Diffusion扩散模型学习1——Pytorch搭建DDPM利用深度卷积神经网络实现图片生成
学习前言
我又死了我又死了我又死了！
源码下载地址
https://github.com/bubbliiiing/ddpm-pytorch
喜欢的可以点个star噢。
网络构建
一、什么是Diffusion
如上图所示。DDPM模型主要分为两个过程：
1、Forward加噪过程（从右往左），数据集的真实图片中逐步加入高斯噪声，最终变成一个杂乱无章的高斯噪声，这个过程一般发生在训练的时候。加噪过程满足一定的数学规律。
2、Reverse去噪过程（从左往右），指对加了噪声的图片逐步去噪，从而还原出真实图片，这个过程一般发生在预测生成的时候。尽管在这里说的是加了噪声的图片，但实际去预测生成的时候，是随机生成一个高斯噪声来去噪。去噪的时候不断根据
XtX_tXt的图片生成Xt−1X_{t-1}Xt−1的噪声，从而实现图片的还原。
1、加噪过程
Forward加噪过程主要符合如下的公式：
xt=αtxt−1+1−αtz1x_t=\sqrt{\alpha_t} x_{t-1}+\sqrt{1-\alpha_t} z_{1}xt=αtxt−1+1−αtz1
其中
αt\sqrt{\alpha_t}αt是预先设定好的超参数，被称为Noise schedule，通常是小于1的值，在论文中αt\alpha_tαt的值从0.9999到0.998。ϵt−1∼N(0,1)\epsilon_{t-1} \sim N(0, 1)ϵt−1∼N(0,1)是高斯噪声。由公式（1）迭代推导。xt=at(at−1xt−2+1−αt−1z2)+1−αtz1=atat−1xt−2+(at(1−αt−1)z2+1−αtz1)x_t=\sqrt{a_t}\left(\sqrt{a_{t-1}} x_{t-2}+\sqrt{1-\alpha_{t-1}} z_2\right)+\sqrt{1-\alpha_t} z_1=\sqrt{a_t a_{t-1}} x_{t-2}+\left(\sqrt{a_t\left(1-\alpha_{t-1}\right)} z_2+\sqrt{1-\alpha_t} z_1\right)xt=at(at−1xt−2+1−αt−1z2)+1−αtz1=atat−1xt−2+(at(1−αt−1)z2+1−αtz1)
其中每次加入的噪声都服从高斯分布
z1,z2,…∼N(0,1)z_1, z_2, \ldots \sim \mathcal{N}(0, 1)z1,z2,…∼N(0,1)，两个高斯分布的相加高斯分布满足公式：N(0,σ12)+N(0,σ22)∼N(0,(σ12+σ22))\mathcal{N}\left(0, \sigma_1^2 \right)+\mathcal{N}\left(0, \sigma_2^2 \right) \sim \mathcal{N}\left(0,\left(\sigma_1^2+\sigma_2^2\right) \right)N(0,σ12)+N(0,σ22)∼N(0,(σ12+σ22))，因此，得到xtx_txt的公式为：xt=atat−1xt−2+1−αtαt−1z2x_t = \sqrt{a_t a_{t-1}} x_{t-2}+\sqrt{1-\alpha_t \alpha_{t-1}} z_2xt=atat−1xt−2+1−αtαt−1z2
因此不断往里面套，就能发现规律了，其实就是累乘
可以直接得出
x0x_0x0到xtx_txt的公式：xt=αt‾x0+1−αt‾ztx_t=\sqrt{\overline{\alpha_t}} x_0+\sqrt{1-\overline{\alpha_t}} z_txt=αtx0+1−αtzt
其中
αt‾=∏itαi\overline{\alpha_t}=\prod_i^t \alpha_iαt=∏itαi，这是随Noise schedule设定好的超参数，zt−1∼N(0,1)z_{t-1} \sim N(0, 1)zt−1∼N(0,1)也是一个高斯噪声。通过上述两个公式，我们可以不断的将图片进行破坏加噪。
2、去噪过程
反向过程就是通过估测噪声，多次迭代逐渐将被破坏的
xtx_txt恢复成x0x_0x0，在恢复时刻，我们已经知道的是xtx_txt，这是图片在ttt时刻的噪声图。一下子从xtx_txt恢复成x0x_0x0是不可能的，我们只能一步一步的往前推，首先从xtx_txt恢复成xt−1x_{t-1}xt−1。根据贝叶斯公式，已知xtx_txt反推xt−1x_{t-1}xt−1：q(xt−1∣xt,x0)=q(xt∣xt−1,x0)q(xt−1∣x0)q(xt∣x0)q\left(x_{t-1} \mid x_t, x_0\right)=q\left(x_t \mid x_{t-1}, x_0\right) \frac{q\left(x_{t-1} \mid x_0\right)}{q\left(x_t \mid x_0\right)}q(xt−1∣xt,x0)=q(xt∣xt−1,x0)q(xt∣x0)q(xt−1∣x0)
右边的三个东西都可以从x_0开始推得到：
q(xt−1∣x0)=aˉt−1x0+1−aˉt−1z∼N(aˉt−1x0,1−aˉt−1)q\left(x_{t-1} \mid x_0\right)=\sqrt{\bar{a}_{t-1}} x_0+\sqrt{1-\bar{a}_{t-1}} z \sim \mathcal{N}\left(\sqrt{\bar{a}_{t-1}} x_0, 1-\bar{a}_{t-1}\right)q(xt−1∣x0)=aˉt−1x0+1−aˉt−1z∼N(aˉt−1x0,1−aˉt−1)q(xt∣x0)=aˉtx0+1−αˉtz∼N(aˉtx0,1−αˉt)q\left(x_t \mid x_0\right) = \sqrt{\bar{a}_t} x_0+\sqrt{1-\bar{\alpha}_t} z \sim \mathcal{N}\left(\sqrt{\bar{a}_t} x_0 , 1-\bar{\alpha}_t\right)q(xt∣x0)=aˉtx0+1−αˉtz∼N(aˉtx0,1−αˉt)q(xt∣xt−1,x0)=atxt−1+1−αtz∼N(atxt−1,1−αt)q\left(x_t \mid x_{t-1}, x_0\right)=\sqrt{a_t} x_{t-1}+\sqrt{1-\alpha_t} z \sim \mathcal{N}\left(\sqrt{a_t} x_{t-1}, 1-\alpha_t\right) \\q(xt∣xt−1,x0)=atxt−1+1−αtz∼N(atxt−1,1−αt)
