diffusion.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sketch Diffusion\n",
    "### A Notebook to play with with diffusion on Googles Quickdraw Dataset using your own Laptop or in Google Colab\n",
    "If running on Colab, make sure to select a GPU Runtime by selecting: Runtime - Change untime type - gpu. \\\n",
    "For Colab we also need to clone the GitHub Repository and install some required dependencies. If you run this locally, you can skip the next cell"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!git clone https://github.com/pitneitemeier/sketch_diffusion.git\n",
    "%cd sketch_diffusion\n",
    "!unzip -q car_64.zip\n",
    "!pip install pytorch_lightning einops --quiet\n",
    "!mkdir plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch \n",
    "from torch import nn\n",
    "import torch.nn.functional as F\n",
    "from torch.utils.data import Dataset, DataLoader, random_split\n",
    "import torchvision\n",
    "import torchvision.transforms as T\n",
    "import pytorch_lightning as pl\n",
    "import matplotlib.pyplot as plt\n",
    "import einops"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Loading and analysing the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "from Sketchdata import SketchData, SingleImageData\n",
    "IMG_SIZE = 32\n",
    "CHANNELS = 1\n",
    "\n",
    "transform = T.Compose([T.ToTensor(), T.Resize(IMG_SIZE)])\n",
    "# to show the images, clamp them to [0,1] and invert the colors to have a white background\n",
    "rev_transform = T.Compose([lambda x : 1 - torch.clamp(x, min=0, max=1)])\n",
    "dataset = SketchData(['car_64'], transform)\n",
    "\n",
    "def get_imgs(num_samples, dataset):\n",
    "    return torch.stack([dataset[i] for i in range(num_samples)], dim=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since we will be looking at lots of grids of Images, we define a helper method that allows us to visualize these."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_grid(imgs, n_row):\n",
    "    '''\n",
    "    Function to plot a grid of images\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    imgs: torch.tensor\n",
    "        expects a tensor of shape [b, c, h, w] of images\n",
    "    n_row: int\n",
    "        the number of images per row\n",
    "    '''\n",
    "    grid = torchvision.utils.make_grid(imgs, nrow=n_row)\n",
    "    fig, ax = plt.subplots(figsize=(2 * n_row, 2 * imgs.shape[0] // n_row))\n",
    "    ax.axis('off')\n",
    "    ax.imshow(grid.permute(1,2,0).cpu(), cmap='gray')\n",
    "    plt.close(fig)\n",
    "    return fig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1440x144 with 1 Axes>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_samples = 10\n",
    "imgs = rev_transform(get_imgs(num_samples, dataset))\n",
    "plot_grid(imgs, num_samples)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Defining the diffusion process\n",
    "A diffusion model is a process inspired by nonequilibrium thermodynamics (see [Deep Unsupervised Learning using Nonequilibrium Thermodynamics](https://arxiv.org/abs/1503.03585)) and is based on the Markov chain depicted in the following figure:\n",
    "\\\n",
    "![The markov chain that characterizes diffusion models. In the forward process Gaussian noise is added to an input image until signal is destroyed. In the reverse process, learned Gaussian transitions are used to subtract the noise.](images/diffusion_markov_chain.svg)\n",
    "\\\n",
    "In this work we largely follow the process introduced by Ho et al. (See [Denoising Diffusion Probabilistic Models](https://arxiv.org/abs/2006.11239)).\n",
    "Starting from an image $x_0 \\sim q(x_0)$ from a ground truth distribution $q$, the \\textit{forward process} gradually adds Gaussian noise to the image according to a variance schedule $\\beta_1, \\ldots, \\beta_T$:\n",
    "\\begin{equation*}\n",
    "    q(x_t | x_{t-1}) = \\mathcal{N}(x_t;\\sqrt{1-\\beta_t}x_{t-1},\\beta_t I).\n",
    "\\end{equation*}\n",
