2021-5-25-今日收获

上午：

照常八点起床
上午满课：操作系统、数据挖掘

下午：

写yolov4的主干特征网络：

class Resblock_body(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(Resblock_body, self).__init__()
        self.out_channels = out_channels

        self.conv1 = BasicConv(in_channels, out_channels, 3)

        self.conv2 = BasicConv(out_channels//2, out_channels//2, 3)
        self.conv3 = BasicConv(out_channels//2, out_channels//2, 3)

        self.conv4 = BasicConv(out_channels, out_channels, 1)
        self.maxpool = nn.MaxPool2d([2,2],[2,2])

    def forward(self, x):
        # 利用一个3x3卷积进行特征整合
        x = self.conv1(x)
        # 引出一个大的残差边route
        route = x
        
        c = self.out_channels
        # 对特征层的通道进行分割，取第二部分作为主干部分。
        x = torch.split(x, c//2, dim = 1)[1]
        # 对主干部分进行3x3卷积
        x = self.conv2(x)
        # 引出一个小的残差边route_1
        route1 = x
        # 对第主干部分进行3x3卷积
        x = self.conv3(x)
        # 主干部分与残差部分进行相接
        x = torch.cat([x,route1], dim = 1) 

        # 对相接后的结果进行1x1卷积
        x = self.conv4(x)
        feat = x
        x = torch.cat([route, x], dim = 1)
        
        # 利用最大池化进行高和宽的压缩
        x = self.maxpool(x)
        return x,feat

class CSPDarkNet(nn.Module):
    def __init__(self):
        super(CSPDarkNet, self).__init__()
        # 首先利用两次步长为2x2的3x3卷积进行高和宽的压缩
        # 416,416,3 -> 208,208,32 -> 104,104,64
        self.conv1 = BasicConv(3, 32, kernel_size=3, stride=2)
        self.conv2 = BasicConv(32, 64, kernel_size=3, stride=2)

        # 104,104,64 -> 52,52,128
        self.resblock_body1 =  Resblock_body(64, 64)
        # 52,52,128 -> 26,26,256
        self.resblock_body2 =  Resblock_body(128, 128)
        # 26,26,256 -> 13,13,512
        self.resblock_body3 =  Resblock_body(256, 256)
        # 13,13,512 -> 13,13,512
        self.conv3 = BasicConv(512, 512, kernel_size=3)


        self.num_features = 1
        # 进行权值初始化
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2. / n))
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()


    def forward(self, x):
        # 416,416,3 -> 208,208,32 -> 104,104,64
        x = self.conv1(x)
        x = self.conv2(x)

        # 104,104,64 -> 52,52,128
        x, _    = self.resblock_body1(x)
        # 52,52,128 -> 26,26,256
        x, _    = self.resblock_body2(x)
        # 26,26,256 -> x为13,13,512
        #           -> feat1为26,26,256
        x, feat1    = self.resblock_body3(x)

        # 13,13,512 -> 13,13,512
        x = self.conv3(x)
        feat2 = x
        return feat1,feat2

def darknet53_tiny(pretrained, **kwargs):
    model = CSPDarkNet()
    if pretrained:
        if isinstance(pretrained, str):
            model.load_state_dict(torch.load(pretrained))
        else:
            raise Exception("darknet request a pretrained path. got [{}]".format(pretrained))
    return model

2. train.py

3. #-------------------------------------#
#       对数据集进行训练
#-------------------------------------#
import os
import numpy as np
import time
import torch
from torch.autograd import Variable
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
from torch.utils.data import DataLoader
from utils.dataloader import yolo_dataset_collate, YoloDataset
from nets.yolo_training import YOLOLoss,Generator
from nets.yolo4_tiny import YoloBody
from tqdm import tqdm
from PIL import Image

def get_lr(optimizer):
    for param_group in optimizer.param_groups:
        return param_group['lr']

#---------------------------------------------------#
#   获得类和先验框
#---------------------------------------------------#
def get_classes(classes_path):
    '''loads the classes'''
    with open(classes_path) as f:
        class_names = f.readlines()
    class_names = [c.strip() for c in class_names]
    return class_names

def get_anchors(anchors_path):
    '''loads the anchors from a file'''
    with open(anchors_path) as f:
        anchors = f.readline()
    anchors = [float(x) for x in anchors.split(',')]
    return np.array(anchors).reshape([-1,3,2])

def fit_one_epoch(net,yolo_losses,epoch,epoch_size,epoch_size_val,gen,genval,Epoch,cuda):
    total_loss = 0
    val_loss = 0

    net.train()
    with tqdm(total=epoch_size,desc=f'Epoch {epoch + 1}/{Epoch}',postfix=dict,mininterval=0.3) as pbar:
        for iteration, batch in enumerate(gen):
            if iteration >= epoch_size:
                break
            images, targets = batch[0], batch[1]
            with torch.no_grad():
                #if cuda:
                images = Variable(torch.from_numpy(images).type(torch.FloatTensor)).cuda()
                targets = [Variable(torch.from_numpy(ann).type(torch.FloatTensor)) for ann in targets]
                #else:
                    #images = Variable(torch.from_numpy(images).type(torch.FloatTensor))
                    #targets = [Variable(torch.from_numpy(ann).type(torch.FloatTensor)) for ann in targets]

