建设网站需要收费吗,django做的网站有哪些,淄博专业做网站,河南南阳油田网站建设第一步#xff1a;准备数据
人像精细分割数据#xff0c;可分割出头发丝#xff0c;为PPM-100开源数据 第二步#xff1a;搭建模型
MODNet网络结构如图所示#xff0c;主要包含3个部分#xff1a;semantic estimation#xff08;S分支#xff09;、detail prediction… 第一步准备数据
人像精细分割数据可分割出头发丝为PPM-100开源数据 第二步搭建模型
MODNet网络结构如图所示主要包含3个部分semantic estimationS分支、detail predictionD分支、semantic-detail fusionF分支。 网络结构简单描述一下
输入一幅图像I送入三个模块S、D、F S模块在低分辨率分支进行语义估计在backbone最后一层输出接上e-ASPP得到语义feature map Sp D模块在高分辨率分支进行细节预测通过融合来自低分辨率分支的信息得到细节feature map Dp F模块融合来自低分辨率分支和高分辨率分支的信息得到alpha matte ap 对S、D、F模块均使用来自GT的显式监督信息进行监督训练。 第三步代码
1损失函数为L2损失
2网络代码
import torch
import torch.nn as nn
import torch.nn.functional as Ffrom .backbones import SUPPORTED_BACKBONES#------------------------------------------------------------------------------
# MODNet Basic Modules
#------------------------------------------------------------------------------class IBNorm(nn.Module): Combine Instance Norm and Batch Norm into One Layerdef __init__(self, in_channels):super(IBNorm, self).__init__()in_channels in_channelsself.bnorm_channels int(in_channels / 2)self.inorm_channels in_channels - self.bnorm_channelsself.bnorm nn.BatchNorm2d(self.bnorm_channels, affineTrue)self.inorm nn.InstanceNorm2d(self.inorm_channels, affineFalse)def forward(self, x):bn_x self.bnorm(x[:, :self.bnorm_channels, ...].contiguous())in_x self.inorm(x[:, self.bnorm_channels:, ...].contiguous())return torch.cat((bn_x, in_x), 1)class Conv2dIBNormRelu(nn.Module): Convolution IBNorm ReLudef __init__(self, in_channels, out_channels, kernel_size, stride1, padding0, dilation1, groups1, biasTrue, with_ibnTrue, with_reluTrue):super(Conv2dIBNormRelu, self).__init__()layers [nn.Conv2d(in_channels, out_channels, kernel_size, stridestride, paddingpadding, dilationdilation, groupsgroups, biasbias)]if with_ibn: layers.append(IBNorm(out_channels))if with_relu:layers.append(nn.ReLU(inplaceTrue))self.layers nn.Sequential(*layers)def forward(self, x):return self.layers(x) class SEBlock(nn.Module): SE Block Proposed in https://arxiv.org/pdf/1709.01507.pdf def __init__(self, in_channels, out_channels, reduction1):super(SEBlock, self).__init__()self.pool nn.AdaptiveAvgPool2d(1)self.fc nn.Sequential(nn.Linear(in_channels, int(in_channels // reduction), biasFalse),nn.ReLU(inplaceTrue),nn.Linear(int(in_channels // reduction), out_channels, biasFalse),nn.Sigmoid())def forward(self, x):b, c, _, _ x.size()w self.pool(x).view(b, c)w self.fc(w).view(b, c, 1, 1)return x * w.expand_as(x)#------------------------------------------------------------------------------
# MODNet Branches
#------------------------------------------------------------------------------class LRBranch(nn.Module): Low Resolution Branch of MODNetdef __init__(self, backbone):super(LRBranch, self).__init__()enc_channels backbone.enc_channelsself.backbone backboneself.se_block SEBlock(enc_channels[4], enc_channels[4], reduction4)self.conv_lr16x Conv2dIBNormRelu(enc_channels[4], enc_channels[3], 5, stride1, padding2)self.conv_lr8x Conv2dIBNormRelu(enc_channels[3], enc_channels[2], 5, stride1, padding2)self.conv_lr Conv2dIBNormRelu(enc_channels[2], 1, kernel_size3, stride2, padding1, with_ibnFalse, with_reluFalse)def forward(self, img, inference):enc_features self.backbone.forward(img)enc2x, enc4x, enc32x enc_features[0], enc_features[1], enc_features[4]enc32x self.se_block(enc32x)lr16x F.interpolate(enc32x, scale_factor2, modebilinear, align_cornersFalse)lr16x self.conv_lr16x(lr16x)lr8x F.interpolate(lr16x, scale_factor2, modebilinear, align_cornersFalse)lr8x self.conv_lr8x(lr8x)pred_semantic Noneif not inference:lr self.conv_lr(lr8x)pred_semantic torch.sigmoid(lr)return pred_semantic, lr8x, [enc2x, enc4x] class HRBranch(nn.Module): High Resolution Branch of MODNetdef __init__(self, hr_channels, enc_channels):super(HRBranch, self).