pytorch版yolov3訓練本身數據集

時間 2019-12-13

標籤 pytorch yolov3 yolov 訓練本身數據简体版

原文原文鏈接

目錄html

1. 環境搭建

將github庫download下來。

git clone https://github.com/ultralytics/yolov3.git

建議在linux環境下使用anaconda進行搭建

conda create -n yolov3 python=3.7

安裝須要的軟件

pip install -r requirements.txt

環境要求：node

python >= 3.7
pytorch >= 1.1
numpy
tqdm
opencv-python

其中只須要注意pytorch的安裝：python

到https://pytorch.org/中根據操做系統，python版本，cuda版本等選擇命令便可。linux

關於深度學習環境搭建請參看：http://www.javashuo.com/article/p-opzashmi-dy.htmlgit

anaconda經常使用用法：http://www.javashuo.com/article/p-vcfrpenk-ey.htmlgithub

2. 數據集構建

1. xml文件生成須要Labelimg軟件

在Windows下使用：

wget https://github.com/pprp/DeepLearning/blob/master/windows_v1.5.1/labelImg.exe

使用快捷鍵：

Ctrl + u  加載目錄中的全部圖像，鼠標點擊Open dir同功能
Ctrl + r  更改默認註釋目標目錄(xml文件保存的地址) 
Ctrl + s  保存
Ctrl + d  複製當前標籤和矩形框
space     將當前圖像標記爲已驗證
w         建立一個矩形框
d         下一張圖片
a         上一張圖片
del       刪除選定的矩形框
Ctrl++    放大
Ctrl--    縮小
↑→↓←        鍵盤箭頭移動選定的矩形框

2. VOC2007 數據集格式

-data
    - VOCdevkit2007
        - VOC2007
            - Annotations (標籤XML文件，用對應的圖片處理工具人工生成的)
            - ImageSets (生成的方法是用sh或者MATLAB語言生成)
                - Main
                    - test.txt
                    - train.txt
                    - trainval.txt
                    - val.txt
            - JPEGImages(原始文件)
            - labels (xml文件對應的txt文件)

經過以上軟件主要構造好JPEGImages和Annotations文件夾中內容,Main文件夾中的txt文件能夠經過python腳本生成：json

import os  
import random  
  
trainval_percent = 0.8
train_percent = 0.8  
xmlfilepath = 'Annotations'  
txtsavepath = 'ImageSets\Main'  
total_xml = os.listdir(xmlfilepath)  
  
num=len(total_xml)  
list=range(num)  
tv=int(num*trainval_percent)  
tr=int(tv*train_percent)  
trainval= random.sample(list,tv)  
train=random.sample(trainval,tr)  
  
ftrainval = open('ImageSets/Main/trainval.txt', 'w')  
ftest = open('ImageSets/Main/test.txt', 'w')  
ftrain = open('ImageSets/Main/train.txt', 'w')  
fval = open('ImageSets/Main/val.txt', 'w')  
  
for i  in list:  
    name=total_xml[i][:-4]+'\n'  
    if i in trainval:  
        ftrainval.write(name)  
        if i in train:  
            ftrain.write(name)  
        else:  
            fval.write(name)  
    else:  
        ftest.write(name)  
  
ftrainval.close()  
ftrain.close()  
fval.close()  
ftest.close()

生成labels文件，voc_label.py文件具體內容以下：windows

# -*- coding: utf-8 -*-
"""
Created on Tue Oct  2 11:42:13 2018
將本文件放到VOC2007目錄下，而後就能夠直接運行
須要修改的地方：
1. sets中替換爲本身的數據集
2. classes中替換爲本身的類別
3. 將本文件放到VOC2007目錄下
4. 直接開始運行
"""

import xml.etree.ElementTree as ET
import pickle
import os
from os import listdir, getcwd
from os.path import join
sets=[('2007', 'train'), ('2007', 'val'), ('2007', 'test')]  #替換爲本身的數據集
classes = ["head", "eye", "nose"]     #修改成本身的類別
#classes = ["eye", "nose"]

