imaginAIry/imaginairy/modules/diffusion/ddim.py

"""SAMPLING ONLY."""
import logging

import numpy as np
import torch
from tqdm import tqdm

from imaginairy.modules.diffusion.util import (
    extract_into_tensor,
    make_ddim_sampling_parameters,
    make_ddim_timesteps,
    noise_like,
)
from imaginairy.utils import get_device

logger = logging.getLogger(__name__)


class DDIMSampler:
    def __init__(self, model, schedule="linear", **kwargs):
        super().__init__()
        self.model = model
        self.ddpm_num_timesteps = model.num_timesteps
        self.schedule = schedule
        self.device_available = get_device()

    def register_buffer(self, name, attr):
        if type(attr) == torch.Tensor:
            if attr.device != torch.device(self.device_available):
                attr = attr.to(torch.float32).to(torch.device(self.device_available))
        setattr(self, name, attr)

    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0.0):
        self.ddim_timesteps = make_ddim_timesteps(
            ddim_discr_method=ddim_discretize,
            num_ddim_timesteps=ddim_num_steps,
            num_ddpm_timesteps=self.ddpm_num_timesteps,
        )
        alphas_cumprod = self.model.alphas_cumprod
        assert (
            alphas_cumprod.shape[0] == self.ddpm_num_timesteps
        ), "alphas have to be defined for each timestep"
        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)

        self.register_buffer("betas", to_torch(self.model.betas))
        self.register_buffer("alphas_cumprod", to_torch(alphas_cumprod))
        self.register_buffer(
            "alphas_cumprod_prev", to_torch(self.model.alphas_cumprod_prev)
        )

        # calculations for diffusion q(x_t | x_{t-1}) and others
        self.register_buffer(
            "sqrt_alphas_cumprod", to_torch(np.sqrt(alphas_cumprod.cpu()))
        )
        self.register_buffer(
            "sqrt_one_minus_alphas_cumprod",
            to_torch(np.sqrt(1.0 - alphas_cumprod.cpu())),
        )
        self.register_buffer(
            "log_one_minus_alphas_cumprod", to_torch(np.log(1.0 - alphas_cumprod.cpu()))
        )
        self.register_buffer(
            "sqrt_recip_alphas_cumprod", to_torch(np.sqrt(1.0 / alphas_cumprod.cpu()))
        )
        self.register_buffer(
            "sqrt_recipm1_alphas_cumprod",
            to_torch(np.sqrt(1.0 / alphas_cumprod.cpu() - 1)),
        )

        # ddim sampling parameters
        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(
            alphacums=alphas_cumprod.cpu(),
            ddim_timesteps=self.ddim_timesteps,
            eta=ddim_eta,
        )
        self.register_buffer("ddim_sigmas", ddim_sigmas)
        self.register_buffer("ddim_alphas", ddim_alphas)
        self.register_buffer("ddim_alphas_prev", ddim_alphas_prev)
        self.register_buffer("ddim_sqrt_one_minus_alphas", np.sqrt(1.0 - ddim_alphas))
        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
            (1 - self.alphas_cumprod_prev)
            / (1 - self.alphas_cumprod)
            * (1 - self.alphas_cumprod / self.alphas_cumprod_prev)
        )
        self.register_buffer(
            "ddim_sigmas_for_original_num_steps", sigmas_for_original_sampling_steps
        )

