本文共 3828 字，大约阅读时间需要 12 分钟。

在darknet版的yoloV4中。data文件中train.txt中只提供了图像的路径，那标签文件是怎么对应读取的呢？

在源码中，找到下面这个函数可以看到：

void replace_image_to_label(const char* input_path, char* output_path){
       find_replace(input_path, "/images/train2014/", "/labels/train2014/", output_path);    // COCO    find_replace(output_path, "/images/val2014/", "/labels/val2014/", output_path);        // COCO    find_replace(output_path, "/JPEGImages/", "/labels/", output_path);    // PascalVOC    find_replace(output_path, "\\images\\train2014\\", "\\labels\\train2014\\", output_path);    // COCO    find_replace(output_path, "\\images\\val2014\\", "\\labels\\val2014\\", output_path);        // COCO    find_replace(output_path, "\\JPEGImages\\", "\\labels\\", output_path);    // PascalVOC    //find_replace(output_path, "/images/", "/labels/", output_path);    // COCO    //find_replace(output_path, "/VOC2007/JPEGImages/", "/VOC2007/labels/", output_path);        // PascalVOC    //find_replace(output_path, "/VOC2012/JPEGImages/", "/VOC2012/labels/", output_path);        // PascalVOC    //find_replace(output_path, "/raw/", "/labels/", output_path);    trim(output_path);    // replace only ext of files    find_replace_extension(output_path, ".jpg", ".txt", output_path);    find_replace_extension(output_path, ".JPG", ".txt", output_path); // error    find_replace_extension(output_path, ".jpeg", ".txt", output_path);    find_replace_extension(output_path, ".JPEG", ".txt", output_path);    find_replace_extension(output_path, ".png", ".txt", output_path);    find_replace_extension(output_path, ".PNG", ".txt", output_path);    find_replace_extension(output_path, ".bmp", ".txt", output_path);    find_replace_extension(output_path, ".BMP", ".txt", output_path);    find_replace_extension(output_path, ".ppm", ".txt", output_path);    find_replace_extension(output_path, ".PPM", ".txt", output_path);    find_replace_extension(output_path, ".tiff", ".txt", output_path);    find_replace_extension(output_path, ".TIFF", ".txt", output_path);    // Check file ends with txt:    if(strlen(output_path) > 4) {
           char *output_path_ext = output_path + strlen(output_path) - 4;        if( strcmp(".txt", output_path_ext) != 0){
               fprintf(stderr, "Failed to infer label file name (check image extension is supported): %s \n", output_path);        }    }else{
           fprintf(stderr, "Label file name is too short: %s \n", output_path);    }}

以PascalVOC数据集为例，程序会自动把图像路径中的/JPEGImages/ 替换为 /labels/

所以我们在任一路径下可以新建两个文件夹

-JPEGImages   #放入所有的训练图片文件-labels   #放入所有的txt文件，会自动生成此文件夹

执行下面的Python程序

import os# iter floderdef iter_files(rootDir, extend_name):    file_list = []    fileName = []    #Iterate over the roots    for root,dirs,files in os.walk(rootDir):        for file_name in files:            root = os.path.normpath(root) #格式化路径 斜杠问题  / -> \            file_path = os.path.join(root, file_name)  # 连接路径            file_path = file_path.replace("\\", "/")            flag = file_path.endswith(extend_name)            if flag:                file_list.append(file_path)                fileName.append(os.path.splitext(file_name)[0])    return file_list, fileName# split train and valdef  split_train_val(fileList):    trainFile = open('train.txt', 'a')    validFile = open('val.txt', 'a')    num = 0    for files in fileList:        num = num + 1        if num % 4 == 0:            validFile.write(files + '\n')        else:            trainFile.write(files + '\n')    trainFile.close()    validFile.close()if __name__ == '__main__':    root = 'D:/Python/images/JEPGImages'  #训练图像的路径    [fileList, fileName] = iter_files(root, 'png')    split_train_val(fileList)

这样我们就直接得到了训练集和验证集的txt文件，前提是你直接有txt格式的标注文件，不是xml的。不然还是需要之前的那篇教程。

后面就简简单单改几个地方，就能训练起来了！

转载地址：http://rxwni.baihongyu.com/