def convert(size, box):
    dw = 1./(size[0])
    dh = 1./(size[1])
    x = (box[0] + box[1])/2.0 - 1
    y = (box[2] + box[3])/2.0 - 1
    w = box[1] - box[0]
    h = box[3] - box[2]
    x = x*dw
    w = w*dw
    y = y*dh
    h = h*dh
    return (x,y,w,h)
def convert_annotation(year, image_id):
    in_file = open('VOC%s/Annotations/%s.xml'%(year, image_id))  #將數據集放於當前目錄下
    out_file = open('VOC%s/labels/%s.txt'%(year, image_id), 'w')
    tree=ET.parse(in_file)
    root = tree.getroot()
    size = root.find('size')
    w = int(size.find('width').text)
    h = int(size.find('height').text)
    for obj in root.iter('object'):
        difficult = obj.find('difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult)==1:
            continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text))
        bb = convert((w,h), b)
        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
wd = getcwd()
for year, image_set in sets:
    if not os.path.exists('VOC%s/labels/'%(year)):
        os.makedirs('VOC%s/labels/'%(year))
    image_ids = open('VOC%s/ImageSets/Main/%s.txt'%(year, image_set)).read().strip().split()
    list_file = open('%s_%s.txt'%(year, image_set), 'w')
    for image_id in image_ids:
        list_file.write('VOC%s/JPEGImages/%s.jpg\n'%(year, image_id))
        convert_annotation(year, image_id)
    list_file.close()   
#os.system("cat 2007_train.txt 2007_val.txt > train.txt")     #修改成本身的數據集用做訓練

到底爲止，VOC格式數據集構造完畢，可是還須要繼續構造符合darknet格式的數據集(coco)。api

須要說明的是：若是打算使用coco評價標準，須要構造coco中json格式，若是要求不高，只須要VOC格式便可，使用做者寫的mAP計算程序便可。bash

voc的xml轉coco的json文件腳本：xml2json.py

# -*- coding: utf-8 -*-
"""
Created on Tue Aug 28 15:01:03 2018
須要改動xml_path and json_path
"""
#!/usr/bin/python
# -*- coding:utf-8 -*-
# @Description: xml轉換到coco數據集json格式
 
import os, sys, json,xmltodict
 
from xml.etree.ElementTree import ElementTree, Element
from collections import OrderedDict
 
XML_PATH = "/home/learner/datasets/VOCdevkit2007/VOC2007/Annotations/test"
JSON_PATH = "./test.json"
json_obj = {}
images = []
annotations = []
categories = []
categories_list = []
annotation_id = 1
 
def read_xml(in_path):
    '''讀取並解析xml文件'''
    tree = ElementTree()
    tree.parse(in_path)
    return tree
 
def if_match(node, kv_map):
    '''判斷某個節點是否包含全部傳入參數屬性
      node: 節點
      kv_map: 屬性及屬性值組成的map'''
    for key in kv_map:
        if node.get(key) != kv_map.get(key):
            return False
    return True
 
def get_node_by_keyvalue(nodelist, kv_map):
    '''根據屬性及屬性值定位符合的節點，返回節點
      nodelist: 節點列表
      kv_map: 匹配屬性及屬性值map'''
    result_nodes = []
    for node in nodelist:
        if if_match(node, kv_map):
            result_nodes.append(node)
    return result_nodes
 
def find_nodes(tree, path):
    '''查找某個路徑匹配的全部節點
      tree: xml樹
      path: 節點路徑'''
    return tree.findall(path)
 
print ("-----------------Start------------------")
xml_names = []
for xml in os.listdir(XML_PATH):
    #os.path.splitext(xml)
    #xml=xml.replace('Cow_','')
    xml_names.append(xml)
    

'''xml_path_list=os.listdir(XML_PATH)
os.path.split
xml_path_list.sort(key=len)'''
xml_names.sort(key=lambda x:int(x[:-4]))
new_xml_names = []
for i in xml_names:
    j = 'Cow_' + i
    new_xml_names.append(j)

#print xml_names
#print new_xml_names
for xml in new_xml_names:
    tree = read_xml(XML_PATH + "/" + xml)
    object_nodes = get_node_by_keyvalue(find_nodes(tree, "object"), {})
    if len(object_nodes) == 0:
        print (xml, "no object")
        continue
    else:
        image = OrderedDict()
        file_name = os.path.splitext(xml)[0];  # 文件名
                #print os.path.splitext(xml)
        para1 = file_name + ".jpg"
               

                height_nodes = get_node_by_keyvalue(find_nodes(tree, "size/height"), {})
        para2 = int(height_nodes[0].text)
                
                width_nodes = get_node_by_keyvalue(find_nodes(tree, "size/width"), {})
        para3 = int(width_nodes[0].text)
        
        fname=file_name[4:]
        para4 = int(fname)
        
