CV计算机视觉核心07-目标检测yolo v2、v3(yolo初始版本的v0和v1版本代码)

CV计算机视觉核心07-目标检测

设计检测类算法的output层，可用已知条件有：

1、检测问题的输出是什么？怎么用数字来表示？
输入是一个矩阵，输出是（x, y, w, h）其中x和y表示目标的左上角坐标，w和h表示目标的长和宽，因此输出是用四个这值来表示的。如果有多个目标N就是N(x, y, w, h)，这样输出就是矩阵。这里需要注意的是（x, y, w, h）只能表示一个矩形，无法表示其他的四边形。

因此为了表示多种四边形，通过四个顶点的方式来表达多种四边形（p1, p2, p3, p4）=> （x1, y1, x2, y2, x3, y3, x4, y4）。因此用八个值表示一个四边形。

除了上面的两种方式外，还可以通过（Cx, Cy, w, h）其中Cx和Cy表示的是矩形的中心，w和h表示的是长和宽。

还可以通过添加angle角度，让矩形有一定的角度旋转。

2、我们已经掌握了分类层的设计方法，是否有用？
有用，通过窗口的内容，进行分类判断，从而确定是否为我们需要目标检测的内容。
目标检测：滑动窗口分类方法。
这种滑动窗口分类主要是遍历性的分类，此为窗口的大小不能固定，有各种尺度的窗口。
检测的性能取决于分类模型的性能，分类模型性能好，检测就好。

目标检测本质上是分类器，那如何训练？如何组织样本训练呢？
分类器本质上就是一个二分类器，将图片送给模型训练就好。

这里会遇到一个问题就是样本不均衡问题。这个问题在之后的检测问题中都是需要我们来处理的，在训练过程中正负样本比例如何控制。

此外还需要注意，所有窗口大小的图片，都需要resize到统一大小来在模型中进行training，要统一尺度来training。

做一次detect检测，我们需要考虑准确率和速度，因此我们在滑动窗口时，可以stride大一些，这样速度就会提高，但是准确度就会下降，因此需要找到一个平衡。

分类器：输入matrix => 输出vector
下图中输出的c表示置信度confidence。

3、其实我们有个隐藏技能：拟合层的设计方法。
把分类层的onehot换成一般向量。
检测问题：直接预测出目标位置（置信度c. x, y, w, h）

以上设计可以解决单类目标的检测，多类目标怎么办？

解决分类问题：增加一个onehot
如果检测10类，那么及时fc[5+10]，其中10就可以同onehot表示。

解决多类别问题：多组输出表示多目标多类别。

多输出与位置强相关：体特征的相对应图片整体位置没有变化，即体特征位置不变性。

多组输出与位置强相关：按照空间位置来映射，N = W’ * H’

同一个位置，有时候出现大目标，有时候出现小目标，怎么办？
如果大目标出现，可以通过中心点确定是在哪个位置。
可以设计两个不同大小的grid，一个网格是控制大目标，一个网格是控制小目标，这样两个loss分开做反向传播。

这个方法就是yolo：

前项计算：

训练的网络直接预测了98个框，如何得到最终的3个目标？

方法1：聚类，聚成3类，然后在这3类中，取置信度得分最大的框
如果两个目标本身就比较接近呢？
两个目标与剩下的一个目标比较远呢？
如果不知道到底有几个目标呢？
方法1失效。

方法2：可以通过p(object)，即c表示置信度confidence，p概率最高的来进行筛选，即score分数的大小。
如果不知道到底有几个目标呢？
方法1和2失效。

方法3：比如说score前三的框，我们无法确定三个框是三个目标、还是两个目标、还是一个目标，只有绘制显示出来人眼才好查看。如果设计一个算法，代替我们人眼来检测，算法算一下这个框是否代表一个目标，如果是一个目标，这个框就不输出，找一下同一个目标score最大的输出；如果不是一个目标，把表示一个目标的框弄到一起，score一个最大的输出。

假设：两个框重合度很高，默认他们表示的是一个物体，如果假设为真，我们可以根据框的重合度来完成“聚类”，两两遍历，认为重合度较大的框们表示一个物体。

如何衡量重合度？loU交并比，交集/并集 [0到1]

两两遍历算重合度，计算量有点大，如何减少计算量：非极大值抑制NMS

以上去除冗余框的过程：非极大值抑制NMS。用于我们不知道有多少个框的时候。

Yolo:


总结检测网络的输出层：检测层的设计。
回归坐标值+onehot分类:
fc=>fc => ( C, x, y, w, h, one-hot )
fc作用是提取全局特征。