    @torch.no_grad()
    def sample(
        self,
        S,
        batch_size,
        shape,
        conditioning=None,
        callback=None,
        normals_sequence=None,
        img_callback=None,
        quantize_x0=False,
        eta=0.0,
        mask=None,
        x0=None,
        temperature=1.0,
        noise_dropout=0.0,
        score_corrector=None,
        corrector_kwargs=None,
        x_T=None,
        log_every_t=100,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
        # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
        **kwargs,
    ):
        if conditioning is not None:
            if isinstance(conditioning, dict):
                cbs = conditioning[list(conditioning.keys())[0]].shape[0]
                if cbs != batch_size:
                    logger.warning(
                        f"Warning: Got {cbs} conditionings but batch-size is {batch_size}"
                    )
            else:
                if conditioning.shape[0] != batch_size:
                    logger.warning(
                        f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}"
                    )

        self.make_schedule(ddim_num_steps=S, ddim_eta=eta)
        # sampling
        C, H, W = shape
        size = (batch_size, C, H, W)
        logger.debug(f"Data shape for DDIM sampling is {size}, eta {eta}")

        samples, intermediates = self.ddim_sampling(
            conditioning,
            size,
            callback=callback,
            img_callback=img_callback,
            quantize_denoised=quantize_x0,
            mask=mask,
            x0=x0,
            ddim_use_original_steps=False,
            noise_dropout=noise_dropout,
            temperature=temperature,
            score_corrector=score_corrector,
            corrector_kwargs=corrector_kwargs,
            x_T=x_T,
            log_every_t=log_every_t,
            unconditional_guidance_scale=unconditional_guidance_scale,
            unconditional_conditioning=unconditional_conditioning,
        )
        return samples, intermediates

    @torch.no_grad()
    def ddim_sampling(
        self,
        cond,
        shape,
        x_T=None,
        ddim_use_original_steps=False,
        callback=None,
        timesteps=None,
        quantize_denoised=False,
        mask=None,
        x0=None,
        img_callback=None,
        log_every_t=100,
        temperature=1.0,
        noise_dropout=0.0,
        score_corrector=None,
        corrector_kwargs=None,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
    ):
        device = self.model.betas.device
        b = shape[0]
        if x_T is None:
            # run on CPU for seed consistency. M1/mps runs were not consistent otherwise
            img = torch.randn(shape, device="cpu").to(device)
        else:
            img = x_T

        if timesteps is None:
            timesteps = (
                self.ddpm_num_timesteps
                if ddim_use_original_steps
                else self.ddim_timesteps
            )
        elif timesteps is not None and not ddim_use_original_steps:
            subset_end = (
                int(
                    min(timesteps / self.ddim_timesteps.shape[0], 1)
                    * self.ddim_timesteps.shape[0]
                )
                - 1
            )
            timesteps = self.ddim_timesteps[:subset_end]

        intermediates = {"x_inter": [img], "pred_x0": [img]}
        time_range = (
            reversed(range(0, timesteps))
            if ddim_use_original_steps
            else np.flip(timesteps)
        )
        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
        logger.info(f"Running DDIM Sampling with {total_steps} timesteps")

        iterator = tqdm(time_range, desc="DDIM Sampler", total=total_steps)

        for i, step in enumerate(iterator):
            index = total_steps - i - 1
            ts = torch.full((b,), step, device=device, dtype=torch.long)

            if mask is not None:
                assert x0 is not None
                img_orig = self.model.q_sample(
                    x0, ts
                )  # TODO: deterministic forward pass?
                img = img_orig * mask + (1.0 - mask) * img

            img, pred_x0 = self.p_sample_ddim(
                img,
                cond,
                ts,
                index=index,
                use_original_steps=ddim_use_original_steps,
                quantize_denoised=quantize_denoised,
                temperature=temperature,
                noise_dropout=noise_dropout,
                unconditional_guidance_scale=unconditional_guidance_scale,
                unconditional_conditioning=unconditional_conditioning,
            )
            if callback:
                callback(i)
            if img_callback:
                img_callback(pred_x0, i)
                img_callback(pred_x0, i)

            if index % log_every_t == 0 or index == total_steps - 1:
                intermediates["x_inter"].append(img)
                intermediates["pred_x0"].append(pred_x0)

        return img, intermediates

    # @torch.no_grad()
    def p_sample_ddim(
        self,
        x,
        c,
        t,
        index,
        repeat_noise=False,
        use_original_steps=False,
        quantize_denoised=False,
        temperature=1.0,
        noise_dropout=0.0,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
        loss_function=None,
    ):
        b, *_, device = *x.shape, x.device