__init__()self.tohr_enc2x Conv2dIBNormRelu(enc_channels[0], hr_channels, 1, stride1, padding0)self.conv_enc2x Conv2dIBNormRelu(hr_channels 3, hr_channels, 3, stride2, padding1)self.tohr_enc4x Conv2dIBNormRelu(enc_channels[1], hr_channels, 1, stride1, padding0)self.conv_enc4x Conv2dIBNormRelu(2 * hr_channels, 2 * hr_channels, 3, stride1, padding1)self.conv_hr4x nn.Sequential(Conv2dIBNormRelu(3 * hr_channels 3, 2 * hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(2 * hr_channels, 2 * hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(2 * hr_channels, hr_channels, 3, stride1, padding1),)self.conv_hr2x nn.Sequential(Conv2dIBNormRelu(2 * hr_channels, 2 * hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(2 * hr_channels, hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(hr_channels, hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(hr_channels, hr_channels, 3, stride1, padding1),)self.conv_hr nn.Sequential(Conv2dIBNormRelu(hr_channels 3, hr_channels, 3, stride1, padding1),Conv2dIBNormRelu(hr_channels, 1, kernel_size1, stride1, padding0, with_ibnFalse, with_reluFalse),)def forward(self, img, enc2x, enc4x, lr8x, inference):img2x F.interpolate(img, scale_factor1/2, modebilinear, align_cornersFalse)img4x F.interpolate(img, scale_factor1/4, modebilinear, align_cornersFalse)enc2x self.tohr_enc2x(enc2x)hr4x self.conv_enc2x(torch.cat((img2x, enc2x), dim1))enc4x self.tohr_enc4x(enc4x)hr4x self.conv_enc4x(torch.cat((hr4x, enc4x), dim1))lr4x F.interpolate(lr8x, scale_factor2, modebilinear, align_cornersFalse)hr4x self.conv_hr4x(torch.cat((hr4x, lr4x, img4x), dim1))hr2x F.interpolate(hr4x, scale_factor2, modebilinear, align_cornersFalse)hr2x self.conv_hr2x(torch.cat((hr2x, enc2x), dim1))pred_detail Noneif not inference:hr F.interpolate(hr2x, scale_factor2, modebilinear, align_cornersFalse)hr self.conv_hr(torch.cat((hr, img), dim1))pred_detail torch.sigmoid(hr)return pred_detail, hr2xclass FusionBranch(nn.Module): Fusion Branch of MODNetdef __init__(self, hr_channels, enc_channels):super(FusionBranch, self).__init__()self.conv_lr4x Conv2dIBNormRelu(enc_channels[2], hr_channels, 5, stride1, padding2)self.conv_f2x Conv2dIBNormRelu(2 * hr_channels, hr_channels, 3, stride1, padding1)self.conv_f nn.Sequential(Conv2dIBNormRelu(hr_channels 3, int(hr_channels / 2), 3, stride1, padding1),Conv2dIBNormRelu(int(hr_channels / 2), 1, 1, stride1, padding0, with_ibnFalse, with_reluFalse),)def forward(self, img, lr8x, hr2x):lr4x F.interpolate(lr8x, scale_factor2, modebilinear, align_cornersFalse)lr4x self.conv_lr4x(lr4x)lr2x F.interpolate(lr4x, scale_factor2, modebilinear, align_cornersFalse)f2x self.conv_f2x(torch.cat((lr2x, hr2x), dim1))f F.interpolate(f2x, scale_factor2, modebilinear, align_cornersFalse)f self.conv_f(torch.cat((f, img), dim1))pred_matte torch.sigmoid(f)return pred_matte#------------------------------------------------------------------------------
# MODNet
#------------------------------------------------------------------------------class MODNet(nn.Module): Architecture of MODNetdef __init__(self, in_channels3, hr_channels32, backbone_archmobilenetv2, backbone_pretrainedTrue):super(MODNet, self).__init__()self.in_channels in_channelsself.hr_channels hr_channelsself.backbone_arch backbone_archself.backbone_pretrained backbone_pretrainedself.backbone SUPPORTED_BACKBONES[self.backbone_arch](self.in_channels)self.lr_branch LRBranch(self.backbone)self.hr_branch HRBranch(self.hr_channels, self.backbone.enc_channels)self.f_branch FusionBranch(self.hr_channels, self.backbone.enc_channels)for m in self.modules():if isinstance(m, nn.Conv2d):self._init_conv(m)elif isinstance(m, nn.BatchNorm2d) or isinstance(m, nn.InstanceNorm2d):self._init_norm(m)if self.backbone_pretrained:self.backbone.load_pretrained_ckpt() def forward(self, img, inference):pred_semantic, lr8x, [enc2x, enc4x] self.lr_branch(img, inference)pred_detail, hr2x self.hr_branch(img, enc2x, enc4x, lr8x, inference)pred_matte self.f_branch(img, lr8x, hr2x)return pred_semantic, pred_detail, pred_mattedef freeze_norm(self):norm_types [nn.BatchNorm2d, nn.InstanceNorm2d]for m in self.modules():for n in norm_types:if isinstance(m, n):m.eval()continuedef _init_conv(self, conv):nn.init.kaiming_uniform_(conv.weight, a0, modefan_in, nonlinearityrelu)if conv.bias is not None:nn.init.constant_(conv.bias, 0)def _init_norm(self, norm):if norm.weight is not None:nn.init.constant_(norm.weight, 1)nn.init.constant_(norm.bias, 0)第四步搭建GUI界面 第五步整个工程的内容
有训练代码和训练好的模型以及训练过程提供数据提供GUI界面代码
代码见基于Pytorch框架的深度学习MODNet网络精细人像分割系统源码 有问题可以私信或者留言有问必答