Batch Normalization : BN层（可以解决离散的问题）
测试时用指数加权平均的方法。
先BN再激活，对不易受训练影响的激活前进行BN。
因为激活函数越是两边越离散，因此需要压缩到均值为0方差为1的情况。
此外还需要注意的是，BN并不完全是均值位0方差为1，因为对于sigmoid激活函数而言，0处是呈现线性的情况，非线性能力较弱，因此需要有一定的偏移和不同情况，使得有大量非线性表现能力。因此可以通过我们loss的需求调整改变BN层的均值和方差。

relu解决sigmoid的梯度离散问题，relu梯度为1，解决了梯度离散的问题。

检侧问题的评估方式：mAP Mean Average Precision

评价一个检测模型需要三个指标，一个是precision 正确/所有输出的框；和另一个是recall 正确/应该输出的宽；以及threshold阈值。
如果要满足recall，那么precision就会下降。

同时使用三个指标来评估检测模型，是比较困难的。因此需要将这三个指标统一起来。

Recall: TP(P表示positive，识别为正的；T表示true)

V2

yolo的损失函数：




这里是降采样5次，2的5次方32，因此输入到网络中的大小最好是32的倍数，且降采样后最好输出的是奇数个大小，比如下面输入的是416416，最后5次降采样为1313（416/32=13）。奇数个会有对应的中心点。



这里是预测的相对位置：






既能在小尺度的检测到，也能在大尺度的检测到：

v3

下图3个不同大小的scale，以及每个scale有3个不同大小的后选框：


1313的做一个上采样得到2626，再跟原来中间的26*26做一个融合拼接：








大于某个阈值的都可以取出来，这样可以预测多标签的任务：

初始版本v0的yolo:

这里所使用的数据集下载地址：https://download.csdn.net/download/m0_37755995/86224061

用于创建自定义数据集的加载PennFudanDataset_main.py

import os
import numpy as np
import torch
from PIL import Image


class PennFudanDataset(object):
    def __init__(self, root, transforms):
        self.root = root
        self.transforms = transforms
        # load all image files, sorting them to
        # ensure that they are aligned
        self.imgs = list(sorted(os.listdir(os.path.join(root, "PNGImages"))))
        self.masks = list(sorted(os.listdir(os.path.join(root, "PedMasks"))))
    # dataset[0]
    def __getitem__(self, idx):
        # load images ad masks
        img_path = os.path.join(self.root, "PNGImages", self.imgs[idx])
        mask_path = os.path.join(self.root, "PedMasks", self.masks[idx])
        img = Image.open(img_path).convert("RGB")
        # note that we haven't converted the mask to RGB,
        # because each color corresponds to a different instance
        # with 0 being background
        mask = Image.open(mask_path)
        # convert the PIL Image into a numpy array
        mask = np.array(mask)
        # instances are encoded as different colors
        obj_ids = np.unique(mask)
        # first id is the background, so remove it
        obj_ids = obj_ids[1:]

        # split the color-encoded mask into a set
        # of binary masks
        masks = mask == obj_ids[:, None, None]

        # get bounding box coordinates for each mask
        num_objs = len(obj_ids)
        boxes = []
        for i in range(num_objs):
            pos = np.where(masks[i])
            xmin = np.min(pos[1])
            xmax = np.max(pos[1])
            ymin = np.min(pos[0])
            ymax = np.max(pos[0])
            boxes.append([xmin, ymin, xmax, ymax])

        # convert everything into a torch.Tensor
        boxes = torch.as_tensor(boxes, dtype=torch.float32)
        # there is only one class
        labels = torch.ones((num_objs,), dtype=torch.int64)
        masks = torch.as_tensor(masks, dtype=torch.uint8)

        image_id = torch.tensor([idx])
        area = (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0])
        # suppose all instances are not crowd
        iscrowd = torch.zeros((num_objs,), dtype=torch.int64)

        target = {}
        target["boxes"] = boxes
        target["labels"] = labels
        target["masks"] = masks
        target["image_id"] = image_id
        target["area"] = area
        target["iscrowd"] = iscrowd

        if self.transforms is not None:
            img, target = self.transforms(img, target)

        return img, target
    # len(dataset)
    def __len__(self):
        return len(self.imgs)


import transforms as T
def get_transform(train):
    transforms = []
    transforms.append(T.ToTensor())
    if train:
        transforms.append(T.RandomHorizontalFlip(0.5))
    return T.Compose(transforms)

v0yolo_model.py

#coding:utf-8
import torch
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
import torch.nn.functional as F
import math

