ICDAR2015数据集标注转为ICDAR2013格式

最新推荐文章于 2024-09-20 19:37:20 发布

老菜鸡mou

最新推荐文章于 2024-09-20 19:37:20 发布

阅读量979

点赞数 3

文章标签：深度学习

本文链接：https://blog.csdn.net/qq_36104270/article/details/112325196

版权

icdar2015为八个角坐标格式，为了训练需要左上角和右下角坐标格式，而且去掉标注为###的框，由于我是训练检测，所以全部标注为text类

测试集标注

需要创建好Test_GT文件夹

for i in range(1,501):
    txts = 'Challenge4_Test_Task1_GT/gt_img_' + str(i) + '.txt'
    w_txts = 'Test_GT/gt_img_' + str(i) + '.txt'
    lines = open(txts,encoding='utf-8-sig').readlines()

    txt_file = open(w_txts,'w')
    for line in lines:
        numbers = line.split(',')
        x1 = min(numbers[0],numbers[2],numbers[4],numbers[6])
        x2 = max(numbers[0],numbers[2],numbers[4],numbers[6])
        y1 = min(numbers[1],numbers[3],numbers[5],numbers[7])
        y2 = max(numbers[1],numbers[3],numbers[5],numbers[7])

        if '###' not in numbers[8]:
            txt_file.write(x1 + ' ' + y1 + ' ' + x2 + ' ' + y2 + ' ' + 'text'  + '\n')

训练集标注

需要创建好Training_GT文件夹

for i in range(1,501):
    txts = 'ch4_training_localization_transcription_gt/gt_img_' + str(i) + '.txt'
    w_txts = 'Training_GT/gt_img_' + str(i) + '.txt'
    lines = open(txts,encoding='utf-8-sig').readlines()

    txt_file = open(w_txts,'w')
    for line in lines:
        numbers = line.split(',')
        x1 = min(numbers[0],numbers[2],numbers[4],numbers[6])
        x2 = max(numbers[0],numbers[2],numbers[4],numbers[6])
        y1 = min(numbers[1],numbers[3],numbers[5],numbers[7])
        y2 = max(numbers[1],numbers[3],numbers[5],numbers[7])

        if '###' not in numbers[8]:
            txt_file.write(x1 + ' ' + y1 + ' ' + x2 + ' ' + y2 + ' ' + 'text'  + '\n')