Vit-V-Net pytorch 代码理解与分析

论文题目：ViT-V-Net: Vision Transformer for Unsupervised Volumetric Medical Image Registration

源码链接：https://github.com/junyuchen245/ViT-V-Net_for_3D_Image_Registration_Pytorch

图1 整体框架

图2 block

图4 代码调用关系

config.py:

配置初始的参数

import ml_collections

def get_3DReg_config():
    config = ml_collections.ConfigDict()
    config.patches = ml_collections.ConfigDict({'size': (8, 8, 8)})
    config.patches.grid = (8, 8, 8)
    config.hidden_size = 252
    config.transformer = ml_collections.ConfigDict()
    config.transformer.mlp_dim = 3072
    config.transformer.num_heads = 12
    config.transformer.num_layers = 12
    config.transformer.attention_dropout_rate = 0.0
    config.transformer.dropout_rate = 0.1
    config.patch_size = 8

    config.conv_first_channel = 512
    config.encoder_channels = (16, 32, 32)
    config.down_factor = 2
    config.down_num = 2
    config.decoder_channels = (96, 48, 32, 32, 16)
    config.skip_channels = (32, 32, 32, 32, 16)
    config.n_dims = 3
    config.n_skip = 5
    return config

models.py:

Multi-head attention

图3  Transformer Encoder

 Attention理论上是上图中的橙色部分

class Attention(nn.Module):
    def __init__(self, config, vis):
        super(Attention, self).__init__()
        self.vis = vis
        self.num_attention_heads = config.transformer["num_heads"]
        self.attention_head_size = int(config.hidden_size / self.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = Linear(config.hidden_size, self.all_head_size)
        self.key = Linear(config.hidden_size, self.all_head_size)
        self.value = Linear(config.hidden_size, self.all_head_size)

        self.out = Linear(config.hidden_size, config.hidden_size)
        self.attn_dropout = Dropout(config.transformer["attention_dropout_rate"])
        self.proj_dropout = Dropout(config.transformer["attention_dropout_rate"])

        self.softmax = Softmax(dim=-1)

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, hidden_states):
        mixed_query_layer = self.query(hidden_states)
        mixed_key_layer = self.key(hidden_states)
        mixed_value_layer = self.value(hidden_states)

        query_layer = self.transpose_for_scores(mixed_query_layer)
        key_layer = self.transpose_for_scores(mixed_key_layer)
        value_layer = self.transpose_for_scores(mixed_value_layer)

        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        attention_probs = self.softmax(attention_scores)
        weights = attention_probs if self.vis else None
        attention_probs = self.attn_dropout(attention_probs)

        context_layer = torch.matmul(attention_probs, value_layer)
        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)
        attention_output = self.out(context_layer)
        attention_output = self.proj_dropout(attention_output)
        return attention_output, weights

Mlp是图三中的黄色部分 ，前向神经网络

class Mlp(nn.Module):#前向神经网络
    def __init__(self, config):
        super(Mlp, self).__init__()
        self.fc1 = Linear(config.hidden_size, config.transformer["mlp_dim"])
        self.fc2 = Linear(config.transformer["mlp_dim"], config.hidden_size)
        self.act_fn = ACT2FN["gelu"]
        self.dropout = Dropout(config.transformer["dropout_rate"])

        self._init_weights()

    def _init_weights(self):
        nn.init.xavier_uniform_(self.fc1.weight)
        nn.init.xavier_uniform_(self.fc2.weight)
        nn.init.normal_(self.fc1.bias, std=1e-6)
        nn.init.normal_(self.fc2.bias, std=1e-6)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act_fn(x)
        x = self.dropout(x)
        x = self.fc2(x)
        x = self.dropout(x)
        return x

Embeddings是下图的部分

embedding这里是用cov3d来进行patch的，输出的是210块，

(B, n_patch, hidden)=([2, 210, 252])，经过transformer也是这个尺寸。

class Embeddings(nn.Module):
    """Construct the embeddings from patch, position embeddings.
    """
    def __init__(self, config, img_size):
        super(Embeddings, self).__init__()
        self.config = config
        down_factor = config.down_factor
        patch_size = _triple(config.patches["size"])
        n_patches = int((img_size[0]/2**down_factor// patch_size[0]) * (img_size[1]/2**down_factor// patch_size[1]) * (img_size[2]/2**down_factor// patch_size[2]))
        self.hybrid_model = CNNEncoder(config, n_channels=2)
        in_channels = config['encoder_channels'][-1]
        self.patch_embeddings = Conv3d(in_channels=in_channels,
                                       out_channels=config.hidden_size,
                                       kernel_size=patch_size,
                                       stride=patch_size)
        self.position_embeddings = nn.Parameter(torch.zeros(1, n_patches, config.hidden_size))

        self.dropout = Dropout(config.transformer["dropout_rate"])

    def forward(self, x):
        x, features = self.hybrid_model(x)
        x = self.patch_embeddings(x)  # (B, hidden. n_patches^(1/2), n_patches^(1/2))
        x = x.flatten(2)
        x = x.transpose(-1, -2)  # (B, n_patches, hidden)
        embeddings = x + self.position_embeddings
        embeddings = self.dropout(embeddings)
        return embeddings, features

