networks.py

# coding=utf-8
import torch
import torch.nn as nn
from torch.nn import init
from torchvision import models
import os

import numpy as np


def weights_init_normal(m):
    classname = m.__class__.__name__
    if classname.find('Conv') != -1:
        init.normal_(m.weight.data, 0.0, 0.02)
    elif classname.find('Linear') != -1:
        init.normal(m.weight.data, 0.0, 0.02)
    elif classname.find('BatchNorm2d') != -1:
        init.normal_(m.weight.data, 1.0, 0.02)
        init.constant_(m.bias.data, 0.0)


def weights_init_xavier(m):
    classname = m.__class__.__name__
    if classname.find('Conv') != -1:
        init.xavier_normal_(m.weight.data, gain=0.02)
    elif classname.find('Linear') != -1:
        init.xavier_normal_(m.weight.data, gain=0.02)
    elif classname.find('BatchNorm2d') != -1:
        init.normal_(m.weight.data, 1.0, 0.02)
        init.constant_(m.bias.data, 0.0)


def weights_init_kaiming(m):
    classname = m.__class__.__name__
    if classname.find('Conv') != -1:
        init.kaiming_normal_(m.weight.data, a=0, mode='fan_in')
    elif classname.find('Linear') != -1:
        init.kaiming_normal_(m.weight.data, a=0, mode='fan_in')
    elif classname.find('BatchNorm2d') != -1:
        init.normal_(m.weight.data, 1.0, 0.02)
        init.constant_(m.bias.data, 0.0)


def init_weights(net, init_type='normal'):
    print('initialization method [%s]' % init_type)
    if init_type == 'normal':
        net.apply(weights_init_normal)
    elif init_type == 'xavier':
        net.apply(weights_init_xavier)
    elif init_type == 'kaiming':
        net.apply(weights_init_kaiming)
    else:
        raise NotImplementedError(
            'initialization method [%s] is not implemented' % init_type)


class FeatureExtraction(nn.Module):
    def __init__(self, input_nc, ngf=64, n_layers=3, norm_layer=nn.BatchNorm2d, use_dropout=False):
        super(FeatureExtraction, self).__init__()
        downconv = nn.Conv2d(input_nc, ngf, kernel_size=4, stride=2, padding=1)
        model = [downconv, nn.ReLU(True), norm_layer(ngf)]
        for i in range(n_layers):
            in_ngf = 2**i * ngf if 2**i * ngf < 512 else 512
            out_ngf = 2**(i+1) * ngf if 2**i * ngf < 512 else 512
            downconv = nn.Conv2d(
                in_ngf, out_ngf, kernel_size=4, stride=2, padding=1)
            model += [downconv, nn.ReLU(True)]
            model += [norm_layer(out_ngf)]
        model += [nn.Conv2d(512, 512, kernel_size=3,
                            stride=1, padding=1), nn.ReLU(True)]
        model += [norm_layer(512)]
        model += [nn.Conv2d(512, 512, kernel_size=3,
                            stride=1, padding=1), nn.ReLU(True)]

        self.model = nn.Sequential(*model)
        init_weights(self.model, init_type='normal')

    def forward(self, x):
        return self.model(x)


class FeatureL2Norm(torch.nn.Module):
    def __init__(self):
        super(FeatureL2Norm, self).__init__()

    def forward(self, feature):
        epsilon = 1e-6
        norm = torch.pow(torch.sum(torch.pow(feature, 2), 1) +
                         epsilon, 0.5).unsqueeze(1).expand_as(feature)
        return torch.div(feature, norm)


class FeatureCorrelation(nn.Module):
    def __init__(self):
        super(FeatureCorrelation, self).__init__()

    def forward(self, feature_A, feature_B):
        b, c, h, w = feature_A.size()
        # reshape features for matrix multiplication
        feature_A = feature_A.transpose(2, 3).contiguous().view(b, c, h*w)
        feature_B = feature_B.view(b, c, h*w).transpose(1, 2)
        # perform matrix mult.
        feature_mul = torch.bmm(feature_B, feature_A)
        correlation_tensor = feature_mul.view(
            b, h, w, h*w).transpose(2, 3).transpose(1, 2)
        return correlation_tensor


class FeatureRegression(nn.Module):
    def __init__(self, input_nc=512, output_dim=6, use_cuda=True):
        super(FeatureRegression, self).__init__()
        self.conv = nn.Sequential(
            nn.Conv2d(input_nc, 512, kernel_size=4, stride=2, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 256, kernel_size=4, stride=2, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
        )
        self.linear = nn.Linear(64 * 4 * 3, output_dim)
        self.tanh = nn.Tanh()
        if use_cuda:
            self.conv.cuda()
            self.linear.cuda()
            self.tanh.cuda()

