Diff_unet_attn.py

import math
import torch
import torch.nn as nn
import torch.nn.functional as F
# This script is from the following repositories
# https://github.com/ermongroup/ddim
# https://github.com/bahjat-kawar/ddrm


def get_timestep_embedding(timesteps, embedding_dim):
    """
    This matches the implementation in Denoising Diffusion Probabilistic Models:
    From Fairseq.
    Build sinusoidal embeddings.
    This matches the implementation in tensor2tensor, but differs slightly
    from the description in Section 3.5 of "Attention Is All You Need".
    """
    assert len(timesteps.shape) == 1

    half_dim = embedding_dim // 2
    emb = math.log(10000) / (half_dim - 1)
    emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
    emb = emb.to(device=timesteps.device)
    emb = timesteps.float()[:, None] * emb[None, :]
    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
    if embedding_dim % 2 == 1:  # zero pad
        emb = torch.nn.functional.pad(emb, (0, 1, 0, 0))
    return emb


def nonlinearity(x):
    # swish
    return x*torch.sigmoid(x)


def Normalize(in_channels):
    return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)


class Upsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            self.conv = torch.nn.Conv2d(in_channels,in_channels,kernel_size=3,stride=1,padding=1)
            #padding=1. The convoution won't change the input shape 

    def forward(self, x):
        x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
        #scale_factor. Perform up sampling to the input. The image size becomes (image_h*2) * (image_w*2)
        if self.with_conv:
            x = self.conv(x)
        return x


class Downsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            # no asymmetric padding in torch conv, must do it ourselves
            self.conv = torch.nn.Conv2d(in_channels,in_channels,kernel_size=3,stride=2,padding=0)

    def forward(self, x):
        if self.with_conv:#whether use convoution to perform downsampling
            pad = (0, 1, 0, 1)
            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
            x = self.conv(x)
        else:
            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
        return x
        #This is down sampling, The image size becomes image_h/2 * image_w/2


class ResnetBlock(nn.Module):
    def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False,
                 dropout, temb_channels=512):
        super().__init__()
        self.in_channels = in_channels
        out_channels = in_channels if out_channels is None else out_channels
        self.out_channels = out_channels
        self.use_conv_shortcut = conv_shortcut

        self.norm1 = Normalize(in_channels)
        self.conv1 = torch.nn.Conv2d(in_channels,out_channels,kernel_size=3,stride=1,padding=1)
        self.temb_proj = torch.nn.Linear(temb_channels,out_channels)#projection
        self.norm2 = Normalize(out_channels)
        self.dropout = torch.nn.Dropout(dropout)
        self.conv2 = torch.nn.Conv2d(out_channels,out_channels,kernel_size=3,stride=1,padding=1)
        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:#whether use convoution to perform residual
                self.conv_shortcut = torch.nn.Conv2d(in_channels,out_channels,kernel_size=3,stride=1,padding=1)
            else:
                self.nin_shortcut = torch.nn.Conv2d(in_channels,out_channels,kernel_size=1,stride=1,padding=0)

    def forward(self, x, temb):
        h = x
        h = self.norm1(h)
        h = nonlinearity(h)
        h = self.conv1(h)
        #temb=token embedding dimension
        h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None]

        h = self.norm2(h)
        h = nonlinearity(h)
        h = self.dropout(h)
        h = self.conv2(h)

        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:
                x = self.conv_shortcut(x)
            else:
                x = self.nin_shortcut(x)

        return x+h


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

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(in_channels, in_channels, kernel_size = 1, stride = 1, padding = 0)
        self.k = torch.nn.Conv2d(in_channels, in_channels, kernel_size = 1, stride = 1, padding = 0)
        self.v = torch.nn.Conv2d(in_channels, in_channels, kernel_size = 1, stride = 1, padding = 0)
        self.proj_out = torch.nn.Conv2d(in_channels, in_channels, kernel_size = 1, stride = 1, padding=0)

    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)
        
