改进2-合并BN层到卷积层代码修改

改进2：合并BN层到卷积层

目的：提升模型前向推理的速度

include/darknet.h

#ifndef ENABLE_FUSE_CONV_BATCHNORM
#define ENABLE_FUSE_CONV_BATCHNORM
#endif

#ifdef ENABLE_FUSE_CONV_BATCHNORM
void fuse_conv_batchnorm(network net);
#endif

src/network.c

#ifdef ENABLE_FUSE_CONV_BATCHNORM
void fuse_conv_batchnorm(network net)
{
    int j;
    for (j = 0; j < net.n; ++j) {
        layer *l = &net.layers[j];

        if (l->type == CONVOLUTIONAL) {
            //printf(" Merges Convolutional-%d and batch_norm \n", j);

            if (l->batch_normalize) {
                int f;
                for (f = 0; f < l->n; ++f)
                {
                    l->biases[f] = l->biases[f] - (double)l->scales[f] * l->rolling_mean[f] / (sqrt((double)l->rolling_variance[f] + .000001));

                    const size_t filter_size = l->size*l->size*l->c / l->groups;
                    int i;
                    for (i = 0; i < filter_size; ++i) {
                        int w_index = f*filter_size + i;

                        l->weights[w_index] = (double)l->weights[w_index] * l->scales[f] / (sqrt((double)l->rolling_variance[f] + .000001));
                    }
                }

                l->batch_normalize = 0;
#ifdef GPU
                if (gpu_index >= 0) {
                    push_convolutional_layer(*l);
                }
#endif
            }
        }
        else {
            //printf(" Fusion skip layer type: %d \n", l->type);
        }
    }
}
#endif //ENABLE_FUSE_CONV_BATCHNORM

src/demo.c

void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const char *filename, char **names, int classes, int delay, char *prefix, int avg_frames, float hier, int w, int h, int frames, int fullscreen)
{
    //demo_frame = avg_frames;
    image **alphabet = load_alphabet();
    demo_names = names;
    demo_alphabet = alphabet;
    demo_classes = classes;
    demo_thresh = thresh;
    demo_hier = hier;
    printf("Demo\n");
    net = load_network(cfgfile, weightfile, 0);
    set_batch_network(net, 1);
    #ifdef ENABLE_FUSE_CONV_BATCHNORM
    fuse_conv_batchnorm(*net);
    #endif
    ......
}

src/detector.c

void validate_detector(char *datacfg, char *cfgfile, char *weightfile, char *outfile)
{
    int j;
    list *options = read_data_cfg(datacfg);
    char *valid_images = option_find_str(options, "valid", "data/train.list");
    char *name_list = option_find_str(options, "names", "data/names.list");
    char *prefix = option_find_str(options, "results", "results");
    char **names = get_labels(name_list);
    char *mapf = option_find_str(options, "map", 0);
    int *map = 0;
    if (mapf) map = read_map(mapf);

    network *net = load_network(cfgfile, weightfile, 0);
    set_batch_network(net, 1);
    #ifdef ENABLE_FUSE_CONV_BATCHNORM
    fuse_conv_batchnorm(*net);
    #endif
    ......
    }

void validate_detector_recall(char *cfgfile, char *weightfile)
{
    network *net = load_network(cfgfile, weightfile, 0);
    set_batch_network(net, 1);
    #ifdef ENABLE_FUSE_CONV_BATCHNORM
    fuse_conv_batchnorm(*net);
    #endif
    ......
}

void test_detector(char *datacfg, char *cfgfile, char *weightfile, char *filename, float thresh, float hier_thresh, char *outfile, int fullscreen)
{
    list *options = read_data_cfg(datacfg);
    char *name_list = option_find_str(options, "names", "data/names.list");
    char **names = get_labels(name_list);

    image **alphabet = load_alphabet();
    network *net = load_network(cfgfile, weightfile, 0);
    set_batch_network(net, 1);
    #ifdef ENABLE_FUSE_CONV_BATCHNORM
    fuse_conv_batchnorm(*net);
    #endif
    ......
    }

训练及性能测试：

./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74

./darknet detector valid cfg/voc.data cfg/yolov3-voc-test.cfg backup/yolov3-voc_final.weights  

python reval_voc.py  --voc_dir /home/bai/darknet/VOCdevkit --year 2007 --image_set test --classes  /home/bai/darknet/data/voc.names testVOC