Github復現YOLO v3（包含常見錯誤）

GitHub鏈接：https://github.com/ultralytics/yolov3
訓練單類鏈接：https://github.com/ultralytics/yolov3/wiki/Example:-Train-Single-Class
參考鏈接：https://blog.csdn.net/qq_44787464/article/details/99736670（其實參考這個就足夠復現了）
我只是把這個過程中遇到的問題說一下，同時也按自己的方式記錄一下，爲以後使用的時候不到處搜做個準備。
我的環境是python3.6，CUDA10.0，torch1.2.0，torchvision0.4.0
torchvision0.4.0鏈接：https://pan.baidu.com/s/1sUi-dteb5Muo0pKQodFVmQ
提取碼：l6vv
torch1.2.0鏈接：https://pan.baidu.com/s/1TIeM-iyijhX4rR-AqqXCsA
提取碼：6b8c
如果你的trochvision版本不是0.4.0你有可能遇到以下錯誤
1.

Traceback (most recent call last):
  File ".\train.py", line 412, in <module>
    train()  # train normally
  File ".\train.py", line 317, in train
    dataloader=testloader)
  File "D:\wcs\yolov3-master\test.py", line 98, in test
    output = non_max_suppression(inf_out, conf_thres=conf_thres, iou_thres=iou_thres)  # nms
  File "D:\wcs\yolov3-master\utils\utils.py", line 560, in non_max_suppression
    i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
AttributeError: module 'torchvision' has no attribute 'ops'

Traceback (most recent call last):
  File ".\train.py", line 3, in <module>
    import torch.distributed as dist
  File "C:\ProgramData\Anaconda3\lib\site-packages\torch\__init__.py", line 81, in <module>
    from torch._C import *
ImportError: DLL load failed: 找不到指定的模塊。

1.數據準備
數據結構就是標準的VOC數據結構
鏈接：https://pan.baidu.com/s/1cAuV3ASpwTNlWaglmEp-9w
提取碼：w3kg

（1）生成train.txt,val.txt,test.txt等（不包含路徑）即：

getText.py

import os
import random
 
trainval_percent = 0.1
train_percent = 0.9
xmlfilepath = './data/cell/Annotations'
txtsavepath = './data/cell/ImageSets'
total_xml = os.listdir(xmlfilepath)
 
num = len(total_xml)
lists = range(num)
tv = int(num * trainval_percent)
tr = int(tv * train_percent)
trainval = random.sample(lists, tv)
train = random.sample(trainval, tr)
 
ftrainval = open('./data/cell/ImageSets/trainval.txt', 'w')
ftest = open('./data/cell/ImageSets/test.txt', 'w')
ftrain = open('./data/cell/ImageSets/train.txt', 'w')
fval = open('./data/cell/ImageSets/val.txt', 'w')
 
for i in lists:
    name = total_xml[i][:-4] + '\n'
    if i in trainval:
        ftrainval.write(name)
        if i in train:
            ftest.write(name)
        else:
            fval.write(name)
    else:
        ftrain.write(name)
 
ftrainval.close()
ftrain.close()
fval.close()
ftest.close()

(2)生成標籤信息以及帶路徑的train.txt,val.txt等文件
voc_label.py

import xml.etree.ElementTree as ET
import pickle
import os
from os import listdir, getcwd
from os.path import join
 
sets = ['train', 'test','val']
 
classes = ["RBC"]#there only one class
 
 
def convert(size, box):
    dw = 1. / size[0]
    dh = 1. / size[1]
    x = (box[0] + box[1]) / 2.0
    y = (box[2] + box[3]) / 2.0
    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(image_id):
    in_file = open('./data/cell/Annotations/%s.xml' % (image_id))
    out_file = open('./data/cell/labels/%s.txt' % (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()
print(wd)
for image_set in sets:
    if not os.path.exists('./data/cell/labels/'):
        os.makedirs('./data/cell/labels/')
    image_ids = open('./data/cell/ImageSets/%s.txt' % (image_set)).read().strip().split()
    list_file = open('./data/cell/%s.txt' % (image_set), 'w')
    for image_id in image_ids:
        list_file.write('./data/cell/images/%s.jpg\n' % (image_id))
        convert_annotation(image_id)
    list_file.close()

產生下面的文件：

（3）配置文件

rbc.data（只有一個類別，eval參數那個不用管，我仔細看了源碼，沒有用到）

rbc.names（類別的名字）

2.訓練
訓練之前主要修改配置文件，電腦不行只能訓練那個yolov3-tiny.cfg

修改的內容主要是類別相關的地方，一開始的鏈接裏有了，我截下圖，原始是80類，下面說了計算最後filters的方法，搜索找到filters=255的地方統一修改成18（一類的計算（4+1+1）*3=18），然後找到所有classes的地方改成1就行了

一個類別修改後的cfg文件

[net]
# Testing
batch=1
subdivisions=1
# Training
# batch=64
# subdivisions=2
width=640
height=480
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.001
burn_in=1000
max_batches = 100200
policy=steps
steps=50000,45000
scales=.1,.1

[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[maxpool]
size=2
stride=1

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

###########

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=18
activation=linear



[yolo]
mask = 3,4,5
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=1
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

[route]
layers = -4

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = -1, 8

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=18
activation=linear

[yolo]
mask = 1,2,3
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=1
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

訓練命令：

python3 .\train.py --data ./data/cell/rbc.data --cfg ./cfg/yolov3-tiny.cfg --weights ./weights/yolov3-tiny.pt --epochs 10

yolov3-tiny.pt文件需要翻牆下載，我把所有的都下了，這裏給你們百度鏈接：https://pan.baidu.com/s/1XGH9OpY-DUKfBqptNyBWjA
提取碼：r4jl

3.預測
把需要預測的圖像複製到sample文件夾下

python3 .\detect.py --names ./data/cell/rbc.names --cfg ./cfg/yolov3-tiny.cfg --weights ./weights/best.pt

運行後自動產生output結果文件

完整
鏈接：https://pan.baidu.com/s/1pe79dyifq8f-BSu1DTCN0g
提取碼：qhag