因此，由于右边三个东西均满足正态分布，
q(xt−1∣xt,x0)q\left(x_{t-1} \mid x_t, x_0\right)q(xt−1∣xt,x0)满足分布如下：∝exp(−12((xt−αtxt−1)2βt+(xt−1−αˉt−1x0)21−αˉt−1−(xt−αˉtx0)21−αˉt))\propto \exp \left(-\frac{1}{2}\left(\frac{\left(x_t-\sqrt{\alpha_t} x_{t-1}\right)^2}{\beta_t}+\frac{\left(x_{t-1}-\sqrt{\bar{\alpha}_{t-1}} x_0\right)^2}{1-\bar{\alpha}_{t-1}}-\frac{\left(x_t-\sqrt{\bar{\alpha}_t} x_0\right)^2}{1-\bar{\alpha}_t}\right)\right)∝exp(−21(βt(xt−αtxt−1)2+1−αˉt−1(xt−1−αˉt−1x0)2−1−αˉt(xt−αˉtx0)2))
把标准正态分布展开后，乘法就相当于加，除法就相当于减，把他们汇总
接下来继续化简，咱们现在要求的是上一时刻的分布
∝exp(−12((xt−αtxt−1)2βt+(xt−1−αˉt−1x0)21−αˉt−1−(xt−αˉtx0)21−αˉt))=exp(−12(xt2−2αtxtxt−1+αtxt−12βt+xt−12−2αˉt−1x0xt−1+αˉt−1x021−αˉt−1−(xt−αˉtx0)21−αˉt))=exp(−12((αtβt+11−αˉt−1)xt−12−(2αtβtxt+2αˉt−11−αˉt−1x0)xt−1+C(xt,x0)))\begin{aligned} & \propto \exp \left(-\frac{1}{2}\left(\frac{\left(x_t-\sqrt{\alpha_t} x_{t-1}\right)^2}{\beta_t}+\frac{\left(x_{t-1}-\sqrt{\bar{\alpha}_{t-1}} x_0\right)^2}{1-\bar{\alpha}_{t-1}}-\frac{\left(x_t-\sqrt{\bar{\alpha}_t} x_0\right)^2}{1-\bar{\alpha}_t}\right)\right) \\ & =\exp \left(-\frac{1}{2}\left(\frac{x_t^2-2 \sqrt{\alpha_t} x_t x_{t-1}+\alpha_t x_{t-1}^2}{\beta_t}+\frac{x_{t-1}^2-2 \sqrt{\bar{\alpha}_{t-1}} x_0 x_{t-1}+\bar{\alpha}_{t-1} x_0^2}{1-\bar{\alpha}_{t-1}}-\frac{\left(x_t-\sqrt{\bar{\alpha}_t} x_0\right)^2}{1-\bar{\alpha}_t}\right)\right) \\ & =\exp \left(-\frac{1}{2}\left(\left(\frac{\alpha_t}{\beta_t}+\frac{1}{1-\bar{\alpha}_{t-1}}\right) x_{t-1}^2-\left(\frac{2 \sqrt{\alpha_t}}{\beta_t} x_t+\frac{2 \sqrt{\bar{\alpha}_{t-1}}}{1-\bar{\alpha}_{t-1}} x_0\right) x_{t-1}+C\left(x_t, x_0\right)\right)\right) \end{aligned}∝exp(−21(βt(xt−αtxt−1)2+1−αˉt−1(xt−1−αˉt−1x0)2−1−αˉt(xt−αˉtx0)2))=exp(−21(βtxt2−2αtxtxt−1+αtxt−12+1−αˉt−1xt−12−2αˉt−1x0xt−1+αˉt−1x02−1−αˉt(xt−αˉtx0)2))=exp(−21((βtαt+1−αˉt−11)xt−12−(βt2αtxt+1−αˉt−12αˉt−1x0)xt−1+C(xt,x0)))
正态分布满足公式，
exp(−(x−μ)22σ2)=exp(−12(1σ2x2−2μσ2x+μ2σ2))\exp \left(-\frac{(x-\mu)^2}{2 \sigma^2}\right)=\exp \left(-\frac{1}{2}\left(\frac{1}{\sigma^2} x^2-\frac{2 \mu}{\sigma^2} x+\frac{\mu^2}{\sigma^2}\right)\right)exp(−2σ2(x−μ)2)=exp(−21(σ21x2−σ22μx+σ2μ2))，其中σ\sigmaσ就是方差，μ\muμ就是均值，配方后我们就可以获得均值和方差。
此时的均值为：