class VGG(nn.Module):
    def __init__(self):
       super(VGG,self).__init__()
       # the vgg's layers
       #self.features = features
       cfg = [64,64,'M',128,128,'M',256,256,256,'M',512,512,512,'M',512,512,512,'M']
       layers= []
       batch_norm = False
       in_channels = 3
       for v in cfg:
           if v == 'M':
               layers += [nn.MaxPool2d(kernel_size=2,stride = 2)]
           else:
               conv2d = nn.Conv2d(in_channels,v,kernel_size=3,padding = 1)
               if batch_norm:
                   layers += [conv2d,nn.Batchnorm2d(v),nn.ReLU(inplace=True)]
               else:
                   layers += [conv2d,nn.ReLU(inplace=True)]
               in_channels = v
       # use the vgg layers to get the feature
       self.features = nn.Sequential(*layers)
       # 全局池化
       self.avgpool = nn.AdaptiveAvgPool2d((7,7))
       # 决策层：分类层
       self.classifier = nn.Sequential(
           nn.Linear(512*7*7,4096),
           nn.ReLU(True),
           nn.Dropout(),
           nn.Linear(4096,4096),
           nn.ReLU(True),
           nn.Dropout(),
           nn.Linear(4096,1000),
       )

       for m in self.modules():
           if isinstance(m,nn.Conv2d):
               nn.init.kaiming_normal_(m.weight,mode='fan_out',nonlinearity='relu')
               if m.bias is not None: 
                   nn.init.constant_(m.bias,0)
           elif isinstance(m,nn.BatchNorm2d):
               nn.init.constant_(m.weight,1)
               nn.init.constant_(m.bias,1)
           elif isinstance(m,nn.Linear):
               nn.init.normal_(m.weight,0,0.01)
               nn.init.constant_(m.bias,0)

    def forward(self,x):
         x = self.features(x)
         x_fea = x
         x = self.avgpool(x)
         x_avg = x
         x = x.view(x.size(0),-1)
         x = self.classifier(x)
         return x,x_fea,x_avg
    def extractor(self,x):
         x = self.features(x)
         return x

class YOLOV0(nn.Module):
    def __init__(self):
       super(YOLOV0,self).__init__()
       vgg = VGG()
       self.extractor = vgg.extractor
       # 这里的avgpool就相当于ROIpooling
       self.avgpool = nn.AdaptiveAvgPool2d((7,7))
       # 决策层：检测层
       self.detector = nn.Sequential(
          #  这里的输入要和线性层能够匹配上。
          #  从25088=>4096
          nn.Linear(512*7*7,4096),
          #  经激活函数
          nn.ReLU(True),
          nn.Dropout(),
          #nn.Linear(4096,1470),
          #  5表示[c,x,y,w,h]
          nn.Linear(4096,5),
       )
       for m in self.modules():
           if isinstance(m,nn.Conv2d):
               nn.init.kaiming_normal_(m.weight,mode='fan_out',nonlinearity='relu')
               if m.bias is not None: 
                   nn.init.constant_(m.bias,0)
           elif isinstance(m,nn.BatchNorm2d):
               nn.init.constant_(m.weight,1)
               nn.init.constant_(m.bias,1)
           elif isinstance(m,nn.Linear):
               nn.init.normal_(m.weight,0,0.01)
               nn.init.constant_(m.bias,0)
    def forward(self,x):
        # 用vgg提取特征
        x = self.extractor(x)
        # x = torch.Size([1, 512, 16, 16])
        # print('x_feature:',x.shape)

        #import pdb
        #pdb.set_trace()

        # 均值pooling，就是roipooling
        x = self.avgpool(x)
        # 减小长宽
        # x_avgpool: torch.Size([1, 512, 7, 7])
        # print('x_avgpool:',x.shape)

        x = x.view(x.size(0),-1)
        # x降维： torch.Size([1, 25088])
        # print('x降维：',x.shape)

        x = self.detector(x)
        # x_detector: torch.Size([1, 5])
        # print('x_detector:',x.shape)

        b,_ = x.shape
        #x = x.view(b,7,7,30) (不输出b,7,7,30了)=> 这里只检测一个目标[1,5]，获得一个框用来直接拟合。
        x = x.view(b,1,1,5)
        return x
     
if __name__ == '__main__':
    vgg = VGG()
    # 这里的x是随机生成的
    x  = torch.randn(1,3,512,512)
    # 将x输入到vgg模型中
    feature,x_fea,x_avg = vgg(x)
    # 打印输出结果
    # torch.Size([1, 1000])
    # torch.Size([1, 512, 16, 16])
    # torch.Size([1, 512, 7, 7])
    print(feature.shape)
    print(x_fea.shape)
    print(x_avg.shape)
 
    yolov0 = YOLOV0()
    # 注意这里是yolo的初始版本，1*1*1*5 其中5表示[c,x,y,w,h]
    feature = yolov0(x)
    # feature_size b*7*7*30
    # torch.Size([1, 1, 1, 5])
    print(feature.shape)
    print(feature)

v0yolotrain.py

#coding:utf-8
from PennFudanDataset_main import *

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.autograd import Variable
from torch.utils.data import DataLoader
from v0yolo_model import *
import cv2
import numpy as np
import time
import sys
import os


