docker 运行DBNet(差分二值化网络文本检测）

尝试用DBNet进行文本检测：
官方代码了： 添加链接描述

之后会上传代码所需所有数据集和模型文件！
首先，建立镜像：

docker build -t dbnet:v1 .

dockerfile文件：
FROM zergmk2/chineseocr:pytorch1.0-gpu-py3.6
USER root
run mkdir /DBNet
copy . /DBNet
#run apt update
#run pip install --upgrade -i https://pypi.douban.com/simple pip
RUN pip install -r /DBNet/requirement1.txt -i https://pypi.douban.com/simple
#run apt install cuda-drivers -y
workdir /DBNet/

requirement文件：

运行train.py文件：

docker run --gpus all -v /....../....  -w /... dbnet:v1 python train.py experiments/seg_detector/totaltext_resnet50_deform_thre.yaml --num_gpus 2

对于数据集：
generate the txt file for train/test in our framework

import os
def get_images(img_path):
    '''
    find image files in data path
    :return: list of files found
    '''
    img_path = os.path.abspath(img_path)
    files = []
    exts = ['jpg', 'png', 'jpeg', 'JPG', 'PNG']
    for parent, dirnames, filenames in os.walk(img_path):
        for filename in filenames:
            for ext in exts:
                if filename.endswith(ext):
                    files.append(os.path.join(parent, filename))
                    break
    print('Find {} images'.format(len(files)))
    return sorted(files)
    
def get_txts(txt_path):
    '''
    find gt files in data path
    :return: list of files found
    '''
    txt_path = os.path.abspath(txt_path)
    files = []
    exts = ['txt']
    for parent, dirnames, filenames in os.walk(txt_path):
        for filename in filenames:
            for ext in exts:
                if filename.endswith(ext):
                    files.append(os.path.join(parent, filename))
                    break
    print('Find {} txts'.format(len(files)))
    return sorted(files)    

if __name__ == '__main__':
    #img_path = './data/ch4_training_images'
    img_path = './data/ch4_test_images'
    files = get_images(img_path)
    #txt_path = './data/ch4_training_localization_transcription_gt'
    txt_path = './data/Challenge4_Test_Task4_GT'
    txts = get_txts(txt_path)
    n = len(files)
    assert len(files) == len(txts)
    with open('dataset_ic13_test.txt', 'w') as f:
        for i in range(n):
            line = files[i] + '\t' + txts[i] + '\n'
            f.write(line)
    
    print('dataset generated ^_^ ')

config.json:

{
    "name": "DB",
    "data_loader": {
        "type": "ImageDataset",
        "args": {
            "dataset": {
                "train_data_path": [
                    [
                        "./dataset_ic13and15_train.txt"
                    ]
                ],
                "train_data_ratio": [
                    1.0
                ],
                "val_data_path": "./dataset_ic13_test.txt",
                "input_size": 640,
                "img_channel": 3,
                "shrink_ratio": 0.4
            },
            "loader": {
                "validation_split": 0.1,
                "train_batch_size": 10,

                "shuffle": false,
                "pin_memory": false,
                "num_workers": 0
            }
        }
    },
    "arch": {
        "type": "DBModel",
        "args": {
            "backbone": "shufflenetv2",
            "fpem_repeat": 2,
            "pretrained": true,
            "segmentation_head": "FPN"
        }
    },
    "loss": {
        "type": "DBLoss",
        "args": {
            "alpha": 1.0,
            "beta": 10.0,
            "ohem_ratio": 3
        }
    },
    "optimizer": {
        "type": "Adam",
        "args": {
            "lr": 0.001,
            "weight_decay": 0,
            "amsgrad": true
        }
    },
    "lr_scheduler": {
        "type": "StepLR",
        "args": {
            "step_size": 300,
            "gamma": 0.1
        }
    },
    "trainer": {
        "seed": 2,
        "gpus": [
            5
        ],
        "epochs": 1000,
        "display_interval": 10,
        "show_images_interval": 100,
        "resume_checkpoint": "",
        "finetune_checkpoint": "",
        "output_dir": "output",
        "tensorboard": false,
        "metrics": "loss"
    }
}