μ~t(xt,x0)=αt(1−αˉt−1)1−αˉtxt+αˉt−1βt1−αˉtx0\tilde{\mu}_t\left(x_t, x_0\right)=\frac{\sqrt{\alpha_t}\left(1-\bar{\alpha}_{t-1}\right)}{1-\bar{\alpha}_t} x_t+\frac{\sqrt{\bar{\alpha}_{t-1}} \beta_t}{1-\bar{\alpha}_t} x_0μ~t(xt,x0)=1−αˉtαt(1−αˉt−1)xt+1−αˉtαˉt−1βtx0。根据之前的公式，xt=αt‾x0+1−αt‾ztx_t=\sqrt{\overline{\alpha_t}} x_0+\sqrt{1-\overline{\alpha_t}} z_txt=αtx0+1−αtzt，我们可以使用xtx_txt反向估计x0x_0x0得到x0x_0x0满足分布x0=1αˉt(xt−1−αˉtzt)x_0=\frac{1}{\sqrt{\bar{\alpha}_t}}\left(\mathrm{x}_t-\sqrt{1-\bar{\alpha}_t} z_t\right)x0=αˉt1(xt−1−αˉtzt)。最终得到均值为μ~t=1at(xt−βt1−aˉtzt)\tilde{\mu}_t=\frac{1}{\sqrt{a_t}}\left(x_t-\frac{\beta_t}{\sqrt{1-\bar{a}_t}} z_t\right)μ~t=at1(xt−1−aˉtβtzt) ，ztz_tzt代表t时刻的噪音是什么。由ztz_tzt无法直接获得，网络便通过当前时刻的xtx_txt经过神经网络计算ztz_tzt。ϵθ(xt,t)\epsilon_\theta\left(x_t, t\right)ϵθ(xt,t)也就是上面提到的ztz_tzt。ϵθ\epsilon_\thetaϵθ代表神经网络。xt−1=1αt(xt−1−αt1−αˉtϵθ(xt,t))+σtzx_{t-1}=\frac{1}{\sqrt{\alpha_t}}\left(x_t-\frac{1-\alpha_t}{\sqrt{1-\bar{\alpha}_t}} \epsilon_\theta\left(x_t, t\right)\right)+\sigma_t zxt−1=αt1(xt−1−αˉt1−αtϵθ(xt,t))+σtz
由于加噪过程中的真实噪声
ϵ\epsilonϵ在复原过程中是无法获得的，因此DDPM的关键就是训练一个由xtx_txt和ttt估测橾声的模型ϵθ(xt,t)\epsilon_\theta\left(x_t, t\right)ϵθ(xt,t)，其中θ\thetaθ就是模型的训练参数，σt\sigma_tσt 也是一个高斯噪声σt∼N(0,1)\sigma_t \sim N(0,1)σt∼N(0,1)，用于表示估测与实际的差距。在DDPM中，使用U-Net作为估测噪声的模型。
本质上，我们就是训练这个Unet模型，该模型输入为
xtx_txt和ttt，输出为xtx_txt时刻的**高斯噪声**。即利用xtx_txt和ttt预测这一时刻的高斯噪声。**这样就可以一步一步的再从噪声回到真实图像。**
二、DDPM网络的构建（Unet网络的构建）
上图是典型的Unet模型结构，仅仅作为示意图，里面具体的数字同学们无需在意，和本文的学习无关。在本文中，Unet的输入和输出shape相同，通道均为3（一般为RGB三通道），宽高相同。
本质上，DDPM最重要的工作就是训练Unet模型，该模型输入为
xtx_txt和ttt，输出为xt−1x_{t-1}xt−1时刻的**高斯噪声**。即利用xtx_txt和ttt预测上一时刻的高斯噪声。**这样就可以一步一步的再从噪声回到真实图像。**
假设我们需要生成一个[64, 64, 3]的图像，在
ttt时刻，我们有一个xtx_txt噪声图，该噪声图的的shape也为[64, 64, 3]，我们将它和ttt一起输入到Unet中。Unet的输出为xt−1x_{t-1}xt−1时刻的[64, 64, 3]的噪声。
实现代码如下，代码中的特征提取模块为残差结构，方便优化：
import mathimport torchimport torch.nn as nnimport torch.nn.functional as Fdefget_norm(norm, num_channels, num_groups):if norm =="in":return nn.InstanceNorm2d(num_channels, affine=True)elif norm =="bn":return nn.BatchNorm2d(num_channels)elif norm =="gn":return nn.GroupNorm(num_groups, num_channels)elif norm isNone:return nn.Identity()else:raise ValueError("unknown normalization type")#------------------------------------------## 计算时间步长的位置嵌入。# 一半为sin，一半为cos。#------------------------------------------#classPositionalEmbedding(nn.Module):def__init__(self, dim, scale=1.0):super().__init__()assert dim %2==0self.dim = dimself.scale = scaledefforward(self, x):device = x.devicehalf_dim = self.dim //2emb = math.log(10000)/ half_dimemb = torch.exp(torch.arange(half_dim, device=device)*-emb)# x * self.scale和emb外积emb = torch.outer(x * self.scale, emb)emb = torch.cat((emb.sin(), emb.cos()), dim=-1)return emb#------------------------------------------## 下采样层，一个步长为2x2的卷积#------------------------------------------#classDownsample(nn.Module):def__init__(self, in_channels):super().