        
                
                for f,i in [("file_name",para1),("height",para2),("width",para3),("id",para4)]:
                    image.setdefault(f,i)

                #print(image)
                images.append(image)    #構建images
          
     
        name_nodes = get_node_by_keyvalue(find_nodes(tree, "object/name"), {})
        xmin_nodes = get_node_by_keyvalue(find_nodes(tree, "object/bndbox/xmin"), {})
        ymin_nodes = get_node_by_keyvalue(find_nodes(tree, "object/bndbox/ymin"), {})
        xmax_nodes = get_node_by_keyvalue(find_nodes(tree, "object/bndbox/xmax"), {})
        ymax_nodes = get_node_by_keyvalue(find_nodes(tree, "object/bndbox/ymax"), {})
           # print ymax_nodes
        for index, node in enumerate(object_nodes):
            annotation = {}
            segmentation = []
            bbox = []
            seg_coordinate = []     #座標
            seg_coordinate.append(int(xmin_nodes[index].text))
            seg_coordinate.append(int(ymin_nodes[index].text))
            seg_coordinate.append(int(xmin_nodes[index].text))
            seg_coordinate.append(int(ymax_nodes[index].text))
            seg_coordinate.append(int(xmax_nodes[index].text))
            seg_coordinate.append(int(ymax_nodes[index].text))
            seg_coordinate.append(int(xmax_nodes[index].text))
            seg_coordinate.append(int(ymin_nodes[index].text))
            segmentation.append(seg_coordinate)
            width = int(xmax_nodes[index].text) - int(xmin_nodes[index].text)
            height = int(ymax_nodes[index].text) - int(ymin_nodes[index].text)
            area = width * height
            bbox.append(int(xmin_nodes[index].text))
            bbox.append(int(ymin_nodes[index].text))
            bbox.append(width)
            bbox.append(height)
     
            annotation["segmentation"] = segmentation
            annotation["area"] = area
            annotation["iscrowd"] = 0
            fname=file_name[4:]
            annotation["image_id"] = int(fname)
            annotation["bbox"] = bbox
            cate=name_nodes[index].text
            if cate=='head':
                category_id=1
            elif cate=='eye':
                category_id=2
            elif cate=='nose':
                category_id=3
            annotation["category_id"] = category_id
            annotation["id"] = annotation_id
            annotation_id += 1
            annotation["ignore"] = 0
            annotations.append(annotation)
     
            if category_id in categories_list:
                pass
            else:
                categories_list.append(category_id)
                categorie = {}
                categorie["supercategory"] = "none"
                categorie["id"] = category_id
                categorie["name"] = name_nodes[index].text
                categories.append(categorie)
     
json_obj["images"] = images
json_obj["type"] = "instances"
json_obj["annotations"] = annotations
json_obj["categories"] = categories
 
f = open(JSON_PATH, "w")
#json.dump(json_obj, f)
json_str = json.dumps(json_obj)
f.write(json_str)
print ("------------------End-------------------")

（運行bash yolov3/data/get_coco_dataset.sh,仿照格式將數據放到其中）

可是這個庫還須要其餘模型：

3. 建立*.names file,

其中保存的是你的全部的類別，每行一個類別，如data/coco.names：

head
eye
nose

4. 更新data/coco.data,其中保存的是不少配置信息

classes = 3 # 改爲你的數據集的類別個數
train = ./data/2007_train.txt # 經過voc_label.py文件生成的txt文件
valid = ./data/2007_test.txt # 經過voc_label.py文件生成的txt文件
names = data/coco.names # 記錄類別
backup = backup/ # 記錄checkpoint存放位置
eval = coco # 選擇map計算方式

5. 更新cfg文件，修改類別相關信息

打開cfg文件夾下的yolov3.cfg文件，大致而言，cfg文件記錄的是整個網絡的結構，是核心部分，具體內容講解請見：https://pprp.github.io/2018/09/20/tricks.html

只須要更改每一個[yolo]層前邊卷積層的filter個數便可：

每個[region/yolo]層前的最後一個卷積層中的 filters=num(yolo層個數)*(classes+5) ,5的意義是5個座標，論文中的tx,ty,tw,th,po

舉個例子：我有三個類，n = 3, 那麼filter = 3 * (3+5) = 21

[convolutional]
size=1
stride=1
pad=1
filters=255 # 改成 21
activation=linear


[yolo]
mask = 6,7,8
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=80 # 改成 3
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