    "The \\textit{reverse process} is another Markov chain with Gaussian transitions that aims to reverse the forward process using a learned mean $\\mu_\\theta(x_t,t)$. Starting from $x_T$ the process samples from\n",
    "\\begin{equation*}\n",
    "    p_\\theta(x_{t-1}|x_t)= \\mathcal{N}(x_{t-1};\\mu_\\theta(x_t,t),\\beta_t I).\n",
    "\\end{equation*}\n",
    "To derive a suitable training loss, we first note that we can draw random samples from the forward process in a closed form, since the sum of two Gaussians with standard deviations $\\sigma_1, \\sigma_2$ is a Gaussian with standard deviation $\\sqrt{\\sigma_1^2 + \\sigma_2^2}$. With $\\epsilon_i, \\epsilon \\sim \\mathcal{N}(0,1)$, $\\alpha_t := \\ 1-\\beta_t$ and $\\overline{\\alpha}_t := \\prod_{i=1}^t \\alpha_i$ it follows that\n",
    "\\begin{align*}\n",
    "    x_t &= \\sqrt{\\alpha_t}x_{t-1} + \\sqrt{1-\\alpha_t} \\epsilon_t \\\\\n",
    "    &= \\sqrt{\\alpha_t}(\\sqrt{\\alpha_{t-1}}x_{t-2} + \\sqrt{1-\\alpha_{t-1}} \\epsilon_{t-1}) + \\sqrt{1-\\alpha_t} \\epsilon_t \\\\\n",
    "    &= \\sqrt{\\overline{\\alpha}_t} x_0 + \\sqrt{1-\\overline{\\alpha}_t} \\epsilon\n",
    "\\end{align*}\n",
    "and therefore $q(x_t|x_0) = \\mathcal{N}(x_t; \\sqrt{\\overline{\\alpha}_t}x_0, (1-\\overline{\\alpha}_t)I)$. From this it is possible to derive the posteriors of the forward process using\n",
    "\\begin{align*}\n",
    "    q(x_{t-1}|x_t,x_0) &= q(x_{t}|x_{t-1}, x_0) \\frac{ q(x_{t-1}|x_0) }{q(x_t|x_0)} \\\\\n",
    "    &= \\mathcal{N}(x_{t-t}; \\tilde{\\mu}(x_t, x_0), \\tilde{\\beta}_t)\\\\\n",
    "    \\text{with } \\tilde{\\mu}(x_t, x_0) &:= \\frac{\\sqrt{\\overline{\\alpha}_{t-1}}\\beta_t x_0 + \\sqrt{\\alpha_t}(1-\\overline{\\alpha}_{t-1})x_t}{1-\\overline{\\alpha}_t} \\\\\n",
    "    \\text{and } \\tilde{\\beta}_t &:= \\frac{1-\\overline{\\alpha}_{t-1}}{1-\\overline{\\alpha}_t} \\beta_t \\approx \\beta_t\n",
    "\\end{align*}\n",
    "With knowledge of the noise $\\epsilon$ that was added in the forward process, we can simplify this using $x_t(x_0, \\epsilon) = \\sqrt{\\overline{\\alpha}_t} x_0 + \\sqrt{1-\\overline{\\alpha}_t} \\epsilon$ and $x_0 = \\frac{1}{\\sqrt{\\overline{\\alpha}_t}}(x_t(x_0,\\epsilon) - \\sqrt{1-\\overline{\\alpha}_t}\\epsilon)$:\n",
    "\\begin{align*}\n",
    "    \\tilde{\\mu}(x_t(x_0,\\epsilon), x_0) = \\frac{1}{\\sqrt{\\alpha_t}}(x_t(x_0,\\epsilon) - \\frac{\\beta_t}{\\sqrt{1-\\overline{\\alpha}_t}}\\epsilon).\n",
    "\\end{align*}\n",
    "Since we want the reverse process to replicate the posterior probability, we can reparametrize $\\mu_\\theta(x_t,x_0)$ as \n",
    "\\begin{align*}\n",
    "    \\mu_\\theta(x_t,x_0) &= \\tilde{\\mu}(x_t, \\frac{1}{\\sqrt{\\overline{\\alpha}_t}}(x_t - \\sqrt{1-\\overline{\\alpha}_t}\\epsilon_\\theta(x_t, t)))\\\\ \n",
    "    &= \\frac{1}{\\sqrt{\\alpha_t}}(x_t - \\frac{\\beta_t}{\\sqrt{1-\\overline{\\alpha}_t}}\\epsilon_\\theta(x_t, t))\n",
    "\\end{align*}\n",
    "We can now define the training loss for a single sample as \n",
    "\\begin{equation*}\n",
    "    L(t, x_0, \\epsilon) := \\lVert{\\epsilon - \\epsilon_\\theta(x_t(x_0,\\epsilon), t)\\rVert}^2.\n",
    "\\end{equation*}\n",
    "Note that it is also possible to derive this loss from a negative log likelihood using the variational lower bound and Kullback-Leibler divergence (See [Denoising Diffusion Probabilistic Models](https://arxiv.org/abs/2006.11239)). During training we then evaluate this loss for a batch of $t \\sim \\text{Uniform}\\{1, \\ldots, T\\}$, $\\epsilon \\sim \\mathcal{N}(0,I)$ and $x_0 \\sim q(x_0)$. The training process looks like this: \n",
    "\n",
    "![Training process](images/diffusion_training.svg)\n",
    "\n",
    "To generate an image, we start from $x_T \\sim \\mathcal{N}(0,I)$ and then step through the full Markov chain in reverse. This process then looks like this:\n",
    "\n",
    "![Sampling process](images/diffusion_sampling.svg)\n",
    "\n",
    "### Noise schedule\n",
    "First we define a class that saves the noise schedule $\\beta_1, \\ldots, \\beta_T$ and provides methods to calculate $\\overline{\\alpha}_t$ and $\\alpha_t$ in training and sampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "class NoiseSchedule:\n",