    def forward(self, x):
        x = self.conv(x)
        x = x.view(x.size(0), -1)
        x = self.linear(x)
        x = self.tanh(x)
        return x


class AffineGridGen(nn.Module):
    def __init__(self, out_h=256, out_w=192, out_ch=3):
        super(AffineGridGen, self).__init__()
        self.out_h = out_h
        self.out_w = out_w
        self.out_ch = out_ch

    def forward(self, theta):
        theta = theta.contiguous()
        batch_size = theta.size()[0]
        out_size = torch.Size(
            (batch_size, self.out_ch, self.out_h, self.out_w))
        return F.affine_grid(theta, out_size)


class TpsGridGen(nn.Module):
    def __init__(self, out_h=256, out_w=192, use_regular_grid=True, grid_size=3, reg_factor=0, use_cuda=True):
        super(TpsGridGen, self).__init__()
        self.out_h, self.out_w = out_h, out_w
        self.reg_factor = reg_factor
        self.use_cuda = use_cuda

        # create grid in numpy
        self.grid = np.zeros([self.out_h, self.out_w, 3], dtype=np.float32)
        # sampling grid with dim-0 coords (Y)
        self.grid_X, self.grid_Y = np.meshgrid(
            np.linspace(-1, 1, out_w), np.linspace(-1, 1, out_h))
        # grid_X,grid_Y: size [1,H,W,1,1]
        self.grid_X = torch.FloatTensor(self.grid_X).unsqueeze(0).unsqueeze(3)
        self.grid_Y = torch.FloatTensor(self.grid_Y).unsqueeze(0).unsqueeze(3)
        if use_cuda:
            self.grid_X = self.grid_X.cuda()
            self.grid_Y = self.grid_Y.cuda()

        # initialize regular grid for control points P_i
        if use_regular_grid:
            axis_coords = np.linspace(-1, 1, grid_size)
            self.N = grid_size*grid_size
            P_Y, P_X = np.meshgrid(axis_coords, axis_coords)
            P_X = np.reshape(P_X, (-1, 1))  # size (N,1)
            P_Y = np.reshape(P_Y, (-1, 1))  # size (N,1)
            P_X = torch.FloatTensor(P_X)
            P_Y = torch.FloatTensor(P_Y)
            self.P_X_base = P_X.clone()
            self.P_Y_base = P_Y.clone()
            self.Li = self.compute_L_inverse(P_X, P_Y).unsqueeze(0)
            self.P_X = P_X.unsqueeze(2).unsqueeze(
                3).unsqueeze(4).transpose(0, 4)
            self.P_Y = P_Y.unsqueeze(2).unsqueeze(
                3).unsqueeze(4).transpose(0, 4)
            if use_cuda:
                self.P_X = self.P_X.cuda()
                self.P_Y = self.P_Y.cuda()
                self.P_X_base = self.P_X_base.cuda()
                self.P_Y_base = self.P_Y_base.cuda()

    def forward(self, theta):
        warped_grid = self.apply_transformation(
            theta, torch.cat((self.grid_X, self.grid_Y), 3))

        return warped_grid

    def compute_L_inverse(self, X, Y):
        N = X.size()[0]  # num of points (along dim 0)
        # construct matrix K
        Xmat = X.expand(N, N)
        Ymat = Y.expand(N, N)
        P_dist_squared = torch.pow(
            Xmat-Xmat.transpose(0, 1), 2)+torch.pow(Ymat-Ymat.transpose(0, 1), 2)
        # make diagonal 1 to avoid NaN in log computation
        P_dist_squared[P_dist_squared == 0] = 1
        K = torch.mul(P_dist_squared, torch.log(P_dist_squared))
        # construct matrix L
        O = torch.FloatTensor(N, 1).fill_(1)
        Z = torch.FloatTensor(3, 3).fill_(0)
        P = torch.cat((O, X, Y), 1)
        L = torch.cat((torch.cat((K, P), 1), torch.cat(
            (P.transpose(0, 1), Z), 1)), 0)
        Li = torch.inverse(L)
        if self.use_cuda:
            Li = Li.cuda()
        return Li

    def apply_transformation(self, theta, points):
        if theta.dim() == 2:
            theta = theta.unsqueeze(2).unsqueeze(3)
        # points should be in the [B,H,W,2] format,
        # where points[:,:,:,0] are the X coords
        # and points[:,:,:,1] are the Y coords

