本文详细介绍了使用keras_retinanet库训练自定义目标检测模型的步骤，包括环境配置、数据集准备、模型训练和目标检测。通过调整参数和处理数据，最终成功训练并应用模型进行目标检测。
最近在学习 keras_retinanet ，下面就记录下用自己的数据集进行的模型训练。
大致分为以下几步：
- 自定义训练数据
- 图片目标标注
- 生成用于训练的图片名称、目标标注位置及目标类别的.csv文件
- 开始训练模型（注意参数调整）
- 转换训练好的模型
- 用转换后的模型进行目标检测
下面就一步一步介绍吧：
目录
由于我之前已经安装了vs2015、Anaconda和Pycharm，就不在此赘述了。
本机配置：
- GTX 1060 3G
- Win10 64
- vs2015
- Anaconda3 5.1.0
- Pycharm
1.下载包，安装环境。
- 从Github上下载Github ——> keras-retinanet这个仓库
- 确保你的环境里有tensorflow、numpy、keras
- 切换到当前目录下运行
pip install . --user或者直接从克隆的仓库中运行代码，但是需要运行python setup.py build_ext --inplace来首先编译Cython代码。
- 如果还不行就把keras_retinanet这整个目录拷贝到你自己环境的D:\Anaconda3-5.0.1\envs\tf-gpu\Lib\site-packages下
测试准备：点击测试模型下载，将用于测试的模型resnet50_coco_best_v2.1.0.h5放在snapshots目录下
下面就可以在jupyter notebook运行examples里的ResNet50RetinaNet.ipynb进行测试，当然也可以在jupyter notebook中将文件保存成.py格式的在pycharm里运行。
2.准备数据集
- retinanet模型训练的数据是按照VOC2007格式处理的，所以你也需要将自己的数据准备成VOC2007的格式
准备如图三个文件夹，JPEGImages放你自己准备训练模型的图片，图片名称最好是按1.jpg,2.jpg这种类型；- Annotations放图片目标的位置和类型标注的.xml文件，这个可以用Github——>labellmg
里的生成目标标注的工具
自动生成.xml的文件（注意使用时将保存路径改到Annotations文件夹）； - ImageSets里的子文件夹Main里放按比例随机抽样切分的训练集、验证集、测试集样本下标值的txt文件，这个可以用如下gen_main_txt.py自动生成
import os import random trainval_percent = 0.8 # 自定义用于训练模型的数据（训练数据和交叉验证数据之和）占全部数据的比例 train_percent = 0.8 # 自定义训练数据占训练数据交叉验证数据之和的比例 xmlfilepath = 'Annotations' txtsavepath = 'ImageSets\Main' total_xml = os.listdir(xmlfilepath) num = len(total_xml) list = range(num) tv = int(num * trainval_percent) tr = int(tv * train_percent) trainval = random.sample(list, tv) train = random.sample(trainval, tr) ftrainval = open('ImageSets/Main/trainval.txt', 'w') ftest = open('ImageSets/Main/test.txt', 'w') ftrain = open('ImageSets/Main/train.txt', 'w') fval = open('ImageSets/Main/val.txt', 'w') for i in list: name = total_xml[i][:-4]+'\n' if i in trainval: ftrainval.write(name) if i in train: ftrain.write(name) else: fval.write(name) else: ftest.write(name) ftrainval.close() ftrain.close() fval.close() ftest .close() - 生成包含文件名、目标位置、目标类型的annotations.csv文件和类别标签的classes.csv文件，这个可以用gen_csv.py自动生成
import csv import os import glob import sys class PascalVOC2CSV(object): def __init__(self, xml=[], ann_path='./annotations.csv', classes_path='./classes.csv'): ''' :param xml: 所有Pascal VOC的xml文件路径组成的列表 :param ann_path: ann_path :param classes_path: classes_path ''' self.xml = xml self.ann_path = ann_path self.classes_path = classes_path self.label = [] self.annotations = [] self.data_transfer() self.write_file() def data_transfer(self): for num, xml_file in enumerate(self.xml): try: # print(xml_file) # 进度输出 sys.stdout.write('\r>> Converting image %d/%d' % ( num + 1, len(self.xml))) sys.stdout.flush() with open(xml_file, 'r') as fp: for p in fp: if '<filename>' in p: self.filen_ame = p.split('>')[1].split('<')[0] if '<object>' in p: # 类别 d = [next(fp).split('>')[1].split('<')[0] for _ in range(9)] self.supercategory = d[0] if self.supercategory not in self.label: self.label.append(self.supercategory) # 边界框 x1 = int(d[-4]); y1 = int(d[-3]); x2 = int(d[-2]); y2 = int(d[-1]) self.annotations.append( [os.path.join('JPEGImages', self.filen_ame), x1, y1, x2, y2, self.supercategory]) except: continue sys.stdout.write('\n') sys.stdout.flush() def write_file(self, ): with open(self.ann_path, 'w', newline='') as fp: csv_writer = csv.writer(fp, dialect='excel') csv_writer.writerows(self.annotations) class_name = sorted(self.label) class_ = [] for num, name in enumerate(class_name): class_.append([name, num]) with open(self.classes_path, 'w', newline='') as fp: csv_writer = csv.writer(fp, dialect='excel') csv_writer.writerows(class_) if __name__ == "__main__": xml_file = glob.glob('./Annotations/*.xml') PascalVOC2CSV(xml_file) - 用keras-retinanet-master\keras_retinanet\bin目录下的debug.py测试数据集是否生成成功，这个程序需要在命令行执行
python D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/keras_retinanet/bin/debug.py csv D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/examples/annotations.csv D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/examples/classes.csv弹出你标注的图片说明数据准备成功。
3.训练模型
- 用keras-retinanet-master\keras_retinanet\bin目录下的train.py来训练模型，需要修改文件里import的相对路径