    "    '''\n",
    "    Class that implements the noise schedule to use with the diffusion\n",
    "    '''\n",
    "    def __init__(self, betas):\n",
    "        '''\n",
    "        Parameters \n",
    "        ----------\n",
    "        betas : torch.tensor\n",
    "            the noise schedule to be used\n",
    "        '''\n",
    "        # calculate alpha_bar directly, since it is the only thing that is used in training\n",
    "        self.device = 'cpu'\n",
    "        self.alphas = 1. - betas\n",
    "        self.alpha_bars = torch.cumprod(self.alphas, dim=0)\n",
    "\n",
    "        self.num_steps = self.alpha_bars.shape[0]\n",
    "\n",
    "    def to(self, device):\n",
    "        self.alpha_bars = self.alpha_bars.to(device)\n",
    "        self.alphas = self.alphas.to(device)\n",
    "        self.device = device\n",
    "    \n",
    "    def get_random_sample(self, x_0):\n",
    "        '''\n",
    "        Parameters\n",
    "        ----------\n",
    "        x_0 : torch.tensor\n",
    "            The training images to generate samples for with shape [batch, channels, height, width]\n",
    "\n",
    "        Returns\n",
    "        -------\n",
    "        t : torch.Tensor\n",
    "            Random times in {0, 1/num_steps, ..., 1}\n",
    "        alpha_bar_t : torch.tensor\n",
    "            the cumulative product of the alphas for each timestep\n",
    "        '''\n",
    "        if self.device != x_0.device:\n",
    "            self.to(x_0.device)\n",
    "        # randomly choose a step in the diffusion chain to train on for each input image and retrieve the corresponding alpha_bar\n",
    "        steps = torch.randint(low=0, high=self.num_steps, size=(x_0.shape[0],), device=x_0.device)\n",
    "        return steps / self.num_steps, self.alpha_bars[steps]\n",
    "    \n",
    "    def get_sample(self, steps):\n",
    "        '''\n",
    "        Parameters\n",
    "        ----------\n",
    "        steps : torch.tensor of ints\n",
    "            The steps in the diffusion chain to get samples for\n",
    "\n",
    "        Returns\n",
    "        -------\n",
    "        t, alpha_t, alpha_bar_t\n",
    "        '''\n",
    "        assert (steps<self.num_steps).all(), f'maximum step should be smaller than the total number of steps'\n",
    "        if self.device != steps.device:\n",
    "          self.to(steps.device)\n",
    "        return steps / self.num_steps, self.alphas[steps], self.alpha_bars[steps]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Forward process\n",
    "In the forward process, we want to access the image at a specific step in the diffusion chain. As derived above we do not have to step through the whole chain, but can use the fact that the sum of Gaussians is again a Gaussian. We define a method that implements this and then show an example of the forward process."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "def noise_image(x_0, alpha_bar_t):\n",
    "    '''\n",
    "    Function to return a noisy version of the image according to the set noise schedule with\n",
    "    x_t ~ N(sqrt(alpha_bar_t)x_0, sqrt(1-alpha_bar_t)I)\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    x_0 : torch.tensor\n",
    "        the original input image with shape [batch, channels, height, width]\n",
    "    alpha_bar : torch.tensor\n",
    "        the accumulated noise schedule with shape [batch,]\n",
    "    \n",
    "    Returns\n",
    "    -------\n",
    "    x_t : torch.tensor\n",
    "        the noisy image\n",
    "    noise : torch.tensor\n",
    "        the noise that was added to the image\n",
    "    '''\n",
    "    noise = torch.randn_like(x_0)\n",
    "    alpha_bar_t = alpha_bar_t.view(-1,1,1,1)\n",
    "    return torch.sqrt(alpha_bar_t) * x_0 + torch.sqrt(1 - alpha_bar_t) * noise, noise"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1440x576 with 1 Axes>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "DIFFUSION_STEPS = 200\n",
