win10下使用mmdet训练自己的数据模型
win10下使用mmdet训练自己的数据模型
- 1.环境配置
- 2.制作自己的coco数据集
- 3.进行训练
- 4.计算测试图像的交并比
- 参考文献
1.环境配置
1.查看自己cuda版本：
2.查看自己python版本
3.安装pytorch
官方地址，按自己的选择复制粘贴到自己的python虚拟环境中。安装完之后在自己的虚拟环境中打开python，输入import torch，如果没有报错，说明自己的pytorch安装成功。
4.安装mmcv和mmdet
在自己虚拟环境中输入pip install mmcv（或者pip install mmcv-full）和pip install mmdet
。如果自己的电脑报错的话可以先安装pycocotools电脑要有visual C++。不过我的电脑没有报错，应该是新版本的会自己下载吧。其实按照git上的步骤应该就是没有问题的。
5.测试是否安装成功下载权重文件，我把他放在了新建的文件夹checkoints中，然后修改并运行以下代码
# coding=utf-8
#改三处
import mmcv
from mmdet.apis import init_detector
from mmdet.apis import inference_detector
from mmdet.apis import show_result_pyplot
# from mmdet.models.detectors.base import BaseDetector# 模型配置文件
config_file = r'C:\Users\ROBOT-773\Desktop\Downloads\mmdetection-master\mmdetection-master\configs\faster_rcnn\faster_rcnn_r50_fpn_1x_coco.py'#1.改1，这里是你fork的文件夹里面的config文件
# 预训练模型文件
# url: https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
checkpoint_file = r'C:\Users\ROBOT-773\Desktop\Downloads\mmdetection-master\mmdetection-master\checkpoints\faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth'#改2，这里是刚刚下载的权重文件
# 通过模型配置文件与预训练文件构建模型
model = init_detector(config_file, checkpoint_file, device='cuda:0')
# 测试单张图片并进行展示
# img = '/home/cv/mmdetection/my_pictures/cars1.jpeg'
img=r'C:\Users\ROBOT-773\Desktop\Downloads\mmdetection-master\mmdetection-master\demo\demo.jpg'#改3，这里是图片的地址
result = inference_detector(model, img)
show_result_pyplot(model, img, result)# 测试一个图像列表并保存结果图像
# imgs = ['test1.jpg', 'test2.jpg', 'test3.jpg']
# for i, result in enumerate(inference_detector(model, imgs)):
# show_result_pyplot(model, imgs[i], result)"""
#
# (3)测试视频和显示测试结果
video = mmcv.VideoReader('demo/Venice-2.mp4')
for frame in video:result = inference_detector(model, frame)show_result(frame, result, model.CLASSES, wait_time=1)
"""
测试成功，下载完成，进行下一步。
2.制作自己的coco数据集
应该是voc和coco数据集都可以，我自己是用了coco数据集。将自己的图片分成训练集和测试集，然后用labelimg分别生成各自的xml文件，然后用下列代码，将xml文件生成json文件
import xml.etree.ElementTree as ET
import os
import jsoncoco = dict()
coco['images'] = []
coco['type'] = 'instances'
coco['annotations'] = []
coco['categories'] = []category_set = dict()
image_set = set()category_item_id = 0
image_id = 20210000000
annotation_id = 0def addCatItem(name):global category_item_idcategory_item = dict()category_item['supercategory'] = 'none'category_item_id += 1category_item['id'] = category_item_idcategory_item['name'] = namecoco['categories'].append(category_item)category_set[name] = category_item_idreturn category_item_iddef addImgItem(file_name, size):global image_idif file_name is None:raise Exception('Could not find filename tag in xml file.')if size['width'] is None:raise Exception('Could not find width tag in xml file.')if size['height'] is None:raise Exception('Could not find height tag in xml file.')image_id += 1image_item = dict()image_item['id'] = image_idimage_item['file_name'] = file_nameimage_item['width'] = size['width']image_item['height'] = size['height']coco['images'].append(image_item)image_set.add(file_name)return image_iddef addAnnoItem(object_name, image_id, category_id, bbox):global annotation_idannotation_item = dict()annotation_item['segmentation'] = []seg = []# bbox[] is x,y,w,h# left_topseg.append(bbox[0])seg.append(bbox[1])# left_bottomseg.append(bbox[0])seg.append(bbox[1] + bbox[3])# right_bottomseg.append(bbox[0] + bbox[2])seg.append(bbox[1] + bbox[3])# right_topseg.append(bbox[0] + bbox[2])seg.append(bbox[1])annotation_item['segmentation'].append(seg)annotation_item['area'] = bbox[2] * bbox[3]annotation_item['iscrowd'] = 0annotation_item['ignore'] = 