        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # compute attention
        b, c, h, w = q.shape
        q = q.reshape(b, c, h*w)
        q = q.permute(0, 2, 1)   # b,hw,c
        k = k.reshape(b, c, h*w)  # b,c,hw
        w_ = torch.bmm(q, k)     # b,hw,hw    w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
        w_ = w_ * (int(c)**(-0.5))
        w_ = torch.nn.functional.softmax(w_, dim=2)

        # attend to values
        v = v.reshape(b, c, h*w)
        w_ = w_.permute(0, 2, 1)   # b,hw,hw (first hw of k, second of q)
        # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
        h_ = torch.bmm(v, w_)
        h_ = h_.reshape(b, c, h, w)

        h_ = self.proj_out(h_)

        return x+h_

class MultiHeadAttnBlock(nn.Module):
    def __init__(self, in_channels):
        super().__init__()
        num_heads = 8
        assert in_channels % num_heads == 0, "in_channels must be divisible by num_heads"
        
        self.in_channels = in_channels
        self.num_heads = num_heads
        self.head_dim = in_channels // num_heads

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
        self.k = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
        self.v = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
        self.proj_out = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)

    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)

        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # Split heads
        b, c, h, w = q.shape
        q = q.reshape(b, self.num_heads, self.head_dim, h * w).permute(0, 1, 3, 2)  # b, num_heads, hw, head_dim
        k = k.reshape(b, self.num_heads, self.head_dim, h * w)  # b, num_heads, head_dim, hw
        v = v.reshape(b, self.num_heads, self.head_dim, h * w)  # b, num_heads, head_dim, hw

        # Compute attention
        w_ = torch.einsum('bnqd,bnkd->bnqk', q, k)  # b, num_heads, hw, hw
        w_ = w_ * (self.head_dim ** -0.5)
        w_ = torch.nn.functional.softmax(w_, dim=-1)

        # Attend to values
        h_ = torch.einsum('bnqk,bnvd->bnqd', w_, v)
        h_ = h_.permute(0, 1, 3, 2).reshape(b, c, h, w)  # Merge heads

        h_ = self.proj_out(h_)

        return x + h_
        
class DiffusionUNet(nn.Module):
    def __init__(self,ch, num_res_blocks, image_size, drop_out):
        super().__init__()
        self.dropout=drop_out

        ch_mult = [1, 1, 2, 2, 4, 4]
        
        attn_resolutions = [32,16,]
        #only when the image becomes attn_resolutions*attn_resolutions, use the self attention.
        resamp_with_conv = True
        #dropout = config.model.dropout
        #in_channels = config.model.in_channels * 2 if config.data.conditional else config.model.in_channels
        in_channels = 2
        out_ch = 1
        resolution = image_size
        
        self.ch = ch
        self.temb_ch = self.ch*4 #the time step embedding dimension
        self.num_resolutions = len(ch_mult) #the nubmer of down-sampling and up-sampling
        #the ch_mult is a tuple, specify the channels in different level of feature extraction (down-sampling or sampling)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels
        
        #self.shearemb = nn.Module()
        #self.shearemb.dense = nn.ModuleList([
        #    torch.nn.Linear(3, self.temb_ch),
        #    torch.nn.Linear(self.temb_ch, self.temb_ch),])
        self.shear_emb = nn.Embedding(num_embeddings=3, embedding_dim=self.temb_ch)
        
        # timestep embedding
        self.temb = nn.Module()
        self.temb.dense = nn.ModuleList([
            torch.nn.Linear(self.ch, self.temb_ch),
            torch.nn.Linear(self.temb_ch, self.temb_ch),])