        if unconditional_conditioning is None or unconditional_guidance_scale == 1.0:
            with torch.no_grad():
                noise_pred = self.model.apply_model(x, t, c)
        else:
            x_in = torch.cat([x] * 2)
            t_in = torch.cat([t] * 2)
            c_in = torch.cat([unconditional_conditioning, c])
            # with torch.no_grad():
            noise_pred_uncond, noise_pred = self.model.apply_model(
                x_in, t_in, c_in
            ).chunk(2)
            noise_pred = noise_pred_uncond + unconditional_guidance_scale * (
                noise_pred - noise_pred_uncond
            )

        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
        alphas_prev = (
            self.model.alphas_cumprod_prev
            if use_original_steps
            else self.ddim_alphas_prev
        )
        sqrt_one_minus_alphas = (
            self.model.sqrt_one_minus_alphas_cumprod
            if use_original_steps
            else self.ddim_sqrt_one_minus_alphas
        )
        sigmas = (
            self.model.ddim_sigmas_for_original_num_steps
            if use_original_steps
            else self.ddim_sigmas
        )
        # select parameters corresponding to the currently considered timestep
        a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
        a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
        sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
        sqrt_one_minus_at = torch.full(
            (b, 1, 1, 1), sqrt_one_minus_alphas[index], device=device
        )

        # current prediction for x_0
        pred_x0 = (x - sqrt_one_minus_at * noise_pred) / a_t.sqrt()
        if quantize_denoised:
            pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
        # direction pointing to x_t
        dir_xt = (1.0 - a_prev - sigma_t**2).sqrt() * noise_pred

        noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
        if noise_dropout > 0.0:
            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
        x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
        return x_prev, pred_x0

    @torch.no_grad()
    def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
        # fast, but does not allow for exact reconstruction
        # t serves as an index to gather the correct alphas
        if use_original_steps:
            sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
            sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
        else:
            sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
            sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas

        if noise is None:
            noise = torch.randn_like(x0)
        return (
            extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0
            + extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise
        )

    @torch.no_grad()
    def decode(
        self,
        x_latent,
        cond,
        t_start,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
        use_original_steps=False,
        img_callback=None,
        score_corrector=None,
        temperature=1.0,
    ):

        timesteps = (
            np.arange(self.ddpm_num_timesteps)
            if use_original_steps
            else self.ddim_timesteps
        )
        timesteps = timesteps[:t_start]

        time_range = np.flip(timesteps)
        total_steps = timesteps.shape[0]
        logger.debug(f"Running DDIM Sampling with {total_steps} timesteps")

        iterator = tqdm(time_range, desc="Decoding image", total=total_steps)
        x_dec = x_latent

        for i, step in enumerate(iterator):
            index = total_steps - i - 1
            ts = torch.full(
                (x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long
            )
            x_dec, pred_x0 = self.p_sample_ddim(
                x_dec,
                cond,
                ts,
                index=index,
                use_original_steps=use_original_steps,
                unconditional_guidance_scale=unconditional_guidance_scale,
                unconditional_conditioning=unconditional_conditioning,
                temperature=temperature,
            )
            # original_loss = ((x_dec - x_latent).abs().mean()*70)
            # sigma_t = torch.full((1, 1, 1, 1), self.ddim_sigmas[index], device=get_device())
            # x_dec = x_dec.detach() + (original_loss * 0.1) ** 2
            # cond_grad = -torch.autograd.grad(original_loss, x_dec)[0]
            # x_dec = x_dec.detach() + cond_grad * sigma_t ** 2
            ## x_dec_alt = x_dec + (original_loss * 0.1) ** 2
            if img_callback:
                img_callback(x_dec, f"x_dec {i}")
                img_callback(pred_x0, f"pred_x0 {i}")
        return x_dec