s)')
parser.add_argument('--data', type=str, default='data/voc.yaml', help='*.data path')
parser.add_argument('--batch-size', type=int, default=2, help='size of each image batch')
parser.add_argument('--save-txt', default='True', action='store_true', help='save results to *.txt')

stay 275 That's ok modify test Methods , Add a path to save test results . So that when the test is done, it can be done in inference\images See the picture of the test , stay inference\output You can see the saved test results in .

Pictured ：

Here are the results of the run ：

 

Click to follow , The first time to learn about Huawei's new cloud technology ~