## 数据处理
#服务器上的地址 /data/2020-722-YOLOV4-Practical-datasets/PenFudanPed
# dataset地址：/Users/zhaomignming/Documents/mmteacher/datasets
#datapath='/Users/zhaomignming/Documents/mmteacher/datasets/PennFudanPed'
# datapath='/Users/zhaomingming/data_sets/PennFudanPed'
datapath = 'zhaomingming'

dataset = PennFudanDataset(datapath, get_transform(train=False))
dataset_test = PennFudanDataset(datapath, get_transform(train=False))

indices = torch.randperm(len(dataset)).tolist()
#dataset = torch.utils.data.Subset(dataset, indices[:-50])
#dataset_test = torch.utils.data.Subset(dataset_test, indices[-50:])
#dataset = torch.utils.data.Subset(dataset, indices[0:1])
#import pdb
#pdb.set_trace()
#dataset = torch.utils.data.Subset(dataset, indices[0:1])
dataset = torch.utils.data.Subset(dataset, [0])
dataset_test = torch.utils.data.Subset(dataset_test, indices[0:2])

def collate_fn(batch):
    return tuple(zip(*batch))
# define training and validation data loaders
train_loader = torch.utils.data.DataLoader(
        dataset, batch_size=1, shuffle=False, num_workers=1,
        collate_fn=collate_fn)

val_loader = torch.utils.data.DataLoader(
        dataset_test, batch_size=2, shuffle=False, num_workers=4,
        collate_fn=collate_fn)


def input_process(batch):
    #import pdb
    #pdb.set_trace()
    #batch[0]，0维是表示图片数量
    batch_size=len(batch[0])
    # 这里是输入，每张图片是3通道448,448
    input_batch= torch.zeros(batch_size,3,448,448)
    for i in range(batch_size):
        inputs_tmp = Variable(batch[0][i])
        inputs_tmp1=cv2.resize(inputs_tmp.permute([1,2,0]).numpy(),(448,448))
        inputs_tmp2=torch.tensor(inputs_tmp1).permute([2,0,1])
        input_batch[i:i+1,:,:,:]= torch.unsqueeze(inputs_tmp2,0)
    return input_batch 

#batch[1][0]['boxes'][0]
def target_process(batch):
    batch_size=len(batch[0])
    target_batch= torch.zeros(batch_size,1,1,5)
    #import pdb
    #pdb.set_trace()
    for i in range(batch_size):
        #只处理batch中的第一张图片
        # batch[1]表示label
        # batch[0]表示image
        bbox=batch[1][i]['boxes'][0]
        #这里是获得图片的channel、x、y
        _,hi,wi = batch[0][i].numpy().shape
        #下面bbox是通过归一化，是一个大于0小于1的数值了
        bbox = bbox/ torch.tensor([wi,hi,wi,hi])
        #这里bbox置信度肯定是1。通过concat实现[c,w,h]
        cbbox =  torch.cat([torch.ones(1),bbox])
        #放到四维矩阵中
        target_batch[i:i+1,:,:,:] = torch.unsqueeze(cbbox,0)
    return target_batch
    

num_classes = 2
n_class    = 2
batch_size = 6
epochs     = 500
lr         = 1e-3
momentum   = 0
w_decay    = 1e-5
step_size  = 50
gamma      = 0.5

# 定义模型
yolov0_model = YOLOV0()
# import pdb
# pdb.set_trace()
# 定义优化算法为sdg:随机梯度下降
optimizer = optim.SGD(yolov0_model.detector.parameters(), lr=lr, momentum=momentum, weight_decay=w_decay)

# 定义学习率变化策略
# 每30个epoch 学习率乘以0.5
scheduler = lr_scheduler.StepLR(optimizer, step_size=step_size, gamma=gamma)  # decay LR by a factor of 0.5 every 30 epochs

# 矩阵形式写法，写法简单，但是可读性不强
def lossfunc(outputs,labels):
    #import  pdb
    #pdb.set_trace()
    tmp = (outputs-labels)**2
    return torch.sum(tmp,0).view(1,5).mm(torch.tensor([10,0.0001,0.0001,0.0001,0.0001]).view(5,1))

