Yolov3训练自己的数据集

安装

（1）github地址

源码

git checkout -b 64efa721ede91cd8ccc18257f98eeba43b73a6afa

https://github.com/AlexeyAB/darknet.git

（2）编译

环境cuda10.0，cudnn7.5.1
修改GPU=1，CUDNN=1，OPENCV=1，修改这三个库的头文件及lib库文件路径

GPU=1
CUDNN=1
CUDNN_HALF=0
OPENCV=1
AVX=0
OPENMP=0
LIBSO=1
ZED_CAMERA=0 # ZED SDK 3.0 and above
ZED_CAMERA_v2_8=0 # ZED SDK 2.X

# set GPU=1 and CUDNN=1 to speedup on GPU
# set CUDNN_HALF=1 to further speedup 3 x times (Mixed-precision on Tensor Cores) GPU: Volta, Xavier, Turing and higher
# set AVX=1 and OPENMP=1 to speedup on CPU (if error occurs then set AVX=0)

USE_CPP=1
DEBUG=1

ARCH= -gencode arch=compute_30,code=sm_30 \
      -gencode arch=compute_35,code=sm_35 \
      -gencode arch=compute_50,code=[sm_50,compute_50] \
      -gencode arch=compute_52,code=[sm_52,compute_52] \
	    -gencode arch=compute_61,code=[sm_61,compute_61]

OS := $(shell uname)

# Tesla V100
# ARCH= -gencode arch=compute_70,code=[sm_70,compute_70]

# GeForce RTX 2080 Ti, RTX 2080, RTX 2070, Quadro RTX 8000, Quadro RTX 6000, Quadro RTX 5000, Tesla T4, XNOR Tensor Cores
# ARCH= -gencode arch=compute_75,code=[sm_75,compute_75]

# Jetson XAVIER
# ARCH= -gencode arch=compute_72,code=[sm_72,compute_72]

# GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4
# ARCH= -gencode arch=compute_61,code=sm_61 -gencode arch=compute_61,code=compute_61

# GP100/Tesla P100 - DGX-1
# ARCH= -gencode arch=compute_60,code=sm_60

# For Jetson TX1, Tegra X1, DRIVE CX, DRIVE PX - uncomment:
# ARCH= -gencode arch=compute_53,code=[sm_53,compute_53]

# For Jetson Tx2 or Drive-PX2 uncomment:
# ARCH= -gencode arch=compute_62,code=[sm_62,compute_62]


VPATH=./src/
EXEC=darknet
OBJDIR=./obj/

ifeq ($(LIBSO), 1)
LIBNAMESO=libdarknet.so
APPNAMESO=uselib
endif

ifeq ($(USE_CPP), 1)
CC=g++
else
CC=gcc
endif

CPP=g++ -std=c++11
NVCC=nvcc
OPTS=-Ofast
LDFLAGS= -lm -pthread
COMMON= -Iinclude/ -I3rdparty/stb/include
CFLAGS=-Wall -Wfatal-errors -Wno-unused-result -Wno-unknown-pragmas -fPIC

ifeq ($(DEBUG), 1)
#OPTS= -O0 -g
#OPTS= -Og -g
COMMON+= -DDEBUG
CFLAGS+= -DDEBUG
else
ifeq ($(AVX), 1)
CFLAGS+= -ffp-contract=fast -mavx -mavx2 -msse3 -msse4.1 -msse4.2 -msse4a
endif
endif

CFLAGS+=$(OPTS)

ifneq (,$(findstring MSYS_NT,$(OS)))
LDFLAGS+=-lws2_32
endif

ifeq ($(OPENCV), 1)
COMMON+= -DOPENCV
CFLAGS+= -DOPENCV
#LDFLAGS+= `pkg-config --libs opencv4 2> /dev/null || pkg-config --libs opencv`
#COMMON+= `pkg-config --cflags opencv4 2> /dev/null || pkg-config --cflags opencv`
LDFLAGS+=-I/home/nfs/admin0/apps/python/anaconda3/envs/py27/include -L/home/nfs/admin0/apps/python/anaconda3/envs/py27/lib/ -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_highgui -lopencv_video -lopencv_videoio
COMMON+=-I/home/nfs/admin0/apps/python/anaconda3/envs/py27/include -L/home/nfs/admin0/apps/python/anaconda3/envs/py27/lib/ -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_highgui -lopencv_video -lopencv_videoio
endif

ifeq ($(OPENMP), 1)
CFLAGS+= -fopenmp
LDFLAGS+= -lgomp
endif

ifeq ($(GPU), 1)
COMMON+= -DGPU -I/usr/local/cuda/include/
CFLAGS+= -DGPU
ifeq ($(OS),Darwin) #MAC
LDFLAGS+= -L/usr/local/cuda/lib -lcuda -lcudart -lcublas -lcurand
else
LDFLAGS+= -L/usr/local/cuda/lib64 -lcuda -lcudart -lcublas -lcurand
endif
endif