    "betas = torch.linspace(1e-4, 2e-2, DIFFUSION_STEPS)\n",
    "noise_schedule = NoiseSchedule(betas)\n",
    "\n",
    "# show an example of the noise process\n",
    "show_steps = 10\n",
    "num_samples=4\n",
    "image = get_imgs(num_samples, dataset)\n",
    "steps = torch.linspace(0,DIFFUSION_STEPS-1, show_steps, dtype=torch.int64)\n",
    "x_0 = image.repeat_interleave(steps.shape[0],0)\n",
    "steps = steps.repeat(image.shape[0],1,1,1)\n",
    "\n",
    "t, _, alpha_bar_t = noise_schedule.get_sample(steps)\n",
    "x_t, _ = noise_image(x_0, alpha_bar_t)\n",
    "x_t = rev_transform(x_t)\n",
    "\n",
    "plot_grid(x_t, show_steps)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reverse process"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "@torch.no_grad()\n",
    "def denoise_step(model, x_t, t, alpha_t, alpha_bar_t):\n",
    "    '''\n",
    "    Function that does a single denoising step\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    x_{t-1} : torch.tensor\n",
    "        the denoised images of shape [batch, channels, height, width]\n",
    "    '''\n",
    "    # reshape inputs for broadcasting\n",
    "    alpha_t = alpha_t.view(-1,1,1,1)\n",
    "    alpha_bar_t = alpha_bar_t.view(-1,1,1,1)\n",
    "\n",
    "    sigma_t = torch.sqrt(1 - alpha_t)\n",
    "    gamma_t = (1 - alpha_t) / torch.sqrt(1 - alpha_bar_t)\n",
    "    noise = torch.randn_like(x_t)\n",
    "\n",
    "    # do not add noise in the last step\n",
    "    add_noise = (t != 0)\n",
    "\n",
    "    return 1/torch.sqrt(alpha_t) * (x_t - gamma_t * model(x_t, t)) + add_noise * sigma_t * noise\n",
    "\n",
    "@torch.no_grad()\n",
    "def generate_images(model, num_examples, show_steps, img_size, channels):\n",
    "    '''\n",
    "    Function that implements Algorithm 2 of the paper 'Denoising Diffusion Probabilistic Models'\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    model : pl.LightningModule\n",
    "        the diffusion model\n",
    "    num_examples : int\n",
    "        the number of images we want to generate\n",
    "    show_steps : int\n",
    "        the number of intermediate generation steps to show\n",
    "    \n",
    "    Returns\n",
    "    -------\n",
    "    imgs : torch.Tensor\n",
    "        the sequence of images that was generated of shape [num_examples * (show_steps + 1), c, h, w]\n",
    "    '''\n",
    "\n",
    "    img =  torch.randn((num_examples, channels, img_size, img_size), device=model.device)\n",
    "    imgs = []\n",
    "    num_steps = model.noise_schedule.num_steps\n",
    "    save_every = num_steps // show_steps if show_steps != 0 else 0\n",
    "    for i in range(num_steps):\n",
    "        if i % save_every  == 0 and show_steps!=0:\n",
    "            imgs.append(img)\n",
    "        t, alpha_t, alpha_bar_t = model.noise_schedule.get_sample(torch.tensor([num_steps - (i+1)], device=model.device))\n",
    "        img = denoise_step(model, img, t, alpha_t, alpha_bar_t)\n",
    "    imgs.append(img)\n",
    "    imgs = torch.stack(imgs, dim=1).reshape(-1, *img.shape[1:])\n",
    "    return imgs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Defining the model\n",
    "The model that we use here is a slightly modified UNet (see [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://arxiv.org/abs/1505.04597))\n",
    "\n",
    "![The model](images/unet_model.svg)\n",
    "\\\n",
    "\n",
    "The ResNet Blocks consist of two 3x3 convolutions with stride 1 that are combined with Groupnorm and a residual connection. To encode the time input, a sinusoidal embedding as presented by Vaswani et al \\cite{Vaswani2017} is used. The embedding is then added onto the image in the channel dimension (so each pixel in a channel gets the same embedding). Before down and upsampling, a self-attention layer is used. The downsampling consists of a 4x4 convolution with stride 2 and the upsampling is a 4x4 transposed convolution with stride 2. The activation in all layers is the GELU function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_upsample(in_channels, out_channels):\n",