0annotation_item['image_id'] = image_idannotation_item['bbox'] = bboxannotation_item['category_id'] = category_idannotation_id += 1annotation_item['id'] = annotation_idcoco['annotations'].append(annotation_item)def parseXmlFiles(xml_path):for f in os.listdir(xml_path):if not f.endswith('.xml'):continuebndbox = dict()size = dict()current_image_id = Nonecurrent_category_id = Nonefile_name = Nonesize['width'] = Nonesize['height'] = Nonesize['depth'] = Nonexml_file = os.path.join(xml_path, f)print(xml_file)tree = ET.parse(xml_file)root = tree.getroot()if root.tag != 'annotation':raise Exception('pascal voc xml root element should be annotation, rather than {}'.format(root.tag))# elem is , , , for elem in root:current_parent = elem.tagcurrent_sub = Noneobject_name = Noneif elem.tag == 'folder':continueif elem.tag == 'filename':file_name = elem.textif file_name in category_set:raise Exception('file_name duplicated')# add img item only after parse tag elif current_image_id is None and file_name is not None and size['width'] is not None:if file_name not in image_set:current_image_id = addImgItem(file_name, size)print('add image with {} and {}'.format(file_name, size))else:raise Exception('duplicated image: {}'.format(file_name))# subelem is , , , , for subelem in elem:bndbox['xmin'] = Nonebndbox['xmax'] = Nonebndbox['ymin'] = Nonebndbox['ymax'] = Nonecurrent_sub = subelem.tagif current_parent == 'object' and subelem.tag == 'name':object_name = subelem.textif object_name not in category_set:current_category_id = addCatItem(object_name)else:current_category_id = category_set[object_name]elif current_parent == 'size':if size[subelem.tag] is not None:raise Exception('xml structure broken at size tag.')size[subelem.tag] = int(subelem.text)# option is , , , , when subelem is for option in subelem:if current_sub == 'bndbox':if bndbox[option.tag] is not None:raise Exception('xml structure corrupted at bndbox tag.')bndbox[option.tag] = int(option.text)# only after parse the if bndbox['xmin'] is not None:if object_name is None:raise Exception('xml structure broken at bndbox tag')if current_image_id is None:raise Exception('xml structure broken at bndbox tag')if current_category_id is None:raise Exception('xml structure broken at bndbox tag')bbox = []# xbbox.append(bndbox['xmin'])# ybbox.append(bndbox['ymin'])# wbbox.append(bndbox['xmax'] - bndbox['xmin'])# hbbox.append(bndbox['ymax'] - bndbox['ymin'])print('add annotation with {},{},{},{}'.format(object_name, current_image_id, current_category_id,bbox))addAnnoItem(object_name, current_image_id, current_category_id, bbox)if __name__ == '__main__':xml_path = r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test'json_file = r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test.json'parseXmlFiles(xml_path)json.dump(coco, open(json_file, 'w'))
3.进行训练
1.先fork商汤的mmdetection，运行python setup.py install。
2.在mmdetction文件下新建data/coco文件夹，如图
3.修改文件（这里我用faster_rcnn_r50_fpn_1x_coco.py文件为例，其实就是修改这个py文件里面的num_classes和classes）
3.1修改mmdetection/mmdet/datasets/coco.py文件里面的CLASSES为自己标签。
3.2修改mmdetection/mmdet/core/evaluation/class_names.py文件里面的coco_classes为自己的标签。
3.3修改F:\mmdetection\mmdetection\configs_base_\models\faster_rcnn_r50_fpn.py的num_classes为自己的类别个数。
3.4修改F:\mmdetection\mmdetection\configs_base_\datasets\coco_detection.py里面的如图所示，还有data_root最好换成绝对路径。
然后再运行一次python setup.py install,再运行如下进行训练。
python tools/train.py configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py
测试模型
python tools/test.py configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py work_dirs/faster_rcnn_r50_fpn_1x_coco/latest.pth --eval bbox --show