ifeq ($(CUDNN), 1)
COMMON+= -DCUDNN
ifeq ($(OS),Darwin) #MAC
CFLAGS+= -DCUDNN -I/usr/local/cuda/include
LDFLAGS+= -L/usr/local/cuda/lib -lcudnn
else
CFLAGS+= -DCUDNN -I/usr/local/cudnn/include
LDFLAGS+= -L/usr/local/cudnn/lib64 -lcudnn
endif
endif

ifeq ($(CUDNN_HALF), 1)
COMMON+= -DCUDNN_HALF
CFLAGS+= -DCUDNN_HALF
ARCH+= -gencode arch=compute_70,code=[sm_70,compute_70]
endif

ifeq ($(ZED_CAMERA), 1)
CFLAGS+= -DZED_STEREO -I/usr/local/zed/include
ifeq ($(ZED_CAMERA_v2_8), 1)
LDFLAGS+= -L/usr/local/zed/lib -lsl_core -lsl_input -lsl_zed
#-lstdc++ -D_GLIBCXX_USE_CXX11_ABI=0
else
LDFLAGS+= -L/usr/local/zed/lib -lsl_zed
#-lstdc++ -D_GLIBCXX_USE_CXX11_ABI=0
endif
endif

OBJ=image_opencv.o http_stream.o gemm.o utils.o dark_cuda.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o darknet.o detection_layer.o captcha.o route_layer.o writing.o box.o nightmare.o normalization_layer.o avgpool_layer.o coco.o dice.o yolo.o detector.o layer.o compare.o classifier.o local_layer.o swag.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o rnn.o rnn_vid.o crnn_layer.o demo.o tag.o cifar.o go.o batchnorm_layer.o art.o region_layer.o reorg_layer.o reorg_old_layer.o super.o voxel.o tree.o yolo_layer.o gaussian_yolo_layer.o upsample_layer.o lstm_layer.o conv_lstm_layer.o scale_channels_layer.o sam_layer.o
ifeq ($(GPU), 1)
LDFLAGS+= -lstdc++
OBJ+=convolutional_kernels.o activation_kernels.o im2col_kernels.o col2im_kernels.o blas_kernels.o crop_layer_kernels.o dropout_layer_kernels.o maxpool_layer_kernels.o network_kernels.o avgpool_layer_kernels.o
endif

OBJS = $(addprefix $(OBJDIR), $(OBJ))
DEPS = $(wildcard src/*.h) Makefile include/darknet.h

all: $(OBJDIR) backup results setchmod $(EXEC) $(LIBNAMESO) $(APPNAMESO)

ifeq ($(LIBSO), 1)
CFLAGS+= -fPIC

$(LIBNAMESO): $(OBJDIR) $(OBJS) include/yolo_v2_class.hpp src/yolo_v2_class.cpp
	$(CPP) -shared -std=c++11 -fvisibility=hidden -DLIB_EXPORTS $(COMMON) $(CFLAGS) $(OBJS) src/yolo_v2_class.cpp -o $@ $(LDFLAGS)

$(APPNAMESO): $(LIBNAMESO) include/yolo_v2_class.hpp src/yolo_console_dll.cpp
	$(CPP) -std=c++11 $(COMMON) $(CFLAGS) -o $@ src/yolo_console_dll.cpp $(LDFLAGS) -L ./ -l:$(LIBNAMESO)
endif

$(EXEC): $(OBJS)
	$(CPP) -std=c++11 $(COMMON) $(CFLAGS) $^ -o $@ $(LDFLAGS)

$(OBJDIR)%.o: %.c $(DEPS)
	$(CC) $(COMMON) $(CFLAGS) -c $< -o $@

$(OBJDIR)%.o: %.cpp $(DEPS)
	$(CPP) -std=c++11 $(COMMON) $(CFLAGS) -c $< -o $@

$(OBJDIR)%.o: %.cu $(DEPS)
	$(NVCC) $(ARCH) $(COMMON) --compiler-options "$(CFLAGS)" -c $< -o $@

$(OBJDIR):
	mkdir -p $(OBJDIR)
backup:
	mkdir -p backup
results:
	mkdir -p results
setchmod:
	chmod +x *.sh

.PHONY: clean

clean:
	rm -rf $(OBJS) $(EXEC) $(LIBNAMESO) $(APPNAMESO)

准备自己数据集

（1）train、test文件夹下分别存放JPEFImages、labels、train.txt(test.txt),train.txt(test.txt)存放图片路径，JPEGImages存放图片，labels存放图片标签；