    "    '''returns a class that doubles the image dimension using a transposed convolution'''\n",
    "    return nn.ConvTranspose2d(in_channels, out_channels, kernel_size=4, stride=2, padding=1)\n",
    "\n",
    "def get_downsample(channels):\n",
    "    '''returns a class that halves the image dimension using a transposed convolution'''\n",
    "    return nn.Conv2d(channels, channels, kernel_size=4, stride=2, padding=1)\n",
    "\n",
    "def get_ConvNormBlock(in_channels, out_channels, num_groups, activation):\n",
    "    '''returns a block of a 3x3 convolution followed by normalization and an activation'''\n",
    "    assert out_channels % num_groups == 0, 'out_channels must be divisible by the number of groups'\n",
    "    return nn.Sequential(\n",
    "        nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1),\n",
    "        nn.GroupNorm(num_groups, out_channels),\n",
    "        activation()\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Generate a base time embedding that is fed into each layer and can then be adjusted to the number of channels in that layer\n",
    "class BaseTimeEmbedding(nn.Module):\n",
    "    '''A class that turns a batch of scalar time inputs into time embedding vectors in the channel dimension using sin and cosine embeddings.'''\n",
    "    def __init__(self, base_time_dim, activation=nn.GELU) -> None:\n",
    "        super().__init__()\n",
    "        self.base_time_dim = base_time_dim\n",
    "        self.layer_1 = nn.Linear(base_time_dim, base_time_dim)\n",
    "        self.layer_2 = nn.Linear(base_time_dim, base_time_dim)\n",
    "        self.activation = activation()\n",
    "\n",
    "    def forward(self, t):\n",
    "        # use a sin/cosine embedding similar to the 'attention is all you need' paper\n",
    "        omega = 10000 ** ( - 2 / self.base_time_dim * torch.arange(1, self.base_time_dim // 2 + 1, 1, device=t.device) )\n",
    "        t = t.unsqueeze(1)\n",
    "        sines = torch.sin( omega * t )\n",
    "        cosines = torch.cos( omega * t )\n",
    "        t = torch.cat([sines, cosines], dim=1)\n",
    "        t = self.activation(self.layer_1(t))\n",
    "        t = self.layer_2(t)\n",
    "        return t\n",
    "\n",
    "class TimeResNetBlock(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels, num_groups, base_time_dim, activation=nn.GELU) -> None:\n",
    "        super().__init__()\n",
    "\n",
    "        self.time_embed = nn.Sequential(\n",
    "            activation(),\n",
    "            nn.Linear(base_time_dim, out_channels)\n",
    "        )\n",
    "\n",
    "        self.conv_norm_block_1 = get_ConvNormBlock(in_channels, out_channels, num_groups, activation=activation)\n",
    "        self.conv_norm_block_2 = get_ConvNormBlock(out_channels, out_channels, num_groups, activation=activation)\n",
    "\n",
    "        # since we allow for different number of input and output channels, a 1x1 convolution has to be used to bring \n",
    "        # input and output to the same shape\n",
    "        self.res_conv = nn.Conv2d(in_channels, out_channels, kernel_size=1) if in_channels != out_channels else nn.Identity()\n",
    "    \n",
    "    def forward(self, x, t):\n",
    "        t = self.time_embed(t)\n",
    "        h = self.conv_norm_block_1(x)\n",
    "        h += t.reshape(*t.shape, 1, 1)\n",
    "        h = self.conv_norm_block_2(h)\n",
    "        return h + self.res_conv(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Implementation of attention and linear attention taken from https://huggingface.co/blog/annotated-diffusion\n",
    "\n",
    "class Attention(nn.Module):\n",
    "    def __init__(self, in_channels, heads, head_channels):\n",
    "        super().__init__()\n",
    "        self.scale = head_channels ** -0.5\n",
    "        self.heads = heads\n",
    "        hidden_channels = head_channels * heads\n",