Block就是图3中的灰色框内的卷积层部分，灰色框并不是全部的transformer，而且block的类里不是12个块，只定义了一个块，12个块的循环在Ecoder类中定义

class Block(nn.Module):
    def __init__(self, config, vis):
        super(Block, self).__init__()
        self.hidden_size = config.hidden_size
        self.attention_norm = LayerNorm(config.hidden_size, eps=1e-6)
        self.ffn_norm = LayerNorm(config.hidden_size, eps=1e-6)
        self.ffn = Mlp(config)
        self.attn = Attention(config, vis)

    def forward(self, x):
        h = x

        x = self.attention_norm(x)
        x, weights = self.attn(x)
        x = x + h

        h = x
        x = self.ffn_norm(x)
        x = self.ffn(x)
        x = x + h
        return x, weights

encoder就是12个block放一起

class Encoder(nn.Module):
    def __init__(self, config, vis):
        super(Encoder, self).__init__()
        self.vis = vis
        self.layer = nn.ModuleList()
        self.encoder_norm = LayerNorm(config.hidden_size, eps=1e-6)
        for _ in range(config.transformer["num_layers"]):
            layer = Block(config, vis)
            self.layer.append(copy.deepcopy(layer))

    def forward(self, hidden_states):
        attn_weights = []
        for layer_block in self.layer:
            hidden_states, weights = layer_block(hidden_states)
            if self.vis:
                attn_weights.append(weights)
        encoded = self.encoder_norm(hidden_states)
        return encoded, attn_weights

DecoderBlock是图1和图2的绿色部分， 用于DecoderCup，

class DecoderBlock(nn.Module):
    def __init__(
            self,
            in_channels,
            out_channels,
            skip_channels=0,
            use_batchnorm=True,
    ):
        super().__init__()
        self.conv1 = Conv3dReLU(
            in_channels + skip_channels,
            out_channels,
            kernel_size=3,
            padding=1,
            use_batchnorm=use_batchnorm,
        )
        self.conv2 = Conv3dReLU(
            out_channels,
            out_channels,
            kernel_size=3,
            padding=1,
            use_batchnorm=use_batchnorm,
        )
        self.up = nn.Upsample(scale_factor=2, mode='trilinear', align_corners=False)

    def forward(self, x, skip=None):
        x = self.up(x)
        if skip is not None:
            x = torch.cat([x, skip], dim=1)
        x = self.conv1(x)
        x = self.conv2(x)
        return x

 DecoderCup是整个的解码过程

class DecoderCup(nn.Module):
    def __init__(self, config, img_size):
        super().__init__()
        self.config = config
        self.down_factor = config.down_factor
        head_channels = config.conv_first_channel
        self.img_size = img_size
        self.conv_more = Conv3dReLU(
            config.hidden_size,
            head_channels,
            kernel_size=3,
            padding=1,
            use_batchnorm=True,
        )
        decoder_channels = config.decoder_channels
        in_channels = [head_channels] + list(decoder_channels[:-1])
        out_channels = decoder_channels
        self.patch_size = _triple(config.patches["size"])
        skip_channels = self.config.skip_channels
        blocks = [
            DecoderBlock(in_ch, out_ch, sk_ch) for in_ch, out_ch, sk_ch in zip(in_channels, out_channels, skip_channels)
        ]
        self.blocks = nn.ModuleList(blocks)

Transformer是整个图3的组成，图1中的橙色部分

class Transformer(nn.Module):
    def __init__(self, config, img_size, vis):
        super(Transformer, self).__init__()
        self.embeddings = Embeddings(config, img_size=img_size)
        self.encoder = Encoder(config, vis)

    def forward(self, input_ids):
        embedding_output, features = self.embeddings(input_ids)
        encoded, attn_weights = self.encoder(embedding_output)  # (B, n_patch, hidden)
        return encoded, attn_weights, features

SpatialTransformer这里和VM一模一样，是图1中的Spatial Transformer蓝色部分，用于使移动图像发生形变。

class SpatialTransformer(nn.Module):
    """
    N-D Spatial Transformer

    Obtained from https://github.com/voxelmorph/voxelmorph
    """

    def __init__(self, size, mode='bilinear'):
        super().__init__()

        self.mode = mode

        # create sampling grid
        vectors = [torch.arange(0, s) for s in size]
        grids = torch.meshgrid(vectors)
        grid = torch.stack(grids)
        grid = torch.unsqueeze(grid, 0)
        grid = grid.type(torch.FloatTensor)

        # registering the grid as a buffer cleanly moves it to the GPU, but it also
        # adds it to the state dict. this is annoying since everything in the state dict
        # is included when saving weights to disk, so the model files are way bigger
        # than they need to be. so far, there does not appear to be an elegant solution.
        # see: https://discuss.pytorch.org/t/how-to-register-buffer-without-polluting-state-dict
        self.register_buffer('grid', grid)

    def forward(self, src, flow):
        # new locations
        #print("self.grid.shape", self.grid.shape)
        #print( "flow.shape", flow.shape )
        new_locs = self.grid + flow
        shape = flow.shape[2:]