# 定义直接拟合的学习率，可读性强
def lossfunc_details(outputs,labels):
    # 判断维度
    assert ( outputs.shape == labels.shape),"outputs shape[%s] not equal labels shape[%s]"%(outputs.shape,labels.shape)
    b,w,h,c = outputs.shape #[b,1,1,5]
    loss = 0
    #遍历每个batch图片的每个cell的loss
    for bi in range(b):
        for wi in range(w):
            for hi in range(h):
                #import pdb
                #pdb.set_trace()
                # detect_vector=[confidence,x,y,w,h]
                detect_vector = outputs[bi,wi,hi]
                gt_dv = labels[bi,wi,hi]
                conf_pred = detect_vector[0]
                conf_gt = gt_dv[0]
                x_pred = detect_vector[1]
                x_gt = gt_dv[1]
                y_pred = detect_vector[2]
                y_gt = gt_dv[2]
                w_pred = detect_vector[3]
                w_gt = gt_dv[3]
                h_pred = detect_vector[4]
                h_gt = gt_dv[4]
                loss_confidence = (conf_pred-conf_gt)**2 
                #loss_geo = (x_pred-x_gt)**2 + (y_pred-y_gt)**2 + (w_pred**0.5-w_gt**0.5)**2 + (h_pred**0.5-h_gt**0.5)**2
                loss_geo = (x_pred-x_gt)**2 + (y_pred-y_gt)**2 + (w_pred-w_gt)**2 + (h_pred-h_gt)**2
                loss_tmp = loss_confidence + 0.3*loss_geo
                #print("loss[%s,%s] = %s,%s"%(wi,hi,loss_confidence.item(),loss_geo.item()))
                loss += loss_tmp
    return loss
# train
def train():
    for epoch in range(epochs):
        ts = time.time()
        # 这里是直接开始train_loader了
        for iter, batch in enumerate(train_loader):
            # 梯度清零
            optimizer.zero_grad()
            # 取图片
            inputs = input_process(batch)
            print('inputs.shape:',inputs.shape)
            # 取标注
            labels = target_process(batch)
            print('labels:',labels)
            #import pdb
            #pdb.set_trace()
            # 将图片输入模型中个，获得输出。
            # 获取得到输出
            outputs = yolov0_model(inputs)
            print('outputs:',outputs)
            #import pdb
            #pdb.set_trace()
            #loss = criterion(outputs, labels)
            # 预测与真实标签计算loss
            # 这里传入的labels是和outputs结构一样的：
            loss = lossfunc_details(outputs,labels)
            loss.backward()
            optimizer.step()
            #print(torch.cat([outputs.detach().view(1,5),labels.view(1,5)],0).view(2,5))
            if iter % 10 == 0:
            #    print(torch.cat([outputs.detach().view(1,5),labels.view(1,5)],0).view(2,5))
                print("epoch{}, iter{}, loss: {}, lr: {}".format(epoch, iter, loss.data.item(),optimizer.state_dict()['param_groups'][0]['lr']))
        
        #print("Finish epoch {}, time elapsed {}".format(epoch, time.time() - ts))
        #print("*"*30)
        #val(epoch)
        scheduler.step()
# inference
def val(epoch):
    yolov0_model.eval()
    total_ious = []
    pixel_accs = []
    for iter, batch in enumerate(val_loader):
        inputs = input_process(batch)
        target,label= target_process(batch)

        output = yolov1_model(inputs)
        output = output.data.cpu().numpy()

        N, _, h, w = output.shape
        pred = output.transpose(0, 2, 3, 1).reshape(-1, n_class).argmax(axis=1).reshape(N, h, w)


if __name__ == "__main__":
    train()

v1版本的yolo:

用于创建自定义数据集的加载PennFudanDataset_main.py

import os
import numpy as np
import torch
from PIL import Image


class PennFudanDataset(object):
    def __init__(self, root, transforms):
        self.root = root
        self.transforms = transforms
        # load all image files, sorting them to
        # ensure that they are aligned
        self.imgs = list(sorted(os.listdir(os.path.join(root, "PNGImages"))))
        self.masks = list(sorted(os.listdir(os.path.join(root, "PedMasks"))))
    # dataset[0]
    def __getitem__(self, idx):
        # load images ad masks
        img_path = os.path.join(self.root, "PNGImages", self.imgs[idx])
        mask_path = os.path.join(self.root, "PedMasks", self.masks[idx])
        img = Image.open(img_path).convert("RGB")
        # note that we haven't converted the mask to RGB,
        # because each color corresponds to a different instance
        # with 0 being background
        mask = Image.open(mask_path)
        # convert the PIL Image into a numpy array
        mask = np.array(mask)
        # instances are encoded as different colors
        obj_ids = np.unique(mask)
        # first id is the background, so remove it
        obj_ids = obj_ids[1:]

        # split the color-encoded mask into a set
        # of binary masks
        masks = mask == obj_ids[:, None, None]

        # get bounding box coordinates for each mask
        num_objs = len(obj_ids)
        boxes = []
        for i in range(num_objs):
            pos = np.where(masks[i])
            xmin = np.min(pos[1])
            xmax = np.max(pos[1])
            ymin = np.min(pos[0])
            ymax = np.max(pos[0])
            boxes.append([xmin, ymin, xmax, ymax])