    "        self.to_qkv = nn.Sequential(\n",
    "            nn.GroupNorm(1, in_channels),\n",
    "            nn.Conv2d(in_channels, hidden_channels * 3, 1, bias=False)\n",
    "        )\n",
    "        self.to_out = nn.Conv2d(hidden_channels, in_channels, 1)\n",
    "\n",
    "    def forward(self, x):\n",
    "        b, c, h, w = x.shape\n",
    "        qkv = self.to_qkv(x).chunk(3, dim=1)\n",
    "        q, k, v = map(\n",
    "            lambda t: einops.rearrange(t, \"b (h c) x y -> b h c (x y)\", h=self.heads), qkv\n",
    "        )\n",
    "        q = q * self.scale\n",
    "\n",
    "        sim = torch.einsum(\"b h d i, b h d j -> b h i j\", q, k)\n",
    "        sim = sim - sim.amax(dim=-1, keepdim=True).detach()\n",
    "        attn = sim.softmax(dim=-1)\n",
    "\n",
    "        out = torch.einsum(\"b h i j, b h d j -> b h i d\", attn, v)\n",
    "        out = einops.rearrange(out, \"b h (x y) d -> b (h d) x y\", x=h, y=w)\n",
    "        return self.to_out(out) + x\n",
    "\n",
    "class LinearAttention(nn.Module):\n",
    "    def __init__(self, in_channels, heads, head_channels):\n",
    "        super().__init__()\n",
    "        self.scale = head_channels ** -0.5\n",
    "        self.heads = heads\n",
    "        hidden_channels = head_channels * heads\n",
    "        self.to_qkv = nn.Sequential(\n",
    "            nn.GroupNorm(1, in_channels),\n",
    "            nn.Conv2d(in_channels, hidden_channels * 3, 1, bias=False)\n",
    "        )\n",
    "\n",
    "        self.to_out = nn.Conv2d(hidden_channels, in_channels, 1)\n",
    "\n",
    "    def forward(self, x):\n",
    "        b, c, h, w = x.shape\n",
    "        qkv = self.to_qkv(x).chunk(3, dim=1)\n",
    "        q, k, v = map(\n",
    "            lambda t: einops.rearrange(t, \"b (h c) x y -> b h c (x y)\", h=self.heads), qkv\n",
    "        )\n",
    "\n",
    "        q = q.softmax(dim=-2)\n",
    "        k = k.softmax(dim=-1)\n",
    "\n",
    "        q = q * self.scale\n",
    "        context = torch.einsum(\"b h d n, b h e n -> b h d e\", k, v)\n",
    "\n",
    "        out = torch.einsum(\"b h d e, b h d n -> b h e n\", context, q)\n",
    "        out = einops.rearrange(out, \"b h c (x y) -> b (h c) x y\", h=self.heads, x=h, y=w)\n",
    "        return self.to_out(out) + x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DownSampleBlock(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels, num_groups, base_time_dim, activation=nn.GELU, attention=True) -> None:\n",
    "        super().__init__()\n",
    "        self.block_1 = TimeResNetBlock(in_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.block_2 = TimeResNetBlock(out_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.attention = LinearAttention(out_channels, 4, out_channels // 4) if attention else nn.Identity()\n",
    "        self.downsample = get_downsample(out_channels)\n",
    "\n",
    "    def forward(self, x, t):\n",
    "        h = self.block_1(x, t)\n",
    "        h = self.block_2(h, t)\n",
    "        h = self.attention(h)\n",
    "\n",
    "        # return downsampled input and the hidden state for residual connection in UNet\n",
    "        return self.downsample(h), h\n",
    "\n",
    "class UpSampleBlock(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels, num_groups, base_time_dim, activation=nn.GELU, attention=True) -> None:\n",
    "        super().__init__()\n",
    "        self.block_1 = TimeResNetBlock(2 * out_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.block_2 = TimeResNetBlock(out_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.attention = LinearAttention(out_channels, 4, out_channels // 4) if attention else nn.Identity()\n",
    "        self.upsample = get_upsample(in_channels, out_channels)\n",
    "\n",
    "    def forward(self, x, h, t):\n",
    "        x = self.upsample(x)\n",
    "\n",
    "        # doing a residual connection by concatenation \n",
    "        # to feed the input image back in at a later stage\n",
    "        x = torch.cat([x,h], dim=1)\n",
