Data reading in yolov3 (1)

yolov3The training of the code from detector.c de train_detector()函数开始,函数如下：

void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear, int dont_show)

If training code for：

*datacfg代表的是 voc.data ,*cfgfile代表的是 yolov3-voc.cfg
voc.data格式：

下面解读train_detector()下的函数.

list *options = read_data_cfg(datacfg);

这个函数的作用是解析voc.data中的数据,And the data deposited in the end nodes as options的双向链表中,
Inside the key function is：

int read_option(char *s, list *options)
{
    size_t i;
    size_t len = strlen(s);
    char *val = 0;
    for(i = 0; i < len; ++i){
        if(s[i] == '='){
            s[i] = '\0';
            val = s+i+1;
            break;
        }
    }
    if(i == len-1) return 0;
    char *key = s;
    option_insert(options, key, val);
    return 1;
}

It can be seen that program willvoc.data中的数据以“=”There are two types of for boundary character,左边为key 右边为val,然后插入到options链表中.在opton_insert()函数中可以看到,作者将key与val包装成一个node节点中的val,val的格式如下：

typedef struct{
    char *key;
    char *val;
    int used;
} kvp;

以上,It can be seen that the author willvoc.dataAccording to each behavior of data in anodeNodes to join aoptions链表中,每个nodeContains three important information,分别为key,val,used（初始为零,标签处理）,Three information into astruct.
下面函数：

char *train_images = option_find_str(options, "train", "data/train.list");

Training images address into the totrain_images中.同时将optionsHead pointer tokey为train的下一行,used设为1.至此,Put the pictures of the training address intotrain_images中.注意：这里train_imagesPointing to aWrite address of imagestxt文件的地址.

A network model of analytic：

char *base = basecfg(cfgfile);

The above function training code（4）cfg/After the data into the*base中.For example as：*base = yolov3-voc.cfg.

转入到parser.c文件中,解读
network parse_network_cfg_custom(char *filename, int batch)
*filename是yolov3-voc.cfg,里面的batch是0;
这里看list *read_cfg()函数：

list *read_cfg(char *filename)
{
    FILE *file = fopen(filename, "r");
    if(file == 0) file_error(filename);
    char *line;
    int nu = 0;
    list *sections = make_list();   //list链表 The neural network layer
    section *current = 0;           //section A neural network layer,类似于卷积层...
    while((line=fgetl(file)) != 0){
        ++ nu;
        strip(line);
        switch(line[0]){
            case '[':
                current = malloc(sizeof(section));
                list_insert(sections, current);
                current->options = make_list();
                current->type = line;
                break;
            case '\0':
            case '#':
            case ';':
                free(line);
                break;
            default:
                if(!read_option(line, current->options)){
                    fprintf(stderr, "Config file error line %d, could parse: %s\n", nu, line);
                    free(line);
                }
                break;
        }
    }
    fclose(file);
    return sections;
}

Watching this program before you put a few price comparison important structures：

typedef struct list{
    int size;
    node *front;
    node *back;
} list;

typedef struct{
    char *type;
    list *options;
}section;

typedef struct node{
    void *val;
    struct node *next;
    struct node *prev;
} node;

typedef struct{
    char *key;
    char *val;
    int used;
} kvp;

This a few structure constitute a list,Below is the program read.
这里将yolov3-voc.cfgThe network layer of the chain into a list,名称为sections,Every node in the linked list islist结构体.At the same time for each layer parameters analytical to asection结构体中,结构体名为current,Put out the following function here：

void list_insert(list *l, void *val)
{
    node *new = malloc(sizeof(node));
    new->val = val;
    new->next = 0;

    if(!l->back){
        l->front = new;
        new->prev = 0;
    }else{
        l->back->next = new;
        new->prev = l->back;
    }
    l->back = new;
    ++l->size;
}

From the function can see,作者将current包装成一个node->val,让后将nodeAs a node is inserted into thelist链表中,这里node->val（current）结构体是section,包含两个变量,tpye与options.
section结构体中包含一个list链表,名为options,Deep to bala bala can see,这个optionsList the main store in order toA network layerDifferent parameters in the,拿第一个[net]层为例.

type中的值是’net’,optionsThe node node中存储着’height=416’这一类数据,node中有key与val ,key = height,val=416.
转入到list *read_cfg(char *filename)Function of the following program segment：

case '[':
                current = malloc(sizeof(section));
                list_insert(sections, current);
                current->options = make_list();
                current->type = line;
                break;

In this section you can see,在识别出"["After this symbol,作者将current插入到sections链表中,转入到list_insert（）这个函数中：

void list_insert(list *l, void *val)
{
    node *new = malloc(sizeof(node));
    new->val = val;
    new->next = 0;

    if(!l->back){
        l->front = new;
        new->prev = 0;
    }else{
        l->back->next = new;
        new->prev = l->back;
    }
    l->back = new;
    ++l->size;
}

Here is a two-way chain table insert program,最后的 ++l->size;表明一个yolov3网络结构有size个网络层,And when the first network layer from1开始的.
将current（Is a representative of the network layer）Head address as node is inserted into thesections（代表的是整个yolov3网络）链表中后,The author in the firstcurrent->optionsThe first address as the head of the list to create a list：

current->options = make_list();

The list of information stored in the procedures section：

            default:
                if(!read_option(line, current->options)){  //read_optionStore the parameters of each network layer
                    fprintf(stderr, "Config file error line %d, could parse: %s\n", nu, line);
                    free(line);
                }
                break;

read_option()函数如下：

int read_option(char *s, list *options)
{
    size_t i;
    size_t len = strlen(s);
    char *val = 0;
    for(i = 0; i < len; ++i){
        if(s[i] == '='){
            s[i] = '\0';
            val = s+i+1;
            break;
        }
    }
    if(i == len-1) return 0;
    char *key = s;
    option_insert(options, key, val);
    return 1;
}

This function mainly will I said abovesection插入到options中.最后在list *read_cfg(char *filename)Many times in the loop,A network layer tocurrent插入到sections中,And the parameters of a networkcurrent->otpions的链表中.
最终：将各个node链成一个options（list结构）链表,Represents a network layer parameters,The network layer name（“net”）赋值给type,而options与type合成为一个current.然后将currentAs a node chain intosections（list）中,sections是整个yolov3网络结构.
至此,The entire network is loaded into the listsections中.