可以查看xml文件画到图片上的效果，用xml_to_jpg.py
import os
import cv2 as cv
import xml.etree.ElementTree as ETdef xml_to_jpg(imgs_path, xmls_path, out_path):imgs_list = os.listdir(imgs_path) #读取图片列表xmls_list = os.listdir(xmls_path) # 读取xml列表if len(imgs_list) <= len(xmls_list): #若图片个数小于或等于xml个数，从图片里面找与xml匹配的for imgName in imgs_list:temp1 = imgName.split('.')[0] #图片名 例如123.jpg 分割之后 temp1 = 123temp1_ = imgName.split('.')[1] #图片后缀if temp1_!='jpg':continuefor xmlName in xmls_list: #遍历xml列表，temp2 = xmlName.split('.')[0] #xml名temp2_ = xmlName.split('.')[1]if temp2_ != 'xml':continueif temp2!=temp1: #判断图片名与xml名是否相同，不同的话跳过下面的步骤 继续找continueelse: #相同的话 开始读取xml坐标信息，并在对应的图片上画框img_path = os.path.join(imgs_path, imgName)xml_path = os.path.join(xmls_path, xmlName)img = cv.imread(img_path)labelled = imgroot = ET.parse(xml_path).getroot()for obj in root.iter('object'):bbox = obj.find('bndbox')xmin = int(bbox.find('xmin').text.strip())ymin = int(bbox.find('ymin').text.strip())xmax = int(bbox.find('xmax').text.strip())ymax = int(bbox.find('ymax').text.strip())labelled = cv.rectangle(labelled, (xmin, ymin), (xmax, ymax), (0, 0, 255), 2)cv.imwrite(out_path + '\\' +imgName, labelled)breakelse: # 若xml个数小于图片个数，从xml里面找与图片匹配的。下面操作与上面差不多for xmlName in xmls_list:temp1 = xmlName.split('.')[0]temp1_ = xmlName.split('.')[1]if temp1_ != 'xml':continuefor imgName in imgs_list:temp2 = imgName.split('.')[0]temp2_ = imgName.split('.')[1] # 图片后缀if temp2_ != 'jpg':continueif temp2 != temp1:continueelse:img_path = os.path.join(imgs_path, imgName)xml_path = os.path.join(xmls_path, xmlName)img = cv.imread(img_path)labelled = imgroot = ET.parse(xml_path).getroot()for obj in root.iter('object'):bbox = obj.find('bndbox')xmin = int(bbox.find('xmin').text.strip())ymin = int(bbox.find('ymin').text.strip())xmax = int(bbox.find('xmax').text.strip())ymax = int(bbox.find('ymax').text.strip())labelled = cv.rectangle(labelled, (xmin, ymin), (xmax, ymax), (0, 0, 255), 1)cv.imwrite(out_path + '\\' +imgName, labelled)break
if __name__ == '__main__':
# 使用英文路径，中文路径读不进来imgs_path =r'C:\Users\ROBOT-773\Desktop\cs\KM1\3c_img' #图片路径xmls_path = r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test' #xml路径retangele_img_path =r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test_label' #保存画框后图片的路径xml_to_jpg(imgs_path, xmls_path, retangele_img_path)4.计算测试图像的交并比
先测试图片，并保存测试的bbox值
#改6处
from mmdet.apis import inference_detector, init_detector
import cv2
import os
import numpy as np
import pandas as pd
import csv
from tqdm import tqdm
import time
os.environ["CUDA_VISIBLE_DEVICES"] = "7"class Model():def __init__(self, root_config, root_checkpoint, **kwargs):self.model = init_detector(root_config, root_checkpoint) # 模型初始化self.thr_ok = kwargs.get('thr_ok', 0.05)self.classes = kwargs.get('classes', None)self.color = self.get_color()self.img_foramt = ['.jpg', '.JPG', '.bmp', '.png']def get_color(self):color = dict(red=(0, 0, 255),green=(0, 255, 0),blue=(255, 0, 0),cyan=(255, 255, 0),yellow=(0, 255, 255),magenta=(255, 0, 255),white=(255, 255, 255),black=(0, 0, 0))return colordef model_test(self, result, img_name, classes, thr_ok=0.05):output_bboxes = []json_dict = []total_bbox = []for id, boxes in enumerate(result): # loop for categoriescategory_id = id + 1if len(boxes) != 0:for box in boxes: # loop for bboxconf = box[4]if conf > thr_ok:total_bbox.append(list(box) + [category_id])bboxes = np.array(total_bbox)best_bboxes = bboxesoutput_bboxes.append(best_bboxes)for bbox in best_bboxes:coord = [round(i, 2) for i in bbox[:4]]conf = bbox[4]category = classes[int(bbox[5]) - 