__init__()self.downsample = nn.Conv2d(in_channels, in_channels,3, stride=2, padding=1)defforward(self, x, time_emb, y):if x.shape[2]%2==1:raise ValueError("downsampling tensor height should be even")if x.shape[3]%2==1:raise ValueError("downsampling tensor width should be even")return self.downsample(x)#------------------------------------------## 上采样层，Upsample+卷积#------------------------------------------#classUpsample(nn.Module):def__init__(self, in_channels):super().__init__()self.upsample = nn.Sequential(nn.Upsample(scale_factor=2, mode="nearest"),nn.Conv2d(in_channels, in_channels,3, padding=1),)defforward(self, x, time_emb, y):return self.upsample(x)#------------------------------------------## 使用Self-Attention注意力机制# 做一个全局的Self-Attention#------------------------------------------#classAttentionBlock(nn.Module):def__init__(self, in_channels, norm="gn", num_groups=32):super().__init__()self.in_channels = in_channelsself.norm = get_norm(norm, in_channels, num_groups)self.to_qkv = nn.Conv2d(in_channels, in_channels *3,1)self.to_out = nn.Conv2d(in_channels, in_channels,1)defforward(self, x):b, c, h, w = x.shapeq, k, v = torch.split(self.to_qkv(self.norm(x)), self.in_channels, dim=1)q = q.permute(0,2,3,1).view(b, h * w, c)k = k.view(b, c, h * w)v = v.permute(0,2,3,1).view(b, h * w, c)dot_products = torch.bmm(q, k)*(c **(-0.5))assert dot_products.shape ==(b, h * w, h * w)attention = torch.softmax(dot_products, dim=-1)out = torch.bmm(attention, v)assert out.shape ==(b, h * w, c)out = out.view(b, h, w, c).permute(0,3,1,2)return self.to_out(out)+ x#------------------------------------------## 用于特征提取的残差结构#------------------------------------------#classResidualBlock(nn.Module):def__init__(self, in_channels, out_channels, dropout, time_emb_dim=None, num_classes=None, activation=F.relu,norm="gn", num_groups=32, use_attention=False,):super().__init__()self.activation = activationself.norm_1 = get_norm(norm, in_channels, num_groups)self.conv_1 = nn.Conv2d(in_channels, out_channels,3, padding=1)self.norm_2 = get_norm(norm, out_channels, num_groups)self.conv_2 = nn.Sequential(nn.Dropout(p=dropout),nn.Conv2d(out_channels, out_channels,3, padding=1),)self.time_bias = nn.Linear(time_emb_dim, out_channels)if time_emb_dim isnotNoneelseNoneself.class_bias = nn.Embedding(num_classes, out_channels)if num_classes isnotNoneelseNoneself.residual_connection = nn.Conv2d(in_channels, out_channels,1)if in_channels != out_channels else nn.Identity()self.attention = nn.Identity()ifnot use_attention else AttentionBlock(out_channels, norm, num_groups)defforward(self, x, time_emb=None, y=None):out = self.activation(self.norm_1(x))# 第一个卷积out = self.conv_1(out)# 对时间time_emb做一个全连接，施加在通道上if self.time_bias isnotNone:if time_emb isNone:raise ValueError("time conditioning was specified but time_emb is not passed")out += self.time_bias(self.activation(time_emb))[:,:,None,None]# 对种类y_emb做一个全连接，施加在通道上if self.class_bias isnotNone:if y isNone:raise ValueError("class conditioning was specified but y is not passed")out += self.class_bias(y)[:,:,None,None]out = self.activation(self.norm_2(out))# 第二个卷积+残差边out = self.conv_2(out)+ self.residual_connection(x)# 最后做个Attentionout = self.attention(out)return out#------------------------------------------## Unet模型#------------------------------------------#classUNet(nn.Module):def__init__(self, img_channels, base_channels=128, channel_mults=(1,2,2,2),num_res_blocks=2, time_emb_dim=128*4, time_emb_scale=1.0, num_classes=None, activation=F.silu,dropout=0.1, attention_resolutions=(1,), norm="gn", num_groups=32, initial_pad=0,):super().__init__()# 使用到的激活函数，一般为SILUself.activation = activation# 是否对输入进行paddingself.initial_pad = initial_pad# 需要去区分的类别数self.num_classes = num_classes# 对时间轴输入的全连接层self.time_mlp = nn.Sequential(PositionalEmbedding(base_channels, time_emb_scale),nn.Linear(base_channels, time_emb_dim),nn.SiLU(),nn.Linear(time_emb_dim, time_emb_dim),)if time_emb_dim isnotNoneelseNone# 对输入图片的第一个卷积self.init_conv = nn.Conv2d(img_channels, base_channels,3, padding=1)# self.downs用于存储下采样用到的层，首先利用ResidualBlock提取特征# 然后利用Downsample降低特征图的高宽self.downs = nn.ModuleList()self.ups = nn.ModuleList()# channels指的是每一个模块处理后的通道数# now_channels是一个中间变量，代表中间的通道数channels =[base_channels]now_channels = base_channelsfor i, mult inenumerate(channel_mults):out_channels = base_channels * multfor _ inrange(num_res_blocks):self.downs.append(ResidualBlock(now_channels, out_channels, dropout,time_emb_dim=time_emb_dim, num_classes=num_classes, activation=activation,norm=norm, num_groups=num_groups, use_attention=i in attention_resolutions,))now_channels = out_channelschannels.append(now_channels)if i !=len(channel_mults)-1:self.downs.append(Downsample(now_channels))channels.append(now_channels)# 可以看作是特征整合，中间的一个特征提取模块self.mid = nn.ModuleList([ResidualBlock(now_channels, now_channels, dropout,time_emb_dim=time_emb_dim, num_classes=num_classes, activation=activation,norm=norm, num_groups=num_groups, use_attention=True,),ResidualBlock(now_channels, now_channels, dropout,time_emb_dim=time_emb_dim, num_classes=num_classes, activation=activation,norm=norm, num_groups=num_groups, use_attention=False,),])# 进行上采样，进行特征融合for i, mult inreversed(list(enumerate(channel_mults))):out_channels = base_channels * multfor _ inrange(num_res_blocks +1):self.ups.append(ResidualBlock(channels.pop()+ now_channels, out_channels, dropout,time_emb_dim=time_emb_dim, num_classes=num_classes, activation=activation,norm=norm, num_groups=num_groups, use_attention=i in attention_resolutions,))now_channels = out_channelsif i !