（2）标签内容内 “类别 (box_center_point_x/box_w) (box_center_point_y/box_h) (box_w/img_w) (box_h/img_h)” ，上述值均为归一化之后的值。一张图片中有多少个标注目标就有多少行以上内容。

def change_txt(ori_txt,dst_txt,ori_img):
    for txt in os.listdir(ori_txt):
        label_infos = open(os.path.join(ori_txt, txt), "r")
        txtlines = label_infos.read().splitlines()

        txt_list = txt.split('.')
        img_name = '.'.join(txt_list[:-1]) + '.jpg'
        srcimg = os.path.join(ori_img,img_name)
        img = cv2.imread(srcimg)
        size = img.shape

        dst_label_info = open(os.path.join(dst_txt,txt),"w")
        for line in txtlines:
            if len(line) == 0:
                dst_label_info.close()
                continue
            arr = line.split(" ")

            center_x = (float(arr[1]) + float(arr[3]))/2.0
            center_y = (float(arr[2]) + float(arr[4]))/2.0
            target_w = float(arr[3]) - float(arr[1])
            target_h = float(arr[4]) - float(arr[2])

            img_w = size[1]
            img_h = size[0]

            cvt_center_x = center_x * 1.0/ img_w
            cvt_center_y = center_y * 1.0/ img_h
            cvt_target_w = target_w * 1.0/ img_w
            cvt_target_h = target_h * 1.0/ img_h

            new_line_list = []
            new_line_list.append('0')
            new_line_list.append(cvt_center_x)
            new_line_list.append(cvt_center_y)
            new_line_list.append(cvt_target_w)
            new_line_list.append(cvt_target_h)

            if(len(new_line_list) != 0):
                print(len(new_line_list))
                str_label = ''
                for i in range(0, (len(new_line_list) - 1)):
                    str_label += str(new_line_list[i]) + ' '
                    # dst_label_info.write(new_line_list[i] + ' ')
                str_label += str(new_line_list[-1]) + '\n'
                # dst_label_info.write(new_line_list[-1] + '\n')
                dst_label_info.write(str_label)
        dst_label_info.close()

（3）求prior box

# coding=utf-8
# k-means ++ for YOLOv2 anchors
# 通过k-means ++ 算法获取YOLOv2需要的anchors的尺寸
import numpy as np

# 定义Box类，描述bounding box的坐标
class Box():
    def __init__(self, x, y, w, h):
        self.x = x
        self.y = y
        self.w = w
        self.h = h


# 计算两个box在某个轴上的重叠部分
# x1是box1的中心在该轴上的坐标
# len1是box1在该轴上的长度
# x2是box2的中心在该轴上的坐标
# len2是box2在该轴上的长度
# 返回值是该轴上重叠的长度
def overlap(x1, len1, x2, len2):
    len1_half = len1 / 2
    len2_half = len2 / 2

    left = max(x1 - len1_half, x2 - len2_half)
    right = min(x1 + len1_half, x2 + len2_half)

    return right - left


# 计算box a 和box b 的交集面积
# a和b都是Box类型实例
# 返回值area是box a 和box b 的交集面积
def box_intersection(a, b):
    w = overlap(a.x, a.w, b.x, b.w)
    h = overlap(a.y, a.h, b.y, b.h)
    if w < 0 or h < 0:
        return 0

    area = w * h
    return area


# 计算 box a 和 box b 的并集面积
# a和b都是Box类型实例
# 返回值u是box a 和box b 的并集面积
def box_union(a, b):
    i = box_intersection(a, b)
    u = a.w * a.h + b.w * b.h - i
    return u


# 计算 box a 和 box b 的 iou
# a和b都是Box类型实例
# 返回值是box a 和box b 的iou
def box_iou(a, b):
    return box_intersection(a, b) / box_union(a, b)


# 使用k-means ++ 初始化 centroids，减少随机初始化的centroids对最终结果的影响
# boxes是所有bounding boxes的Box对象列表
# n_anchors是k-means的k值
# 返回值centroids 是初始化的n_anchors个centroid
def init_centroids(boxes,n_anchors):
    centroids = []
    boxes_num = len(boxes)

    centroid_index = np.random.choice(boxes_num, 1)
    centroids.append(boxes[centroid_index])

    print(centroids[0].w,centroids[0].h)

    for centroid_index in range(0,n_anchors-1):

        sum_distance = 0
        distance_thresh = 0
        distance_list = []
        cur_sum = 0

        for box in boxes:
            min_distance = 1
            for centroid_i, centroid in enumerate(centroids):
                distance = (1 - box_iou(box, centroid))
                if distance < min_distance:
                    min_distance = distance
            sum_distance += min_distance
            distance_list.append(min_distance)