        # need to normalize grid values to [-1, 1] for resampler
        for i in range(len(shape)):
            new_locs[:, i, ...] = 2 * (new_locs[:, i, ...] / (shape[i] - 1) - 0.5)

        # move channels dim to last position
        # also not sure why, but the channels need to be reversed
        if len(shape) == 2:
            new_locs = new_locs.permute(0, 2, 3, 1)
            new_locs = new_locs[..., [1, 0]]
        elif len(shape) == 3:
            new_locs = new_locs.permute(0, 2, 3, 4, 1)
            new_locs = new_locs[..., [2, 1, 0]]

        return nnf.grid_sample(src, new_locs, align_corners=True, mode=self.mode)

DoubleConv双重卷积，就是2层卷积，代码没有用leakyrelu,只用了relu。

class DoubleConv(nn.Module):
    """(convolution => [BN] => ReLU) * 2"""

    def __init__(self, in_channels, out_channels, mid_channels=None):
        super().__init__()
        if not mid_channels:
            mid_channels = out_channels
        self.double_conv = nn.Sequential(
            nn.Conv3d(in_channels, mid_channels, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv3d(mid_channels, out_channels, kernel_size=3, padding=1),
            nn.ReLU(inplace=True)
        )

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

Down在2层卷积后加入池化层，池化层可以有效的缩小参数矩阵的尺寸，从而减少最后连接层的中的参数数量。所以加入池化层可以加快计算速度和防止过拟合的作用

class Down(nn.Module):
    """Downscaling with maxpool then double conv"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.maxpool_conv = nn.Sequential(
            nn.MaxPool3d(2),
            DoubleConv(in_channels, out_channels)
        )

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

 Conv3dReLU 用于解码中 ，是一个带有relu激活函数的3D卷积，为什么归一化这里，代码用BatchNorm？而图中是instance norm。而且也没用leakyrelu

class Conv3dReLU(nn.Sequential):
    def __init__(
            self,
            in_channels,
            out_channels,
            kernel_size,
            padding=0,
            stride=1,
            use_batchnorm=True,
    ):
        conv = nn.Conv3d(
            in_channels,
            out_channels,
            kernel_size,
            stride=stride,
            padding=padding,
            bias=not (use_batchnorm),
        )
        relu = nn.ReLU(inplace=True)

        bn = nn.BatchNorm3d(out_channels)

        super(Conv3dReLU, self).__init__(conv, bn, relu)

CNNEncoder，就是这3层卷积，输出权重和一个特征图，之后接着embedding

 疑问：为什么图中每层只有一次池化，这里又加入了三个池化？而且上图中instance norm在代码里根本没有

class CNNEncoder(nn.Module):
    def __init__(self, config, n_channels=2):
        super(CNNEncoder, self).__init__()
        self.n_channels = n_channels
        decoder_channels = config.decoder_channels
        encoder_channels = config.encoder_channels
        self.down_num = config.down_num
        self.inc = DoubleConv(n_channels, encoder_channels[0])
        self.down1 = Down(encoder_channels[0], encoder_channels[1])
        self.down2 = Down(encoder_channels[1], encoder_channels[2])
        self.width = encoder_channels[-1]
    def forward(self, x):
        features = []
        x1 = self.inc(x)
        features.append(x1)
        x2 = self.down1(x1)
        features.append(x2)
        feats = self.down2(x2)
        features.append(feats)
        feats_down = feats
        for i in range(self.down_num):
            feats_down = nn.MaxPool3d(2)(feats_down)
            features.append(feats_down)
        return feats, features[::-1]

RegistrationHead配准，用于主框架中

class RegistrationHead(nn.Sequential):
    def __init__(self, in_channels, out_channels, kernel_size=3, upsampling=1):
        conv3d = nn.Conv3d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2)
        conv3d.weight = nn.Parameter(Normal(0, 1e-5).sample(conv3d.weight.shape))
        conv3d.bias = nn.Parameter(torch.zeros(conv3d.bias.shape))
        super().__init__(conv3d)

ViTVNet：整体网络框架

class ViTVNet(nn.Module):
    def __init__(self, config, img_size=(64, 256, 256), int_steps=7, vis=False, mode='bilinear'):
        super(ViTVNet, self).__init__()
        self.transformer = Transformer(config, img_size, vis)
        self.decoder = DecoderCup(config, img_size)
        self.reg_head = RegistrationHead(
            in_channels=config.decoder_channels[-1],
            out_channels=config['n_dims'],
            kernel_size=3,
        )
        self.spatial_trans = SpatialTransformer(img_size, mode)
        self.config = config
        #self.integrate = VecInt(img_size, int_steps)
    def forward(self, x):

        source = x[:,0:1,:,:]
        x, attn_weights, features = self.transformer(x)  # (B, n_patch, hidden)
        x = self.decoder(x, features)
        flow = self.reg_head(x)
        #flow = self.integrate(x)
        #img = x[0].cuda( )
        #flow = x[1].cuda( )
        
        out = self.spatial_trans(source, flow)
        #out = self.spatial_trans( source, flow )
        return out, flow