        # convert everything into a torch.Tensor
        boxes = torch.as_tensor(boxes, dtype=torch.float32)
        # there is only one class
        labels = torch.ones((num_objs,), dtype=torch.int64)
        masks = torch.as_tensor(masks, dtype=torch.uint8)

        image_id = torch.tensor([idx])
        area = (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0])
        # suppose all instances are not crowd
        iscrowd = torch.zeros((num_objs,), dtype=torch.int64)

        target = {}
        target["boxes"] = boxes
        target["labels"] = labels
        target["masks"] = masks
        target["image_id"] = image_id
        target["area"] = area
        target["iscrowd"] = iscrowd

        if self.transforms is not None:
            img, target = self.transforms(img, target)

        return img, target
    # len(dataset)
    def __len__(self):
        return len(self.imgs)


import transforms as T
def get_transform(train):
    transforms = []
    transforms.append(T.ToTensor())
    if train:
        transforms.append(T.RandomHorizontalFlip(0.5))
    return T.Compose(transforms)

v1yolomodel.py

#coding:utf-8
import torch
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
import torch.nn.functional as F
import math

class VGG(nn.Module):
    def __init__(self):
       super(VGG,self).__init__()
       # the vgg's layers
       #self.features = features
       cfg = [64,64,'M',128,128,'M',256,256,256,'M',512,512,512,'M',512,512,512,'M']
       layers= []
       batch_norm = False
       in_channels = 3
       for v in cfg:
           if v == 'M':
               layers += [nn.MaxPool2d(kernel_size=2,stride = 2)]
           else:
               conv2d = nn.Conv2d(in_channels,v,kernel_size=3,padding = 1)
               if batch_norm:
                   layers += [conv2d,nn.Batchnorm2d(v),nn.ReLU(inplace=True)]
               else:
                   layers += [conv2d,nn.ReLU(inplace=True)]
               in_channels = v
       # use the vgg layers to get the feature
       self.features = nn.Sequential(*layers)
       # 全局池化
       self.avgpool = nn.AdaptiveAvgPool2d((7,7))
       # 决策层：分类层
       self.classifier = nn.Sequential(
           nn.Linear(512*7*7,4096),
           nn.ReLU(True),
           nn.Dropout(),
           nn.Linear(4096,4096),
           nn.ReLU(True),
           nn.Dropout(),
           nn.Linear(4096,1000),
       )

       for m in self.modules():
           if isinstance(m,nn.Conv2d):
               nn.init.kaiming_normal_(m.weight,mode='fan_out',nonlinearity='relu')
               if m.bias is not None: 
                   nn.init.constant_(m.bias,0)
           elif isinstance(m,nn.BatchNorm2d):
               nn.init.constant_(m.weight,1)
               nn.init.constant_(m.bias,1)
           elif isinstance(m,nn.Linear):
               nn.init.normal_(m.weight,0,0.01)
               nn.init.constant_(m.bias,0)

    def forward(self,x):
         x = self.features(x)
         x_fea = x
         x = self.avgpool(x)
         x_avg = x
         x = x.view(x.size(0),-1)
         x = self.classifier(x)
         return x,x_fea,x_avg
    def extractor(self,x):
         x = self.features(x)
         return x

class YOLOV1(nn.Module):
    def __init__(self):
       super(YOLOV1,self).__init__()
       vgg = VGG()
       self.extractor = vgg.extractor
       self.avgpool = nn.AdaptiveAvgPool2d((7,7))
       # 决策层：检测层
       self.detector = nn.Sequential(
          nn.Linear(512*7*7,4096),
          nn.ReLU(True),
          nn.Dropout(),
          #nn.Linear(4096,1470),
          nn.Linear(4096,245),
          #nn.Linear(4096,5),
       )
       for m in self.modules():
           if isinstance(m,nn.Conv2d):
               nn.init.kaiming_normal_(m.weight,mode='fan_out',nonlinearity='relu')
               if m.bias is not None: 
                   nn.init.constant_(m.bias,0)
           elif isinstance(m,nn.BatchNorm2d):
               nn.init.constant_(m.weight,1)
               nn.init.constant_(m.bias,1)
           elif isinstance(m,nn.Linear):
               nn.init.normal_(m.weight,0,0.01)
               nn.init.constant_(m.bias,0)
    def forward(self,x):
        x = self.extractor(x)
        #import pdb
        #pdb.set_trace()
        x = self.avgpool(x)
        x = x.view(x.size(0),-1)
        x = self.detector(x)
        # detector: torch.Size([1, 245]) 这里的245是7*7*5所得
        print('detector:',x.shape)
        b,_ = x.shape
        #x = x.view(b,7,7,30)
        # 这里我们只有预测框，没有分类
        x = x.view(b,7,7,5)
        
        #x = x.view(b,1,1,5)
        return x
     
if __name__ == '__main__':
    vgg = VGG()
    x  = torch.randn(1,3,512,512)
    feature,x_fea,x_avg = vgg(x)
    print(feature.shape)
    print(x_fea.shape)
    print(x_avg.shape)
 
    yolov1 = YOLOV1()
    feature = yolov1(x)
    # feature_size b*7*7*30
    # feature.shape: torch.Size([1, 7, 7, 5])
    print('feature.shape:',feature.shape)
    print(feature)
   

v1yolotrain.py

#coding:utf-8
from PennFudanDataset_main import *

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.autograd import Variable
from torch.utils.data import DataLoader
from v1yolomodel import *
import cv2
import numpy as np
import time
import sys
import os


