sampler.py

import torch
import numpy as np

class DDPMSampler:

    def __init__(self, generator: torch.Generator, num_training_steps=1000, beta_start: float = 0.00085, beta_end: float = 0.0120):
        # Params "beta_start" and "beta_end" taken from: https://github.com/CompVis/stable-diffusion/blob/21f890f9da3cfbeaba8e2ac3c425ee9e998d5229/configs/stable-diffusion/v1-inference.yaml#L5C8-L5C8
        # For the naming conventions, refer to the DDPM paper (https://arxiv.org/pdf/2006.11239.pdf)
        self.betas = torch.linspace(beta_start ** 0.5, beta_end ** 0.5, num_training_steps, dtype=torch.float32) ** 2  #beta
        self.alphas = 1.0 - self.betas
        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)  # alpha bar
        self.one = torch.tensor(1.0)

        self.generator = generator

        self.num_train_timesteps = num_training_steps
        self.timesteps = torch.from_numpy(np.arange(0, num_training_steps)[::-1].copy()) ##[999, 998, ...0]

    def set_inference_timesteps(self, num_inference_steps=50):
      # num_inference_steps = 50
      # step ratio = num_training_steps // inference_steps = 20
        self.num_inference_steps = num_inference_steps
        step_ratio = self.num_train_timesteps // self.num_inference_steps # 1000/50 = 20
        timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.int64)  #[980, 960, ..0]
        self.timesteps = torch.from_numpy(timesteps)

    def _get_previous_timestep(self, timestep: int) -> int:
        prev_t = timestep - self.num_train_timesteps // self.num_inference_steps  #eg: t = 960, t-1 = 960-20 = 940
        return prev_t

    def _get_variance(self, timestep: int) -> torch.Tensor:
        prev_t = self._get_previous_timestep(timestep)  #t-1

        alpha_prod_t = self.alphas_cumprod[timestep] #alpha bar t
        alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one  #alpha bar t minus 1
        current_beta_t = 1 - alpha_prod_t / alpha_prod_t_prev  #beta t

        # For t > 0, compute predicted variance βt (see formula (6) and (7) from https://arxiv.org/pdf/2006.11239.pdf)
        # and sample from it to get previous sample
        # x_{t-1} ~ N(pred_prev_sample, variance) == add variance to pred_sample
        variance = (1 - alpha_prod_t_prev) / (1 - alpha_prod_t) * current_beta_t  #variance#

        # we always take the log of variance, so clamp it to ensure it's not 0
        variance = torch.clamp(variance, min=1e-20)

        return variance

    def set_strength(self, strength=1):
        """
            Set how much noise to add to the input image.
            More noise (strength ~ 1) means that the output will be further from the input image.
            Less noise (strength ~ 0) means that the output will be closer to the input image.
        """
        # more strength -> start step is approximately 0 that is model starts from pure noise and generates the image from it, strength = 1,  start step = 50 - (50 * 1) = 0
        # less strenght -> start step is skipped till 50 so model has the less noisified image a time step 50, model reconstructs the image from the less noisified image, strength = 0, start_step = 50

        # start_step is the number of noise levels to skip
        #eg inf_steps = 50, strength = 1,  start step = 50 - (50 * 1) = 0, strength = 0, start_step = 50
        start_step = self.num_inference_steps - int(self.num_inference_steps * strength)
        self.timesteps = self.timesteps[start_step:]  #skip time_steps, if start_step = 50 8#
        self.start_step = start_step  #50, in this case

    def step(self, timestep: int, latents: torch.Tensor, model_output: torch.Tensor):
        t = timestep #t
        prev_t = self._get_previous_timestep(t)  #t-1

        # 1. compute alphas, betas
        alpha_prod_t = self.alphas_cumprod[t] #alpha_bar_t
        alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one  #alpha_bar_t-1
        beta_prod_t = 1 - alpha_prod_t  #beta_bar_t
        beta_prod_t_prev = 1 - alpha_prod_t_prev #beta_bar_t-1
        current_alpha_t = alpha_prod_t / alpha_prod_t_prev  #alpha_t
        current_beta_t = 1 - current_alpha_t  #beta_t

        # 2. compute predicted original sample from predicted noise also called
        # "predicted x_0" of formula (15) from https://arxiv.org/pdf/2006.11239.pdf
        pred_original_sample = (latents - beta_prod_t ** (0.5) * model_output) / alpha_prod_t ** (0.5)  #x_0 - gaussian noise

        # 4. Compute coefficients for pred_original_sample x_0 and current sample x_t
        # See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
        pred_original_sample_coeff = (alpha_prod_t_prev ** (0.5) * current_beta_t) / beta_prod_t  #coeff_x_0
        current_sample_coeff = current_alpha_t ** (0.5) * beta_prod_t_prev / beta_prod_t  #coff_x_t

        # 5. Compute predicted previous sample µ_t
        # See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
        pred_prev_sample = pred_original_sample_coeff * pred_original_sample + current_sample_coeff * latents  #

        # 6. Add noise
        variance = 0
        if t > 0:
            device = model_output.device
            noise = torch.randn(model_output.shape, generator=self.generator, device=device, dtype=model_output.dtype)
            # Compute the variance as per formula (7) from https://arxiv.org/pdf/2006.11239.pdf
            variance = (self._get_variance(t) ** 0.5) * noise

        # sample from N(mu, sigma) = X can be obtained by X = mu + sigma * N(0, 1)
        # the variable "variance" is already multiplied by the noise N(0, 1)
        pred_prev_sample = pred_prev_sample + variance  #predicted xt-1

        return pred_prev_sample

    def add_noise(
        self,
        original_samples: torch.FloatTensor,
        timesteps: torch.IntTensor,
    ) -> torch.FloatTensor:
    #forward noisification
    #q(xt | x_not) = N(xt; sqrt(alpha_cumprod); (1 - alpha_cumprod)I)
        alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype)  #alpha_bar
        timesteps = timesteps.to(original_samples.device)

        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5  #sqrt(alpha_bar_t)
        sqrt_alpha_prod = sqrt_alpha_prod.flatten()  #flatten
        while len(sqrt_alpha_prod.shape) < len(original_samples.shape):  #for boardcasting
            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)

        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5  #sqrt(1 - alpha_bar_t)
        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
        while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape):
            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)

        # Sample from q(x_t | x_0) as in equation (4) of https://arxiv.org/pdf/2006.11239.pdf
        # Because N(mu, sigma) = X can be obtained by X = mu + sigma * N(0, 1)
        # here mu = sqrt_alpha_prod * original_samples and sigma = sqrt_one_minus_alpha_prod
        noise = torch.randn(original_samples.shape, generator=self.generator, device=original_samples.device, dtype=original_samples.dtype)  #noise
        noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise  #noisy samples
        return noisy_samples