            #----------------------#
            #   清零梯度
            #----------------------#
            optimizer.zero_grad()
            #----------------------#
            #   前向传播
            #----------------------#
            outputs = net(images)
            losses = []
            num_pos_all = 0
            #----------------------#
            #   计算损失
            #----------------------#
            for i in range(2):
                loss_item, num_pos = yolo_losses[i](outputs[i], targets)
                losses.append(loss_item)
                num_pos_all += num_pos

            loss = sum(losses) / num_pos_all
            #----------------------#
            #   反向传播
            #----------------------#
            loss.backward()
            optimizer.step()

            total_loss += loss.item()
            
            pbar.set_postfix(**{'total_loss': total_loss / (iteration + 1), 
                                'lr'        : get_lr(optimizer)})
            pbar.update(1)

    net.eval()
    print('Start Validation')
    with tqdm(total=epoch_size_val, desc=f'Epoch {epoch + 1}/{Epoch}',postfix=dict,mininterval=0.3) as pbar:
        for iteration, batch in enumerate(genval):
            if iteration >= epoch_size_val:
                break
            images_val, targets_val = batch[0], batch[1]

            with torch.no_grad():
                if cuda:
                    images_val = Variable(torch.from_numpy(images_val).type(torch.FloatTensor)).cuda()
                    targets_val = [Variable(torch.from_numpy(ann).type(torch.FloatTensor)) for ann in targets_val]
                else:
                    images_val = Variable(torch.from_numpy(images_val).type(torch.FloatTensor))
                    targets_val = [Variable(torch.from_numpy(ann).type(torch.FloatTensor)) for ann in targets_val]
                optimizer.zero_grad()
                outputs = net(images_val)
                losses = []
                num_pos_all = 0
                for i in range(2):
                    loss_item, num_pos = yolo_losses[i](outputs[i], targets_val)
                    losses.append(loss_item)
                    num_pos_all += num_pos
                loss = sum(losses) / num_pos_all
                val_loss += loss.item()
            pbar.set_postfix(**{'total_loss': val_loss / (iteration + 1)})
            pbar.update(1)
    print('Finish Validation')
    print('Epoch:'+ str(epoch+1) + '/' + str(Epoch))
    print('Total Loss: %.4f || Val Loss: %.4f ' % (total_loss/(epoch_size+1),val_loss/(epoch_size_val+1)))

    print('Saving state, iter:', str(epoch+1))
    torch.save(model.state_dict(), 'logs/Epoch%d-Total_Loss%.4f-Val_Loss%.4f.pth'%((epoch+1),total_loss/(epoch_size+1),val_loss/(epoch_size_val+1)))


if __name__ == "__main__":

    cuda = True

    Use_Data_Loader = True

    normalize = False
  
    input_shape = (416,416)


    anchors_path = 'model_data/yolo_anchors.txt'
    classes_path = 'model_data/new_classes1.txt'   
    #----------------------------------------------------#
    #   获取classes和anchor
    #----------------------------------------------------#
    class_names = get_classes(classes_path)
    anchors = get_anchors(anchors_path)
    num_classes = len(class_names)
    
    mosaic = False
    Cosine_lr = False
    smoooth_label = 0
    
    #------------------------------------------------------#
    #   创建yolo模型
    #   训练前一定要修改classes_path和对应的txt文件
    #------------------------------------------------------#
    model = YoloBody(len(anchors[0]), num_classes)

    #------------------------------------------------------#
    #   权值文件
    #------------------------------------------------------#
    model_path = "model_data/yolov4_tiny_weights_coco.pth"
    # 加快模型训练的效率
    print('Loading weights into state dict...')
    device = torch.device('cuda')
    model_dict = model.state_dict()
    pretrained_dict = torch.load(model_path, map_location=device)
    pretrained_dict = {k: v for k, v in pretrained_dict.items() if np.shape(model_dict[k]) ==  np.shape(v)}
    model_dict.update(pretrained_dict)
    model.load_state_dict(model_dict)
    print('Finished!')

    net = model.train()

    if cuda:
        net = torch.nn.DataParallel(model)
        cudnn.benchmark = True
        net = net.cuda()

    # 建立loss函数
    yolo_losses = []
    for i in range(2):
        yolo_losses.append(YOLOLoss(np.reshape(anchors,[-1,2]),num_classes, \
                                (input_shape[1], input_shape[0]), smoooth_label, cuda, normalize))

    #----------------------------------------------------#
    #   获得图片路径和标签
    #----------------------------------------------------#
    annotation_path = '2007_train.txt'

    val_split = 0.0
    with open(annotation_path) as f:
        lines = f.readlines()
    np.random.seed(10101)
    np.random.shuffle(lines)
    np.random.seed(None)
    num_val = int(len(lines)*val_split)
    num_train = len(lines) - num_val
    
  
    if True:
        lr = 1e-3
        Batch_size = 32
        Init_Epoch = 0
        Freeze_Epoch = 50
        