    "        x = self.block_1(x, t)\n",
    "        x = self.block_2(x, t)\n",
    "        x = self.attention(x)\n",
    "        return x\n",
    "\n",
    "class MidBlock(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels, num_groups, base_time_dim, activation=nn.GELU) -> None:\n",
    "        super().__init__()\n",
    "        self.block_1 = TimeResNetBlock(in_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.block_2 = TimeResNetBlock(out_channels, out_channels, num_groups, base_time_dim, activation=activation)\n",
    "        self.attention = Attention(out_channels, 4, out_channels // 4)\n",
    "\n",
    "    def forward(self, x, t):\n",
    "        h = self.block_1(x, t)\n",
    "        h = self.block_2(h, t)\n",
    "        h = self.attention(h)\n",
    "        return h"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "class UNet(pl.LightningModule):\n",
    "    def __init__(self,\n",
    "        channel_list,\n",
    "        lr,\n",
    "        patience,\n",
    "        noise_schedule,\n",
    "        optim=torch.optim.Adam\n",
    "    ):\n",
    "        super().__init__()\n",
    "        self.save_hyperparameters()\n",
    "        self.lr = lr\n",
    "        self.patience = patience\n",
    "        self.noise_schedule = noise_schedule\n",
    "        \n",
    "        base_time_dim = max(channel_list)\n",
    "        self.time_embedding = BaseTimeEmbedding(base_time_dim)\n",
    "\n",
    "        num_layers = len(channel_list)\n",
    "        self.downsample = nn.ModuleList([\n",
    "            DownSampleBlock(channel_list[i], channel_list[i+1], num_groups=8, base_time_dim=base_time_dim) \n",
    "            for i in range(num_layers-2)\n",
    "            ])\n",
    "\n",
    "        self.upsample = nn.ModuleList([\n",
    "            UpSampleBlock(channel_list[num_layers-(i+1)], channel_list[num_layers-(i+2)], num_groups=8, base_time_dim=base_time_dim) \n",
    "            for i in range(num_layers-2)\n",
    "            ])\n",
    "        \n",
    "        self.middle = MidBlock(channel_list[-2], channel_list[-1], num_groups=8, base_time_dim=base_time_dim)\n",
    "        self.final_layer = nn.Conv2d(channel_list[1], channel_list[0], kernel_size=3, padding=1)\n",
    "\n",
    "    def forward(self, x, t):\n",
    "        t = self.time_embedding(t)\n",
    "        \n",
    "        # save hidden states for residual connections to upsampling\n",
    "        h_list = []\n",
    "        for layer in self.downsample:\n",
    "            x, h = layer(x, t)\n",
    "            h_list.append(h)\n",
    "\n",
    "        x = self.middle(x,t)\n",
    "\n",
    "        for layer in self.upsample:\n",
    "            h = h_list.pop()\n",
    "            x = layer(x, h, t)\n",
    "\n",
    "        out = self.final_layer(x)\n",
    "        return out\n",
    "\n",
    "    def training_step(self, x_0, _):\n",
    "        t, alpha_bar_t = self.noise_schedule.get_random_sample(x_0)\n",
    "        x_t, noise = noise_image(x_0, alpha_bar_t)\n",
    "        predicted_noise = self(x_t, t)\n",
    "        \n",
    "        loss = F.mse_loss(noise, predicted_noise)\n",
    "        self.log(\"train_loss\", loss)\n",
    "        return loss\n",
    "\n",
    "    def validation_step(self, x_0, _):\n",
    "        t, alpha_bar_t = self.noise_schedule.get_random_sample(x_0)\n",
    "        x_t, noise = noise_image(x_0, alpha_bar_t)\n",
    "        predicted_noise = self(x_t, t)\n",
    "        loss = F.mse_loss(noise, predicted_noise)\n",
    "\n",
    "        self.log(\"val_loss\", loss, prog_bar=True)\n",
    "        return loss\n",
    "\n",
    "    def configure_optimizers(self):\n",
    "        optimizer = torch.optim.Adam(self.parameters(), self.lr)\n",
    "        scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor=.2, patience=self.patience, verbose=True, min_lr=1e-6)\n",