## 数据处理
#服务器上的地址 /data/2020-722-YOLOV4-Practical-datasets/PenFudanPed
# dataset地址：/Users/zhaomignming/Documents/mmteacher/datasets
#datapath='/Users/zhaomignming/Documents/mmteacher/datasets/PennFudanPed'
datapath='/Users/zhaomingming/data_sets/PennFudanPed'
# datapath = 'zhaomingming'
dataset = PennFudanDataset(datapath, get_transform(train=False))
dataset_test = PennFudanDataset(datapath, get_transform(train=False))

indices = torch.randperm(len(dataset)).tolist()
#dataset = torch.utils.data.Subset(dataset, indices[:-50])
#dataset_test = torch.utils.data.Subset(dataset_test, indices[-50:])
#dataset = torch.utils.data.Subset(dataset, indices[0:1])
#import pdb
#pdb.set_trace()
#dataset = torch.utils.data.Subset(dataset, indices[0:1])
dataset = torch.utils.data.Subset(dataset, [0])
dataset_test = torch.utils.data.Subset(dataset_test, indices[0:2])

def collate_fn(batch):
    return tuple(zip(*batch))
# define training and validation data loaders
train_loader = torch.utils.data.DataLoader(
        dataset, batch_size=1, shuffle=False, num_workers=1,
        collate_fn=collate_fn)

val_loader = torch.utils.data.DataLoader(
        dataset_test, batch_size=2, shuffle=False, num_workers=4,
        collate_fn=collate_fn)


def input_process(batch):
    #import pdb
    #pdb.set_trace()
    batch_size=len(batch[0])
    input_batch= torch.zeros(batch_size,3,448,448)
    for i in range(batch_size):
        inputs_tmp = Variable(batch[0][i])
        inputs_tmp1=cv2.resize(inputs_tmp.permute([1,2,0]).numpy(),(448,448))
        inputs_tmp2=torch.tensor(inputs_tmp1).permute([2,0,1])
        input_batch[i:i+1,:,:,:]= torch.unsqueeze(inputs_tmp2,0)
    return input_batch 

#batch[1][0]['boxes'][0]
def target_process(batch,grid_number=7):
    #中心点落在哪个框中，这个框的执行度就是1。其余全为0。
    # batch[1]表示label
    # batch[0]表示image
    batch_size=len(batch[0])
    target_batch= torch.zeros(batch_size,grid_number,grid_number,5)
    #import pdb
    #pdb.set_trace()
    for i in range(batch_size):
        labels = batch[1]
        batch_labels = labels[i]
        #import pdb
        #pdb.set_trace()
        number_box = len(batch_labels['boxes'])
        for wi in range(grid_number):
            for hi in range(grid_number):
                # 遍历每个标注的框
                for bi in range(number_box):
                    bbox=batch_labels['boxes'][bi]
                    _,himg,wimg = batch[0][i].numpy().shape
                    # 框归一化一下
                    bbox = bbox/ torch.tensor([wimg,himg,wimg,himg])
                    #import pdb
                    #pdb.set_trace()
                    # 计算框的中心点
                    center_x= (bbox[0]+bbox[2])*0.5
                    center_y= (bbox[1]+bbox[3])*0.5
                    #print("[%s,%s,%s],[%s,%s,%s]"%(wi/grid_number,center_x,(wi+1)/grid_number,hi/grid_number,center_y,(hi+1)/grid_number))
                    # 判断中心点有没有落在当前cell中，这里是7*7*5,如果是7*7*10时,就需要判断一下是大框还是小框。
                    if center_x<=(wi+1)/grid_number and center_x>=wi/grid_number and center_y<=(hi+1)/grid_number and center_y>= hi/grid_number:
                        #pdb.set_trace()
                        cbbox =  torch.cat([torch.ones(1),bbox])
                        # 中心点落在grid内，
                        target_batch[i:i+1,wi:wi+1,hi:hi+1,:] = torch.unsqueeze(cbbox,0)
                    #else:
                        #cbbox =  torch.cat([torch.zeros(1),bbox])
                #import pdb
                #pdb.set_trace()
                #rint(target_batch[i:i+1,wi:wi+1,hi:hi+1,:])
                #target_batch[i:i+1,wi:wi+1,hi:hi+1,:] = torch.unsqueeze(cbbox,0)
    return target_batch
    

num_classes = 2
n_class    = 2
batch_size = 6
epochs     = 500
lr         = 1e-3
momentum   = 0
w_decay    = 1e-5
step_size  = 50
gamma      = 0.5