1] if classes is not None else int(bbox[5])json_dict.append({'img_name': img_name, 'cats': category, 'bbox': coord, 'score': conf})det_df = pd.DataFrame(json_dict, columns=['img_name', 'cats', 'bbox', 'score'])return det_dfdef single_test(self, root_img): # 单张图片模型测试img = cv2.imread(root_img)model_result = inference_detector(self.model, img)return model_resultdef run(self, img_root):model_result = self.single_test(img_root)img_name = self.get_strfile(img_root, pos=-1)result_df = self.model_test(model_result, img_name, self.classes, thr_ok=self.thr_ok)img_name_lst, cat_lst, box_lst, score_lst = self.pd2lst(result_df)return img_name_lst, cat_lst, box_lst, score_lstdef pd2lst(self, result_df):img_name_lst, cat_lst, box_lst, score_lst = [], [], [], []if len(result_df) > 0:for i in range(len(result_df)):img_name_lst.append(result_df.loc[i]['img_name'])cat_lst.append(result_df.loc[i]['cats'])box_lst.append(result_df.loc[i]['bbox'])score_lst.append(result_df.loc[i]['score'])return img_name_lst, cat_lst, box_lst, score_lstdef draw_bbox(self, img, cat_lst, box_lst, score_lst,bbox_color='green',text_color='green',thickness=1,font_scale=0.5):for j, cat in enumerate(cat_lst):x1, y1, x2, y2 = np.array(box_lst[j]).astype(np.int32)bbox_color_new = self.color[bbox_color]cv2.rectangle(img, (x1, y1), (x2, y2), bbox_color_new, thickness=thickness)score = round(score_lst[j], 4)text_color_new = self.color[text_color]label_text = '{}:{}'.format(str(cat), str(score))cv2.putText(img, label_text, (x1, y1 - 2), cv2.FONT_HERSHEY_COMPLEX, font_scale, text_color_new)return img,x1,y1,x2,y2def get_strfile(self, file_str, pos=-1):# 得到file_str / or \\ 的最后一个名称endstr_f_filestr = file_str.split('\\')[pos] if '\\' in file_str else file_str.split('/')[pos]return endstr_f_filestrdef build_dir(self,out_dir):# 构建文件if not os.path.exists(out_dir):os.makedirs(out_dir)return out_dirdef show_img(self,img):import matplotlib.pyplot as pltplt.imshow(img)plt.show()def get_files_root(root):''':return: 寻找root下面有文件夹的路径，输出所有文件夹绝对路径的列表'''files_lst = [root]result_lst = files_lstif build_files(root) == []:result_lst = files_lstelse:is_while = Truefiles_all_path = [file for file in files_lst]while is_while:for file_root in files_lst:F1 = build_files(file_root)for F1 in F1:files_all_path.append(F1)is_while = False# 排除主文件夹record = np.ones((len(files_all_path)))for i, F3 in enumerate(files_all_path):F3 = files_all_path[i]for j, F4 in enumerate(files_all_path):if F3 + '\\' in F4 or F3 + '/' in F4:record[i] = 0break# 将需要循环聚集files_lst = []for i, F3 in enumerate(files_all_path):if record[i] == 1:files_lst.append(F3)# 判断是否有子文件夹for F4 in files_lst:file_judge = build_files(F4)if file_judge != []:is_while = Truebreakresult_lst = files_lstreturn result_lstdef build_files(root):''':得到该路径下的所有文件'''files = [os.path.join(root, file) for file in os.listdir(root)]files_true = []for file in files:if not os.path.isfile(file):files_true.append(file)return files_truedef single_main(model,root_img,work_dir,csv_bb):# 一张图片测试所有集合img_name_lst, cat_lst, box_lst, score_lst = model.run(root_img) #img = cv2.imread(root_img)img,x1,y1,x2,y2 = model.draw_bbox(img, cat_lst, box_lst, score_lst)file_name = model.get_strfile(root_img, pos=-2)out_file = model.build_dir(os.path.join(work_dir, file_name))img_name = model.get_strfile(root_img, pos=-1)cv2.imwrite(os.path.join(out_file, img_name), img)csv_bb.append([img_name,x1,y1,x2,y2])return csv_bbdef main(root,model,work_dir):root_files=get_files_root(root)csv_bb = []num=0for file_path in tqdm(root_files):for name in tqdm(os.listdir(file_path)):if name[-4:] in