使用谷歌colab训练crnn模型

准备工作

使用colab训练crnn模型

训练代码

将训练代码和数据集上传至colab，注意图片要定长，不然可能会出现loss nan

安装pytorch1.2.0

pip uninstall torch
pip install torch===1.2.0 torchvision===0.4.0 -f https://download.pytorch.org/whl/torch_stable.html
import torch
print(torch.__version__)

将colab连接到google driver

!apt-get install opam
!opam init
!opam update
!opam install depext
!opam depext google-drive-ocamlfuse
!opam install google-drive-ocamlfuse

#进行授权操作
from google.colab import auth
auth.authenticate_user()
from oauth2client.client import GoogleCredentials
creds = GoogleCredentials.get_application_default()
import getpass
!/root/.opam/system/bin/google-drive-ocamlfuse -headless -id={creds.client_id} -secret={creds.client_secret} < /dev/null 2>&1 | grep URL
vcode = getpass.getpass()
!echo {vcode} | /root/.opam/system/bin/google-drive-ocamlfuse -headless -id={creds.client_id} -secret={creds.client_secret}
#!!!注意，里面的/root/.opam/system/bin/google-drive-ocamlfuse换成你自己的路径,一般来说你也会得到和我一样的结果

# 指定Google Drive云端硬盘的根目录，名为drive
!mkdir -p drive
!/root/.opam/system/bin/google-drive-ocamlfuse drive

切换到对应目录

%cd /content/drive/colab/crnn
%ls

安装需要的包

pip install tensorboardX

运行训练代码

修改yaml文件中的相关参数，例如图片路径，batchsize，开始结束epoch等

!python train.py --cfg lib/config/OWN_config.yaml

测试模型效果

!python demonew.py --image_path 1_en.jpg

这里需要将demo.py的代码进行修改，第一个代码段中的路径修改为自己的路径

    parser.add_argument('--checkpoint', type=str, default='output/checkpoints/base.pth',
                        help='the path to your checkpoints')

model = crnn.get_crnn(config).to(device)
print('loading pretrained model from {0}'.format(args.checkpoint))
##model.load_state_dict(torch.load(args.checkpoint))  ##作者训练的模型
checkpoint = torch.load(args.checkpoint, map_location='cpu') ##针对训练时新保存的模型
model.load_state_dict(checkpoint['state_dict'])