理解darknet如何训练目标检测模型
定位源码
参考 darknet训练自己的yolo_v3模型 这篇博客，训练用这行命令：
./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74darknet.c 中对应源码：
else if (0 == strcmp(argv[1], "detector")){
run_detector(argc, argv);run_detector() 中对应源码：
else if(0==strcmp(argv[2], "train")) train_detector(datacfg, cfg, weights, gpus, ngpus, clear);解读源码
默认使用含GPU预定义的代码。
void train_detector(char *datacfg, char *cfgfile, char *weightfile, int *gpus, int ngpus, int clear)
{
list *options = read_data_cfg(datacfg);
char *train_images = option_find_str(options, "train", "data/train.list");
char *backup_directory = option_find_str(options, "backup", "/backup/");
读取datacfg数据集配置文件，寻找训练集索引文件名，赋值给 train_images 字符串；寻找备份文件夹名，赋值给 backup_directory 字符串。
以 cfg/voc.data 为例，其内容如下：
指定训练集索引文件为 /home/pjreddie/data/voc/train.txt ，备份文件夹为 bachup 。
srand(time(0));
char *base = basecfg(cfgfile);
printf("%s\n", base);
float avg_loss = -1;
network **nets = calloc(ngpus, sizeof(network));time(0)代表当前时间，srand(time(0))代表用当前时间来设定rand函数所用的随机数产生演算法的种子值。
读取网络配置文件内容到 base 字符串。申请ngpus（gpu个数）个network结构体大小的连续内存，返回首地址指针给nets。
srand(time(0));
int seed = rand();
int i;
for(i = 0; i < ngpus; ++i){
srand(seed);
#ifdef GPU
cuda_set_device(gpus[i]);
#endif
nets[i] = load_network(cfgfile, weightfile, clear);
nets[i]->learning_rate *= ngpus;
}
srand(time(0));
network *net = nets[0];利用cfggile(网络配置)和weightfile(网络权重)载入网络。如果用到多个GPU，则多次载入网络，且学习率要乘上GPU个数，我尚不知道为什么这么做。
int imgs = net->batch * net->subdivisions * ngpus;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net->learning_rate, net->momentum, net->decay);
data train, buffer;
layer l = net->layers[net->n - 1];
int classes = l.classes;
float jitter = l.jitter;
list *plist = get_paths(train_images);
//int N = plist->size;
char **paths = (char **)list_to_array(plist);
load_args args = get_base_args(net);
args.coords = l.coords;
args.paths = paths;
args.n = imgs;
args.m = plist->size;
args.classes = classes;
args.jitter = jitter;
args.num_boxes = l.max_boxes;
args.d = &buffer;
args.type = DETECTION_DATA;
//args.type = INSTANCE_DATA;
args.threads = 64;主要是载入目标检测网络的一些参数。参数很多，牵扯到一些目标检测的操作，都可以单独写一篇博客，这里暂时跳过。
pthread_t load_thread = load_data(args);
double time;
int count = 0;
//while(i*imgs < N*120){
while(get_current_batch(net) < net->max_batches){
if(l.random && count++%10 == 0){
printf("Resizing\n");
int dim = (rand() % 10 + 10) * 32;
if (get_current_batch(net)+200 > net->max_batches) dim = 608;
//int dim = (rand() % 4 + 16) * 32;
printf("%d\n", dim);
args.w = dim;
args.h = dim;
pthread_join(load_thread, 0);
train = buffer;
free_data(train);
load_thread = load_data(args);
#pragma omp parallel for
for(i = 0; i < ngpus; ++i){
resize_network(nets[i], dim, dim);
}
net = nets[0];
}
time=what_time_is_it_now();
pthread_join(load_thread, 0);
train = buffer;
load_thread = load_data(args);
printf("Loaded: %lf seconds\n", what_time_is_it_now()-time);
time=what_time_is_it_now();
float loss = 0;
#ifdef GPU
if(ngpus == 1){
loss = train_network(net, train);
} else {
loss = train_networks(nets, ngpus, train, 4);
}
#else
loss = train_network(net, train);
#endif
if (avg_loss < 0) avg_loss = loss;
avg_loss = avg_loss*.9 + loss*.1;
i = get_current_batch(net);
printf("%ld: %f, %f avg, %f rate, %lf seconds, %d images\n", get_current_batch(net), loss, avg_loss, get_current_rate(net), what_time_is_it_now()-time, i*imgs);
if(i%100==0){
#ifdef GPU
if(ngpus != 1) sync_nets(nets, ngpus, 0);
#endif
char buff[256];
sprintf(buff, "%s/%s.backup", backup_directory, base);
save_weights(net, buff);
}
if(i%10000==0 || (i < 1000 && i%100 == 0)){
#ifdef GPU
if(ngpus != 1) sync_nets(nets, ngpus, 0);
#endif
char buff[256];
sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
save_weights(net, buff);
}
free_data(train);
}这是训练的主体。这里用到了多线程读取多张图片，然后结合成一个batch，赋给data结构体变量 “tain” 中。
训练的迭代次数是训练的批次数。
训练过程中，每训练了100个批次，把网络权重保存到后缀名为backup的文件中，当做备份。
训练批次在1000以内中的每100批次，或者训练批次为100000时，保存权重到后缀名为weights的文件。
当训练批次到达 net->max_batches 时，会停止训练，并把训练得到的权重保存下来。
训练产生的权重文件保存在backup目录下。