        distance_thresh = sum_distance*np.random.random()

        for i in range(0,boxes_num):
            cur_sum += distance_list[i]
            if cur_sum > distance_thresh:
                centroids.append(boxes[i])
                print(boxes[i].w, boxes[i].h)
                break

    return centroids


# 进行 k-means 计算新的centroids
# boxes是所有bounding boxes的Box对象列表
# n_anchors是k-means的k值
# centroids是所有簇的中心
# 返回值new_centroids 是计算出的新簇中心
# 返回值groups是n_anchors个簇包含的boxes的列表
# 返回值loss是所有box距离所属的最近的centroid的距离的和
def do_kmeans(n_anchors, boxes, centroids):
    loss = 0
    groups = []
    new_centroids = []
    for i in range(n_anchors):
        groups.append([])
        new_centroids.append(Box(0, 0, 0, 0))

    for box in boxes:
        min_distance = 1
        group_index = 0
        for centroid_index, centroid in enumerate(centroids):
            distance = (1 - box_iou(box, centroid))
            if distance < min_distance:
                min_distance = distance
                group_index = centroid_index
        groups[group_index].append(box)
        loss += min_distance
        new_centroids[group_index].w += box.w
        new_centroids[group_index].h += box.h

    for i in range(n_anchors):
        new_centroids[i].w /= len(groups[i])
        new_centroids[i].h /= len(groups[i])

    return new_centroids, groups, loss


# 计算给定bounding boxes的n_anchors数量的centroids
# label_path是训练集列表文件地址
# n_anchors 是anchors的数量
# loss_convergence是允许的loss的最小变化值
# grid_size * grid_size 是栅格数量
# iterations_num是最大迭代次数
# plus = 1时启用k means ++ 初始化centroids
def compute_centroids(label_path,n_anchors,loss_convergence,grid_size,iterations_num,plus):

    boxes = []
    label_files = []
    f = open(label_path)
    for line in f:
        label_path = line.rstrip().replace('img', 'txt_yolo')
        label_path = label_path.replace('JPEGImages', 'labels')
        label_path = label_path.replace('.jpg', '.txt')
        label_path = label_path.replace('.JPEG', '.txt')
        label_files.append(label_path)
    f.close()

    for label_file in label_files:
        f = open(label_file)
        for line in f:
            temp = line.strip().split(" ")
            if len(temp) > 1:
                boxes.append(Box(0, 0, float(temp[3]), float(temp[4])))

    if plus:
        centroids = init_centroids(boxes, n_anchors)
    else:
        centroid_indices = np.random.choice(len(boxes), n_anchors)
        centroids = []
        for centroid_index in centroid_indices:
            centroids.append(boxes[centroid_index])

    # iterate k-means
    centroids, groups, old_loss = do_kmeans(n_anchors, boxes, centroids)
    iterations = 1
    while (True):
        centroids, groups, loss = do_kmeans(n_anchors, boxes, centroids)
        iterations = iterations + 1
        print("loss = %f" % loss)
        if abs(old_loss - loss) < loss_convergence or iterations > iterations_num:
            break
        old_loss = loss

        for centroid in centroids:
            print(centroid.w * grid_size, centroid.h * grid_size)

    # print result
    for centroid in centroids:
        print("k-means result：\n")
        print(centroid.w * grid_size, centroid.h * grid_size)


label_path = "/data_2/LYD/红绿灯检测/原始数据/ssd数据制作脚本/data_rename/train1.txt"
n_anchors = 9
loss_convergence = 1e-5
grid_size = 288
iterations_num = 200
plus = 0
compute_centroids(label_path,n_anchors,loss_convergence,grid_size,iterations_num,plus)

训练

(1)修改cfg/voc.data内容

classes= 1
train  = /home/LightDetection_Yolov3/data_yolo/train/train.txt
valid  = /home/LightDetection_Yolov3/data_yolo/test/test.txt
names = data/voc.names
backup = /home/LightDetection_Yolov3/darknet/backup/

(2)修改data/voc.names

light

(3)修改cfg/yolov3.cfg

//第3行，训练时batch=32
batch=32
//第8、9行，width、height=288
width=288
height=288
//第22、23行，修改学习率下降策略
steps=100000,200000,350000，450000
scales=.5，.2，.5，.2
//第603、689、776行，filter改为18，只有一类，（80+5）x3替换为（1+5）x3 = 18
filters=18
//第609、695、782行，修改anchors为准备数据集中第(3)步求出来的prior box
anchors = 6,13， 9,21， 17,8， 14,31， 19,44， 29,13， 26,63， 49,21， 57,94

(4)开始训练

./darknet detector train cfg/voc.data cfg/yolov3.cfg