=0:self.ups.append(Upsample(now_channels))assertlen(channels)==0self.out_norm = get_norm(norm, base_channels, num_groups)self.out_conv = nn.Conv2d(base_channels, img_channels,3, padding=1)defforward(self, x, time=None, y=None):# 是否对输入进行paddingip = self.initial_padif ip !=0:x = F.pad(x,(ip,)*4)# 对时间轴输入的全连接层if self.time_mlp isnotNone:if time isNone:raise ValueError("time conditioning was specified but tim is not passed")time_emb = self.time_mlp(time)else:time_emb =Noneif self.num_classes isnotNoneand y isNone:raise ValueError("class conditioning was specified but y is not passed")# 对输入图片的第一个卷积x = self.init_conv(x)# skips用于存放下采样的中间层skips =[x]for layer in self.downs:x = layer(x, time_emb, y)skips.append(x)# 特征整合与提取for layer in self.mid:x = layer(x, time_emb, y)# 上采样并进行特征融合for layer in self.ups:ifisinstance(layer, ResidualBlock):x = torch.cat([x, skips.pop()], dim=1)x = layer(x, time_emb, y)# 上采样并进行特征融合x = self.activation(self.out_norm(x))x = self.out_conv(x)if self.initial_pad !=0:return x[:,:, ip:-ip, ip:-ip]else:return x
三、Diffusion的训练思路
Diffusion的训练思路比较简单，首先随机给每个batch里每张图片都生成一个t，代表我选择这个batch里面第t个时刻的噪声进行拟合。代码如下：
t = torch.randint(0, self.num_timesteps,(b,), device=device)
生成batch_size个噪声，计算施加这个噪声后模型在t个时刻的噪声图片是怎么样的，如下所示：
defperturb_x(self, x, t, noise):return(extract(self.sqrt_alphas_cumprod, t, x.shape)* x +extract(self.sqrt_one_minus_alphas_cumprod, t, x.shape)* noise)defget_losses(self, x, t, y):# x, noise [batch_size, 3, 64, 64]noise = torch.randn_like(x)perturbed_x = self.perturb_x(x, t, noise)
之后利用这个噪声图片、t和网络模型计算预测噪声，利用预测噪声和实际噪声进行拟合。
defget_losses(self, x, t, y):# x, noise [batch_size, 3, 64, 64]noise = torch.randn_like(x)perturbed_x = self.perturb_x(x, t, noise)estimated_noise = self.model(perturbed_x, t, y)if self.loss_type =="l1":loss = F.l1_loss(estimated_noise, noise)elif self.loss_type =="l2":loss = F.mse_loss(estimated_noise, noise)return loss
利用DDPM生成图片
DDPM的库整体结构如下：
一、数据集的准备
在训练前需要准备好数据集，数据集保存在datasets文件夹里面。
二、数据集的处理
打开txt_annotation.py，默认指向根目录下的datasets。运行txt_annotation.py。
此时生成根目录下面的train_lines.txt。
三、模型训练
在完成数据集处理后，运行train.py即可开始训练。
训练过程中，可在results文件夹内查看训练效果：
本文转载自: https://blog.csdn.net/weixin_44791964/article/details/128604816
版权归原作者 Bubbliiiing 所有， 如有侵权，请联系我们删除。
版权归原作者 Bubbliiiing 所有， 如有侵权，请联系我们删除。