model.img_foramt:root_img=os.path.join(file_path,name)csv_bb=single_main(model,root_img,work_dir,csv_bb)num+=1print('num of images:',num)f = open(r'F:\mmdetection\mmdetection\data\coco\test2021_gen.csv', 'w')#改6，生成目标框xmin,ymin,xmax,ymax的csvwriter = csv.writer(f)for i in csv_bb:writer.writerow(i)f.close()if __name__ == '__main__':root_config = r'F:\mmdetection\mmdetection\configs\faster_rcnn\faster_rcnn_r50_fpn_1x_coco.py'#改1，配置文件root_checkpoint = r'F:\mmdetection\mmdetection\work_dirs\faster_rcnn_r50_fpn_1x_coco\latest.pth'#改2，模型生成文件root = r'F:\mmdetection\mmdetection\data\coco\test2021' # 改3，测试文件夹work_dir=r'F:\mmdetection\mmdetection\data\coco\test2021_gen'#改4，测试生成文件夹info = {'classes': None}#改5，目标名称time_start = time.time()model = Model(root_config, root_checkpoint, **info) # 类实列化，也是初始化# img_name_lst, cat_lst, box_lst, score_lst = model.run(root_img) #单张图片的预测main(root, model, work_dir)time_end=time.time()time_gap=time_end-time_startprint('time gap:',time_gap)
然后读取测试文件的xml和测试出来的csv文件，计算交并比。
import csv
import pandas as pd
import xml.etree.ElementTree as ET
import os
import glob
#读xml文件的bbox
os.chdir(r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test1')
def xml_bbox(path):xml_list = []for xml_file in glob.glob(path + '\\*.xml'):tree = ET.parse(xml_file)#print(tree)root = tree.getroot()#hh=root.findall('object')#print(hh[0][0].text)#print(root.find('size')[0])for member in root.findall('object'):#print(member,member[0].text)try:value = (root.find('filename').text,int(root.find('size')[0].text),int(root.find('size')[1].text),member[0].text,int(member[4][0].text),int(member[4][1].text),int(member[4][2].text),int(member[4][3].text))except:pass#print(value)xml_list.append(value)#column_name = ['filename', 'width', 'height', 'class', 'xmin', 'ymin', 'xmax', 'ymax']#xml_df = pd.DataFrame(xml_list, columns=column_name)return xml_list
#读csv文件的bbox
def csv_bbox(path,name):with open(path,'r') as f:csv.list=[]reader=csv.reader(f)for row in reader:if row == []:continueif row[0] == name:return row#根据两个bbox计算交并比def calculate_IoU(predicted_bound, ground_truth_bound):px1, py1, px2, py2 = map(float,predicted_bound)print("预测框P的坐标是：({}, {}, {}, {})".format(px1, py1, px2, py2))gx1, gy1, gx2, gy2 = map(float,ground_truth_bound)print("原标记框G的坐标是：({}, {}, {}, {})".format(gx1, gy1, gx2, gy2))parea = (px2 - px1) * (py2 - py1) # 计算P的面积garea = (gx2 - gx1) * (gy2 - gy1) # 计算G的面积print("预测框P的面积是：{}；原标记框G的面积是：{}".format(parea, garea))x1 = max(px1, gx1)y1 = max(py1, gy1)x2 = min(px2, gx2)y2 = min(py2, gy2)w = x2 - x1h = y2 - y1if w <=0 or h <= 0:return 0area = w * h # G∩P的面积print("G∩P的面积是：{}".format(area))# 并集的面积 = 两个矩形面积 - 交集面积IoU = area / (parea + garea - area)return IoU
def main():iou_list=[]xml_path = r'C:\Users\ROBOT-773\Desktop\cs\KM1\XML\test1'csv_path=r'F:\mmdetection\mmdetection\data\coco\test2021_gen.csv'#xmls_list = os.listdir(xml_path) # 读取xml列表#for xmlname in xmls_list:# temp2 = xmlname.split('.')[0] # xml名xml_list = xml_bbox(xml_path)for xml in xml_list:csv_df = csv_bbox(csv_path,xml[0])IoU = calculate_IoU( (xml[4],xml[5],xml[6],xml[7]), (csv_df[1], csv_df[2], csv_df[3], csv_df[4]))iou_list.append(IoU)print("IoU是：{}".format(IoU))print(iou_list)print(sum(iou_list)/350)
if __name__ =="__main__":main()
参考文献
【1】https://blog.csdn.net/weixin_42341590/article/details/119511297
【2】https://blog.csdn.net/qq_41690533/article/details/115868856
【3】https://blog.csdn.net/qq_33897832/article/details/103995636
【4】https://www.cnblogs.com/tangjunjun/p/14847291.html
本文来自互联网用户投稿，文章观点仅代表作者本人，不代表本站立场，不承担相关法律责任。如若转载，请注明出处。 如若内容造成侵权/违法违规/事实不符，请点击【内容举报】进行投诉反馈！