    "        return {'optimizer':optimizer, 'lr_scheduler':scheduler, 'monitor':'val_loss'}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# split the data into training and validation data. We do not care about test data, since the model is tested by generating images\n",
    "val_percent = .1\n",
    "val_samples = int(len(dataset) * val_percent)\n",
    "train_samples = len(dataset) - val_samples\n",
    "\n",
    "train_dataset, val_dataset = random_split(dataset, [train_samples, val_samples])\n",
    "batch_size = 64\n",
    "train_loader = DataLoader(train_dataset, batch_size, shuffle=True)\n",
    "val_loader = DataLoader(val_dataset, batch_size)\n",
    "\n",
    "# alternatively for training on low end hardware (or cpus) we can use a single image of the dataset for training\n",
    "# to do this, just uncomment the next line\n",
    "#train_dataset = SingleImageData(dataset[0], epoch_len=5000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To monitor the training in Tensorboard or by saving image files we write a Callback to pass to the Pytorch Lightning Trainer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "class GenerateCallback(pl.Callback):\n",
    "    def __init__(self, num_examples, show_steps, every_n_epochs, img_size, channels, reverse_transform, plot_folder = None):\n",
    "        super().__init__()\n",
    "        self.num_examples = num_examples\n",
    "        self.show_steps = show_steps\n",
    "        self.img_size = img_size\n",
    "        self.reverse_transform = reverse_transform\n",
    "        self.channels=channels\n",
    "        self.every_n_epochs = every_n_epochs\n",
    "        self.plot_folder = plot_folder\n",
    "\n",
    "    def on_train_epoch_end(self, trainer, model):\n",
    "        if trainer.current_epoch % self.every_n_epochs == 0:\n",
    "            with torch.no_grad():\n",
    "                model.eval()\n",
    "                imgs = generate_images(model, self.num_examples, self.show_steps, self.img_size, self.channels)\n",
    "                imgs = self.reverse_transform(imgs)\n",
    "                grid = torchvision.utils.make_grid(imgs,nrow=self.show_steps+1)\n",
    "                model.train()\n",
    "            # Plot and add to tensorboard\n",
    "            trainer.logger.experiment.add_image(f\"Generated images\", grid, global_step=trainer.global_step)\n",
    "            if self.plot_folder is not None:\n",
    "                fig = plot_grid(imgs, self.show_steps+1)\n",
    "                fig.savefig(self.plot_folder + 'gen_imgs_epoch_%s.jpeg'%trainer.current_epoch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = UNet(channel_list=[1,32,64,128], lr=5e-4, patience=2, noise_schedule=noise_schedule)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
    "\n",
    "# if using a single image for training, remove the val_loader from the trainer and set num_examples=1\n",
    "generate_callback = GenerateCallback(\n",
    "    num_examples=10, \n",
    "    show_steps=5, \n",
    "    every_n_epochs=1, \n",
    "    img_size=IMG_SIZE, \n",
    "    channels=CHANNELS, \n",
    "    reverse_transform=rev_transform,\n",
    "    plot_folder='plots/'\n",
    ")\n",
    "\n",
    "trainer = pl.Trainer(\n",
    "    accelerator=device,\n",
    "    devices=1,\n",
    "    max_epochs=20,\n",
    "    callbacks=[generate_callback]\n",
    "    )\n",
    "trainer.fit(model, train_loader, val_loader)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can look at some more images after the training has finished:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_imgs = 10\n",
    "show_steps = 5\n",
    "\n",
    "model.eval()\n",
    "model.to(device)\n",
    "\n",
    "imgs = generate_images(model, num_imgs, show_steps, IMG_SIZE, channels=CHANNELS)\n",
    "imgs = rev_transform(imgs)\n",
    "plot_grid(imgs, show_steps + 1)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9.13 ('dml')",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.13"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "0df9bd60cd56d52c4f76f8cba168bd4f32969d0e9b16258ab68a4deb1d6f9b21"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}