        # input are the corresponding control points P_i
        batch_size = theta.size()[0]
        # split theta into point coordinates
        Q_X = theta[:, :self.N, :, :].squeeze(3)
        Q_Y = theta[:, self.N:, :, :].squeeze(3)
        Q_X = Q_X + self.P_X_base.expand_as(Q_X)
        Q_Y = Q_Y + self.P_Y_base.expand_as(Q_Y)

        # get spatial dimensions of points
        points_b = points.size()[0]
        points_h = points.size()[1]
        points_w = points.size()[2]

        # repeat pre-defined control points along spatial dimensions of points to be transformed
        P_X = self.P_X.expand((1, points_h, points_w, 1, self.N))
        P_Y = self.P_Y.expand((1, points_h, points_w, 1, self.N))

        # compute weigths for non-linear part
        W_X = torch.bmm(self.Li[:, :self.N, :self.N].expand(
            (batch_size, self.N, self.N)), Q_X)
        W_Y = torch.bmm(self.Li[:, :self.N, :self.N].expand(
            (batch_size, self.N, self.N)), Q_Y)
        # reshape
        # W_X,W,Y: size [B,H,W,1,N]
        W_X = W_X.unsqueeze(3).unsqueeze(4).transpose(
            1, 4).repeat(1, points_h, points_w, 1, 1)
        W_Y = W_Y.unsqueeze(3).unsqueeze(4).transpose(
            1, 4).repeat(1, points_h, points_w, 1, 1)
        # compute weights for affine part
        A_X = torch.bmm(self.Li[:, self.N:, :self.N].expand(
            (batch_size, 3, self.N)), Q_X)
        A_Y = torch.bmm(self.Li[:, self.N:, :self.N].expand(
            (batch_size, 3, self.N)), Q_Y)
        # reshape
        # A_X,A,Y: size [B,H,W,1,3]
        A_X = A_X.unsqueeze(3).unsqueeze(4).transpose(
            1, 4).repeat(1, points_h, points_w, 1, 1)
        A_Y = A_Y.unsqueeze(3).unsqueeze(4).transpose(
            1, 4).repeat(1, points_h, points_w, 1, 1)

        # compute distance P_i - (grid_X,grid_Y)
        # grid is expanded in point dim 4, but not in batch dim 0, as points P_X,P_Y are fixed for all batch
        points_X_for_summation = points[:, :, :, 0].unsqueeze(
            3).unsqueeze(4).expand(points[:, :, :, 0].size()+(1, self.N))
        points_Y_for_summation = points[:, :, :, 1].unsqueeze(
            3).unsqueeze(4).expand(points[:, :, :, 1].size()+(1, self.N))

        if points_b == 1:
            delta_X = points_X_for_summation-P_X
            delta_Y = points_Y_for_summation-P_Y
        else:
            # use expanded P_X,P_Y in batch dimension
            delta_X = points_X_for_summation - \
                P_X.expand_as(points_X_for_summation)
            delta_Y = points_Y_for_summation - \
                P_Y.expand_as(points_Y_for_summation)

        dist_squared = torch.pow(delta_X, 2)+torch.pow(delta_Y, 2)
        # U: size [1,H,W,1,N]
        dist_squared[dist_squared == 0] = 1  # avoid NaN in log computation
        U = torch.mul(dist_squared, torch.log(dist_squared))

        # expand grid in batch dimension if necessary
        points_X_batch = points[:, :, :, 0].unsqueeze(3)
        points_Y_batch = points[:, :, :, 1].unsqueeze(3)
        if points_b == 1:
            points_X_batch = points_X_batch.expand(
                (batch_size,)+points_X_batch.size()[1:])
            points_Y_batch = points_Y_batch.expand(
                (batch_size,)+points_Y_batch.size()[1:])

        points_X_prime = A_X[:, :, :, :, 0] + \
            torch.mul(A_X[:, :, :, :, 1], points_X_batch) + \
            torch.mul(A_X[:, :, :, :, 2], points_Y_batch) + \
            torch.sum(torch.mul(W_X, U.expand_as(W_X)), 4)

        points_Y_prime = A_Y[:, :, :, :, 0] + \
            torch.mul(A_Y[:, :, :, :, 1], points_X_batch) + \
            torch.mul(A_Y[:, :, :, :, 2], points_Y_batch) + \
            torch.sum(torch.mul(W_Y, U.expand_as(W_Y)), 4)

        return torch.cat((points_X_prime, points_Y_prime), 3)