        #----------------------------------------------------------------------------#
        #   我在实际测试时，发现optimizer的weight_decay起到了反作用，
        #   所以去除掉了weight_decay，大家也可以开起来试试，一般是weight_decay=5e-4
        #----------------------------------------------------------------------------#
        optimizer = optim.Adam(net.parameters(),lr)
        if Cosine_lr:
            lr_scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=5, eta_min=1e-5)
        else:
            lr_scheduler = optim.lr_scheduler.StepLR(optimizer,step_size=1,gamma=0.92)

        if Use_Data_Loader:
            train_dataset = YoloDataset(lines[:num_train], (input_shape[0], input_shape[1]), mosaic=mosaic, is_train=True)
            val_dataset = YoloDataset(lines[num_train:], (input_shape[0], input_shape[1]), mosaic=False, is_train=False)
            gen = DataLoader(train_dataset, shuffle=False, batch_size=Batch_size, num_workers=2, pin_memory=True,
                                    drop_last=True, collate_fn=yolo_dataset_collate)
            gen_val = DataLoader(val_dataset, shuffle=False, batch_size=Batch_size, num_workers=2,pin_memory=True, 
                                    drop_last=True, collate_fn=yolo_dataset_collate)
        else:
            gen = Generator(Batch_size, lines[:num_train],
                            (input_shape[0], input_shape[1])).generate(train=True, mosaic = mosaic)
            gen_val = Generator(Batch_size, lines[num_train:],
                            (input_shape[0], input_shape[1])).generate(train=False, mosaic = mosaic)

        epoch_size = max(1, num_train//Batch_size)
        epoch_size_val = num_val//Batch_size
        #------------------------------------#
        #   冻结一定部分训练
        #------------------------------------#
        for param in model.backbone.parameters():
            param.requires_grad = False

        for epoch in range(Init_Epoch,Freeze_Epoch):
            fit_one_epoch(net,yolo_losses,epoch,epoch_size,epoch_size_val,gen,gen_val,Freeze_Epoch,cuda)
            lr_scheduler.step()

    if True:
        lr = 1e-4
        Batch_size = 16
        Freeze_Epoch = 50
        Unfreeze_Epoch = 100

        #----------------------------------------------------------------------------#
        #   我在实际测试时，发现optimizer的weight_decay起到了反作用，
        #   所以去除掉了weight_decay，大家也可以开起来试试，一般是weight_decay=5e-4
        #----------------------------------------------------------------------------#
        optimizer = optim.Adam(net.parameters(),lr)
        if Cosine_lr:
            lr_scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=5, eta_min=1e-5)
        else:
            lr_scheduler = optim.lr_scheduler.StepLR(optimizer,step_size=1,gamma=0.92)

        if Use_Data_Loader:
            train_dataset = YoloDataset(lines[:num_train], (input_shape[0], input_shape[1]), mosaic=mosaic, is_train=True)
            val_dataset = YoloDataset(lines[num_train:], (input_shape[0], input_shape[1]), mosaic=False, is_train=False)
            gen = DataLoader(train_dataset, shuffle=False, batch_size=Batch_size, num_workers=2, pin_memory=True,
                                    drop_last=True, collate_fn=yolo_dataset_collate)
            gen_val = DataLoader(val_dataset, shuffle=False, batch_size=Batch_size, num_workers=2,pin_memory=True, 
                                    drop_last=True, collate_fn=yolo_dataset_collate)
        else:
            gen = Generator(Batch_size, lines[:num_train],
                            (input_shape[0], input_shape[1])).generate(train=True, mosaic = mosaic)
            gen_val = Generator(Batch_size, lines[num_train:],
                            (input_shape[0], input_shape[1])).generate(train=False, mosaic = mosaic)

        epoch_size = max(1, num_train//Batch_size)
        epoch_size_val = num_val//Batch_size


        for param in model.backbone.parameters():
            param.requires_grad = True

        for epoch in range(Freeze_Epoch,Unfreeze_Epoch):
            fit_one_epoch(net,yolo_losses,epoch,epoch_size,epoch_size_val,gen,gen_val,Unfreeze_Epoch,cuda)
            lr_scheduler.step()

3. 调整参数
   这个就不放了，因为要比赛，提前放出来对团队不尊重。

晚上：

1. 软工大作业的demo写完了
2. 查看yolov4代码，看不懂的搜索
3. 开会、确定下次的团队内容，我已经把表打出来了

自我控制：

1. 克制住调侃学弟的冲动，因为我感觉那样不好，像挖苦一样
2. 说话越发有礼貌，向大佬请教问题，他还挺愿意教我的
3. 另一个大佬帮我搭建了VPN，接下来就看我了！

总结：

这两天直播很好，注意力很集中的了，然后现在明确明天的学习内容：

1. 看高学长给的学习内容
2. 潘老师给的pytorch手机离线运行课题研究