6. 數據集格式說明

- yolov3
    - data
      - 2007_train.txt
      - 2007_test.txt
      - coco.names
      - coco.data
      - annotations(json files)
      - images(將2007_train.txt中的圖片放到train2014文件夾中，test同理)
        - train2014
          - 0001.jpg
          - 0002.jpg
        - val2014
          - 0003.jpg
          - 0004.jpg
      - labels（voc_labels.py生成的內容須要從新組織一下）
        - train2014
          - 0001.txt
          - 0002.txt
        - val2014
          - 0003.txt
          - 0004.txt
      - samples(存放待測試圖片)

2007_train.txt內容示例：

/home/dpj/yolov3-master/data/images/val2014/Cow_1192.jpg
/home/dpj/yolov3-master/data/images/val2014/Cow_1196.jpg
.....

注意images和labels文件架構一致性，由於txt是經過簡單的替換獲得的：

images -> labels
.jpg -> .txt

3. 訓練模型

預訓練模型：

Darknet *.weights format: https://pjreddie.com/media/files/yolov3.weights
PyTorch *.pt format: https://drive.google.com/drive/folders/1uxgUBemJVw9wZsdpboYbzUN4bcRhsuAI

開始訓練：

python train.py --data data/coco.data --cfg cfg/yolov3.cfg

若是日誌正常輸出那證實能夠運行了

若是中斷了，能夠恢復訓練

python train.py --data data/coco.data --cfg cfg/yolov3.cfg --resume

4. 測試模型

將待測試圖片放到data/samples中，而後運行

python detect.py --weights weights/best.pt

5. 評估模型

python test.py --weights weights/latest.pt

若是使用cocoAPI使用如下命令：

git clone https://github.com/cocodataset/cocoapi && cd cocoapi/PythonAPI && make && cd ../.. && cp -r cocoapi/PythonAPI/pycocotools yolov3
cd yolov3
 
python3 test.py --save-json --img-size 416
Namespace(batch_size=32, cfg='cfg/yolov3-spp.cfg', conf_thres=0.001, data_cfg='data/coco.data', img_size=416, iou_thres=0.5, nms_thres=0.5, save_json=True, weights='weights/yolov3-spp.weights')
Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16130MB)
               Class    Images   Targets         P         R       mAP        F1
Calculating mAP: 100%|█████████████████████████████████████████| 157/157 [05:59<00:00,  1.71s/it]
                 all     5e+03  3.58e+04     0.109     0.773      0.57     0.186
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.335
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.565
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.349
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.151
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.360
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.493
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.280
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.432
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.458
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.255
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.494
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.620

python3 test.py --save-json --img-size 608 --batch-size 16
Namespace(batch_size=16, cfg='cfg/yolov3-spp.cfg', conf_thres=0.001, data_cfg='data/coco.data', img_size=608, iou_thres=0.5, nms_thres=0.5, save_json=True, weights='weights/yolov3-spp.weights')
Using CUDA device0 _CudaDeviceProperties(name='Tesla V100-SXM2-16GB', total_memory=16130MB)
               Class    Images   Targets         P         R       mAP        F1
Computing mAP: 100%|█████████████████████████████████████████| 313/313 [06:11<00:00,  1.01it/s]
                 all     5e+03  3.58e+04      0.12      0.81     0.611     0.203
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.366
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.607
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.386
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.207
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.391
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.485
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.296
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.464
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.494
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.331
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.517
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.618

6. 可視化

可使用python -c from utils import utils;utils.plot_results()

建立drawLog.py

def plot_results():
    # Plot YOLO training results file 'results.txt'
    import glob
    import numpy as np
    import matplotlib.pyplot as plt
    #import os; os.system('rm -rf results.txt && wget https://storage.googleapis.com/ultralytics/results_v1_0.txt')

    plt.figure(figsize=(16, 8))
    s = ['X', 'Y', 'Width', 'Height', 'Objectness', 'Classification', 'Total Loss', 'Precision', 'Recall', 'mAP']
    files = sorted(glob.glob('results.txt'))
    for f in files:
        results = np.loadtxt(f, usecols=[2, 3, 4, 5, 6, 7, 8, 17, 18, 16]).T  # column 16 is mAP
        n = results.shape[1]
        for i in range(10):
            plt.subplot(2, 5, i + 1)
            plt.plot(range(1, n), results[i, 1:], marker='.', label=f)
            plt.title(s[i])
            if i == 0:
                plt.legend()
    plt.savefig('./plot.png')
if __name__ == "__main__":
    plot_results()