# Defines the Unet generator.
# |num_downs|: number of downsamplings in UNet. For example,
# if |num_downs| == 7, image of size 128x128 will become of size 1x1
# at the bottleneck


class UnetGenerator(nn.Module):
    def __init__(self, input_nc, output_nc, num_downs, ngf=64,
                 norm_layer=nn.BatchNorm2d, use_dropout=False):
        super(UnetGenerator, self).__init__()
        # construct unet structure
        unet_block = UnetSkipConnectionBlock(
            ngf * 8, ngf * 8, input_nc=None, submodule=None, norm_layer=norm_layer, innermost=True)
        for i in range(num_downs - 5):
            unet_block = UnetSkipConnectionBlock(
                ngf * 8, ngf * 8, input_nc=None, submodule=unet_block, norm_layer=norm_layer, use_dropout=use_dropout)
        unet_block = UnetSkipConnectionBlock(
            ngf * 4, ngf * 8, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
        unet_block = UnetSkipConnectionBlock(
            ngf * 2, ngf * 4, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
        unet_block = UnetSkipConnectionBlock(
            ngf, ngf * 2, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
        unet_block = UnetSkipConnectionBlock(
            output_nc, ngf, input_nc=input_nc, submodule=unet_block, outermost=True, norm_layer=norm_layer)

        self.model = unet_block

    def forward(self, input):
        return self.model(input)


# Defines the submodule with skip connection.
# X -------------------identity---------------------- X
#   |-- downsampling -- |submodule| -- upsampling --|
class UnetSkipConnectionBlock(nn.Module):
    def __init__(self, outer_nc, inner_nc, input_nc=None,
                 submodule=None, outermost=False, innermost=False, norm_layer=nn.BatchNorm2d, use_dropout=False):
        super(UnetSkipConnectionBlock, self).__init__()
        self.outermost = outermost
        use_bias = norm_layer == nn.InstanceNorm2d

        if input_nc is None:
            input_nc = outer_nc
        downconv = nn.Conv2d(input_nc, inner_nc, kernel_size=4,
                             stride=2, padding=1, bias=use_bias)
        downrelu = nn.LeakyReLU(0.2, True)
        downnorm = norm_layer(inner_nc)
        uprelu = nn.ReLU(True)
        upnorm = norm_layer(outer_nc)

        if outermost:
            upsample = nn.Upsample(scale_factor=2, mode='bilinear')
            upconv = nn.Conv2d(inner_nc * 2, outer_nc,
                               kernel_size=3, stride=1, padding=1, bias=use_bias)
            down = [downconv]
            up = [uprelu, upsample, upconv, upnorm]
            model = down + [submodule] + up
        elif innermost:
            upsample = nn.Upsample(scale_factor=2, mode='bilinear')
            upconv = nn.Conv2d(inner_nc, outer_nc, kernel_size=3,
                               stride=1, padding=1, bias=use_bias)
            down = [downrelu, downconv]
            up = [uprelu, upsample, upconv, upnorm]
            model = down + up
        else:
            upsample = nn.Upsample(scale_factor=2, mode='bilinear')
            upconv = nn.Conv2d(inner_nc*2, outer_nc, kernel_size=3,
                               stride=1, padding=1, bias=use_bias)
            down = [downrelu, downconv, downnorm]
            up = [uprelu, upsample, upconv, upnorm]

            if use_dropout:
                model = down + [submodule] + up + [nn.Dropout(0.5)]
            else:
                model = down + [submodule] + up

        self.model = nn.Sequential(*model)

    def forward(self, x):
        if self.outermost:
            return self.model(x)
        else:
            return torch.cat([x, self.model(x)], 1)


class Vgg19(nn.Module):
    def __init__(self, requires_grad=False):
        super(Vgg19, self).__init__()
        vgg_pretrained_features = models.vgg19(pretrained=True).features
        self.slice1 = torch.nn.Sequential()
        self.slice2 = torch.nn.Sequential()
        self.slice3 = torch.nn.Sequential()
        self.slice4 = torch.nn.Sequential()
        self.slice5 = torch.nn.Sequential()
        for x in range(2):
            self.slice1.add_module(str(x), vgg_pretrained_features[x])
        for x in range(2, 7):
            self.slice2.add_module(str(x), vgg_pretrained_features[x])
        for x in range(7, 12):
            self.slice3.add_module(str(x), vgg_pretrained_features[x])
        for x in range(12, 21):
            self.slice4.add_module(str(x), vgg_pretrained_features[x])
        for x in range(21, 30):
            self.slice5.add_module(str(x), vgg_pretrained_features[x])
        if not requires_grad:
            for param in self.parameters():
                param.requires_grad = False