        # downsampling
        self.conv_in = torch.nn.Conv2d(in_channels, self.ch, kernel_size = 3, stride = 1, padding = 1) #the first feature extraction
        
        curr_res = resolution
        in_ch_mult = (1,)+tuple(ch_mult)
        self.down = nn.ModuleList()
        block_in = None
        for i_level in range(self.num_resolutions):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_in = ch*in_ch_mult[i_level]
            block_out = ch*ch_mult[i_level]
            for i_block in range(self.num_res_blocks):
                block.append(ResnetBlock(in_channels = block_in, out_channels = block_out, temb_channels = self.temb_ch, dropout = self.dropout))
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(AttnBlock(block_in))
            down = nn.Module()
            down.block = block
            down.attn = attn
            if i_level != self.num_resolutions-1:
                down.downsample = Downsample(block_in, resamp_with_conv)
                curr_res = curr_res // 2 #afte one down-sampling, the resolution becomes half.
            self.down.append(down)

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(in_channels=block_in,out_channels=block_in,temb_channels=self.temb_ch,dropout=self.dropout)
        self.mid.attn_1 = AttnBlock(block_in)
        self.mid.block_2 = ResnetBlock(in_channels=block_in,out_channels=block_in,temb_channels=self.temb_ch,dropout=self.dropout)

        # upsampling
        self.up = nn.ModuleList()
        for i_level in reversed(range(self.num_resolutions)):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_out = ch*ch_mult[i_level]
            skip_in = ch*ch_mult[i_level]
            for i_block in range(self.num_res_blocks+1):
                if i_block == self.num_res_blocks:
                    skip_in = ch*in_ch_mult[i_level]
                block.append(ResnetBlock(in_channels=block_in+skip_in,out_channels=block_out,temb_channels=self.temb_ch,dropout=self.dropout))
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(AttnBlock(block_in))
            up = nn.Module()
            up.block = block
            up.attn = attn
            if i_level != 0:
                up.upsample = Upsample(block_in, resamp_with_conv)
                curr_res = curr_res * 2 #after one up-sampling, the resolution becomes twice.
            self.up.insert(0, up)  # prepend to get consistent order

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(block_in,out_ch,kernel_size=3,stride=1,padding=1)
        
    def forward(self, x, t, shear):
        #condition is the initial condition used as key and value in cross attention
        
        assert x.shape[2] == x.shape[3] == self.resolution
        #print(x.shape)
        # timestep embedding
        temb = get_timestep_embedding(t, self.ch)
        temb = self.temb.dense[0](temb)
        temb = nonlinearity(temb)
        temb = self.temb.dense[1](temb)

        #shear embedding
        shear_emb = self.shear_emb(shear)
        '''
        shear_emb=nn.functional.one_hot(shear, num_classes=3).type(torch.float) #one hot
        shear_emb=self.shearemb.dense[0](shear_emb)
        shear_emb = nonlinearity(shear_emb)
        shear_emb=self.shearemb.dense[1](shear_emb)
        '''

        temb = temb + shear_emb
        
        # downsampling
        hs = [self.conv_in(x)] #the first feature extraction
        for i_level in range(self.num_resolutions):
            for i_block in range(self.num_res_blocks):
                h = self.down[i_level].block[i_block](hs[-1], temb) #perform one Restblock
                if len(self.down[i_level].attn) > 0: #check if there is Attblock
                    h = self.down[i_level].attn[i_block](h)
                hs.append(h)
            if i_level != self.num_resolutions-1:
                hs.append(self.down[i_level].downsample(hs[-1])) #perform down-sampling
            #store the each level down-sampling ouput for skip connection when up-sampling

        # middle
        h = hs[-1]
        h = self.mid.block_1(h, temb)
        h = self.mid.attn_1(h)
        h = self.mid.block_2(h, temb)

        # upsampling
        for i_level in reversed(range(self.num_resolutions)):
            for i_block in range(self.num_res_blocks+1):
                h = self.up[i_level].block[i_block](
                    torch.cat([h, hs.pop()], dim=1), temb) #this is skip connection
                if len(self.up[i_level].attn) > 0:
                    h = self.up[i_level].attn[i_block](h)
            if i_level != 0:
                h = self.up[i_level].upsample(h)

        # end
        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        return h