# 定义模型
yolov1_model = YOLOV1()
import pdb
pdb.set_trace()
# 定义优化算法为sdg:随机梯度下降
optimizer = optim.SGD(yolov1_model.detector.parameters(), lr=lr, momentum=momentum, weight_decay=w_decay)

# 定义学习率变化策略
# 每30个epoch 学习率乘以0.5
scheduler = lr_scheduler.StepLR(optimizer, step_size=step_size, gamma=gamma)  # decay LR by a factor of 0.5 every 30 epochs

# 矩阵形式写法，写法简单，但是可读性不强
def lossfunc(outputs,labels):
    #import  pdb
    #pdb.set_trace()
    tmp = (outputs-labels)**2
    return torch.sum(tmp,0).view(1,5).mm(torch.tensor([10,0.0001,0.0001,0.0001,0.0001]).view(5,1))

# 定义直接拟合的学习率，可读性强
def lossfunc_details(outputs,labels):
    # 判断维度
    assert ( outputs.shape == labels.shape),"outputs shape[%s] not equal labels shape[%s]"%(outputs.shape,labels.shape)
    #import pdb
    #pdb.set_trace()
    b,w,h,c = outputs.shape
    loss = 0
    #import pdb
    #pdb.set_trace()
    conf_loss_matrix = torch.zeros(b,w,h)
    geo_loss_matrix = torch.zeros(b,w,h)
    loss_matrix = torch.zeros(b,w,h)
    
    for bi in range(b):
        for wi in range(w):
            for hi in range(h):
                #import pdb
                #pdb.set_trace()
                # detect_vector=[confidence,x,y,w,h]
                detect_vector = outputs[bi,wi,hi]
                gt_dv = labels[bi,wi,hi]
                conf_pred = detect_vector[0]
                conf_gt = gt_dv[0]
                x_pred = detect_vector[1]
                x_gt = gt_dv[1]
                y_pred = detect_vector[2]
                y_gt = gt_dv[2]
                w_pred = detect_vector[3]
                w_gt = gt_dv[3]
                h_pred = detect_vector[4]
                h_gt = gt_dv[4]
                loss_confidence = (conf_pred-conf_gt)**2 
                #loss_geo = (x_pred-x_gt)**2 + (y_pred-y_gt)**2 + (w_pred**0.5-w_gt**0.5)**2 + (h_pred**0.5-h_gt**0.5)**2
            
                loss_geo = (x_pred-x_gt)**2 + (y_pred-y_gt)**2 + (w_pred-w_gt)**2 + (h_pred-h_gt)**2
                loss_geo = conf_gt*loss_geo
                loss_tmp = loss_confidence + 0.3*loss_geo
                #print("loss[%s,%s] = %s,%s"%(wi,hi,loss_confidence.item(),loss_geo.item()))
                loss += loss_tmp
                conf_loss_matrix[bi,wi,hi]=loss_confidence
                geo_loss_matrix[bi,wi,hi]=loss_geo
                loss_matrix[bi,wi,hi]=loss_tmp
    #打印出batch中每张片的位置loss,和置信度输出
    print(geo_loss_matrix)
    print(outputs[0,:,:,0]>0.5)
    return loss,loss_matrix,geo_loss_matrix,conf_loss_matrix
# train
def train():
    for epoch in range(epochs):
        ts = time.time()
        for iter, batch in enumerate(train_loader):
            optimizer.zero_grad()
            # 取图片
            inputs = input_process(batch)
            # 取标注
            labels = target_process(batch)
            
            # 获取得到输出
            outputs = yolov1_model(inputs)
            #import pdb
            #pdb.set_trace()
            #loss = criterion(outputs, labels)
            loss,lm,glm,clm = lossfunc_details(outputs,labels)
            loss.backward()
            optimizer.step()
            #print(torch.cat([outputs.detach().view(1,5),labels.view(1,5)],0).view(2,5))
            if iter % 10 == 0:
            #    print(torch.cat([outputs.detach().view(1,5),labels.view(1,5)],0).view(2,5))
                print("epoch{}, iter{}, loss: {}, lr: {}".format(epoch, iter, loss.data.item(),optimizer.state_dict()['param_groups'][0]['lr']))
        
        #print("Finish epoch {}, time elapsed {}".format(epoch, time.time() - ts))
        #print("*"*30)
        #val(epoch)
        scheduler.step()
# inference
def val(epoch):
    yolov1_model.eval()
    total_ious = []
    pixel_accs = []
    for iter, batch in enumerate(val_loader):
        inputs = input_process(batch)
        target,label= target_process(batch)

        output = yolov1_model(inputs)
        output = output.data.cpu().numpy()

        N, _, h, w = output.shape
        pred = output.transpose(0, 2, 3, 1).reshape(-1, n_class).argmax(axis=1).reshape(N, h, w)
        

if __name__ == "__main__":
    train()