    def forward(self, X):
        h_relu1 = self.slice1(X)
        h_relu2 = self.slice2(h_relu1)
        h_relu3 = self.slice3(h_relu2)
        h_relu4 = self.slice4(h_relu3)
        h_relu5 = self.slice5(h_relu4)
        out = [h_relu1, h_relu2, h_relu3, h_relu4, h_relu5]
        return out


class VGGLoss(nn.Module):
    def __init__(self, layids=None):
        super(VGGLoss, self).__init__()
        self.vgg = Vgg19()
        self.vgg.cuda()
        self.criterion = nn.L1Loss()
        self.weights = [1.0/32, 1.0/16, 1.0/8, 1.0/4, 1.0]
        self.layids = layids

    def forward(self, x, y):
        x_vgg, y_vgg = self.vgg(x), self.vgg(y)
        loss = 0
        if self.layids is None:
            self.layids = list(range(len(x_vgg)))
        for i in self.layids:
            loss += self.weights[i] * \
                self.criterion(x_vgg[i], y_vgg[i].detach())
        return loss


class DT(nn.Module):
    def __init__(self):
        super(DT, self).__init__()

    def forward(self, x1, x2):
        dt = torch.abs(x1 - x2)
        return dt


class DT2(nn.Module):
    def __init__(self):
        super(DT, self).__init__()

    def forward(self, x1, y1, x2, y2):
        dt = torch.sqrt(torch.mul(x1 - x2, x1 - x2) +
                        torch.mul(y1 - y2, y1 - y2))
        return dt


class GicLoss(nn.Module):
    def __init__(self, opt):
        super(GicLoss, self).__init__()
        self.dT = DT()
        self.opt = opt

    def forward(self, grid):
        Gx = grid[:, :, :, 0]
        Gy = grid[:, :, :, 1]
        Gxcenter = Gx[:, 1:self.opt.fine_height - 1, 1:self.opt.fine_width - 1]
        Gxup = Gx[:, 0:self.opt.fine_height - 2, 1:self.opt.fine_width - 1]
        Gxdown = Gx[:, 2:self.opt.fine_height, 1:self.opt.fine_width - 1]
        Gxleft = Gx[:, 1:self.opt.fine_height - 1, 0:self.opt.fine_width - 2]
        Gxright = Gx[:, 1:self.opt.fine_height - 1, 2:self.opt.fine_width]

        Gycenter = Gy[:, 1:self.opt.fine_height - 1, 1:self.opt.fine_width - 1]
        Gyup = Gy[:, 0:self.opt.fine_height - 2, 1:self.opt.fine_width - 1]
        Gydown = Gy[:, 2:self.opt.fine_height, 1:self.opt.fine_width - 1]
        Gyleft = Gy[:, 1:self.opt.fine_height - 1, 0:self.opt.fine_width - 2]
        Gyright = Gy[:, 1:self.opt.fine_height - 1, 2:self.opt.fine_width]

        dtleft = self.dT(Gxleft, Gxcenter)
        dtright = self.dT(Gxright, Gxcenter)
        dtup = self.dT(Gyup, Gycenter)
        dtdown = self.dT(Gydown, Gycenter)

        return torch.sum(torch.abs(dtleft - dtright) + torch.abs(dtup - dtdown))


class GMM(nn.Module):
    """ Geometric Matching Module
    """

    def __init__(self, opt):
        super(GMM, self).__init__()
        self.extractionA = FeatureExtraction(
            22, ngf=64, n_layers=3, norm_layer=nn.BatchNorm2d)
        self.extractionB = FeatureExtraction(
            1, ngf=64, n_layers=3, norm_layer=nn.BatchNorm2d)
        self.l2norm = FeatureL2Norm()
        self.correlation = FeatureCorrelation()
        self.regression = FeatureRegression(
            input_nc=192, output_dim=2*opt.grid_size**2, use_cuda=True)
        self.gridGen = TpsGridGen(
            opt.fine_height, opt.fine_width, use_cuda=True, grid_size=opt.grid_size)

    def forward(self, inputA, inputB):
        featureA = self.extractionA(inputA)
        featureB = self.extractionB(inputB)
        featureA = self.l2norm(featureA)
        featureB = self.l2norm(featureB)
        correlation = self.correlation(featureA, featureB)

        theta = self.regression(correlation)
        grid = self.gridGen(theta)
        return grid, theta


def save_checkpoint(model, save_path):
    if not os.path.exists(os.path.dirname(save_path)):
        os.makedirs(os.path.dirname(save_path))

    torch.save(model.cpu().state_dict(), save_path)
    model.cuda()


def load_checkpoint(model, checkpoint_path):
    if not os.path.exists(checkpoint_path):
        return
    model.load_state_dict(torch.load(checkpoint_path))
    model.cuda()