from keras_retinanet import layers # noqa: F401 from keras_retinanet import losses from keras_retinanet import models from keras_retinanet.callbacks import RedirectModel from keras_retinanet.callbacks.eval import Evaluate from keras_retinanet.models.retinanet import retinanet_bbox from keras_retinanet.preprocessing.csv_generator import CSVGenerator from keras_retinanet.preprocessing.kitti import KittiGenerator from keras_retinanet.preprocessing.open_images import OpenImagesGenerator from keras_retinanet.preprocessing.pascal_voc import PascalVocGenerator from keras_retinanet.utils.anchors import make_shapes_callback from keras_retinanet.utils.config import read_config_file, parse_anchor_parameters from keras_retinanet.utils.keras_version import check_keras_version from keras_retinanet.utils.model import freeze as freeze_model from keras_retinanet.utils.transform import random_transform_generator - 根据你自己GPU的性能微调参数，如果报tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[1,256,100,100]错误说明你的GPU内存不够用，可以通过降低batch-size、将image-min-side,image-max-side改小、改小网络结构等方式解决
tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[1,256,100,100] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc [[{ {node training/Adam/gradients/classification_submodel/pyramid_classification_3/convolution_grad/Conv2DBackpropInput}} = Conv2DBackpropInput[T=DT_FLOAT, _class=["loc:@training/Adam/cond_85/Switch_2"], data_format="NCHW", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:GPU:0"](training/Adam/gradients/classification_submodel/pyramid_classification_3/convolution_grad/ShapeN, pyramid_classification_3/kernel/read, training/Adam/gradients/classification_submodel/pyramid_classification_3/Relu_grad/ReluGrad)]] Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.参数调整：
- 修改workers参数（不知道为什么多线程会出错，就暂且不开启吧）
- 在命令行运行train.py文件
python D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/keras_retinanet/bin/train.py csv D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/examples/annotations.csv D:/PyCharm/PycharmProjects/tf-gpu-env/project/keras-retinanet-master/examples/classes.csv创建模型，网络结构如下
Creating model, this may take a second...
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, None, None, 3 0
__________________________________________________________________________________________________
padding_conv1 (ZeroPadding2D) (None, None, None, 3 0 input_1[0][0]
__________________________________________________________________________________________________
conv1 (Conv2D) (None, None, None, 6 9408 padding_conv1[0][0]
__________________________________________________________________________________________________
bn_conv1 (BatchNormalization) (None, None, None, 6 256 conv1[0][0]
__________________________________________________________________________________________________
conv1_relu (Activation) (None, None, None, 6 0 bn_conv1[0][0]
__________________________________________________________________________________________________
pool1 (MaxPooling2D) (None, None, None, 6 0 conv1_relu[0][0]
__________________________________________________________________________________________________
res2a_branch2a (Conv2D) (None, None, None, 6 4096 pool1[0][0]
__________________________________________________________________________________________________
bn2a_branch2a (BatchNormalizati (None, None, None, 6 256 res2a_branch2a[0][0]
__________________________________________________________________________________________________
res2a_branch2a_relu (Activation (None, None, None, 6 0 bn2a_branch2a[0][0]
__________________________________________________________________________________________________
padding2a_branch2b (ZeroPadding (None, None, None, 6 0 res2a_branch2a_relu[0][0]
__________________________________________________________________________________________________
res2a_branch2b (Conv2D) (None, None, None, 6 36864 padding2a_branch2b[0][0]
__________________________________________________________________________________________________
bn2a_branch2b (BatchNormalizati (None, None, None, 6 256 res2a_branch2b[0][0]
__________________________________________________________________________________________________
res2a_branch2b_relu (Activation (None, None, None, 6 0 bn2a_branch2b[0][0]
____________________________________________________________________________________________
最低0.47元/天 解锁文章
扫一扫
4447
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
