TorchJaekwon/Model/Diffusion/Sampler/DDIM.py

from typing import Literal, Optional

from tqdm import tqdm
import numpy as np
import torch

from TorchJaekwon.Util.UtilTorch import UtilTorch
from TorchJaekwon.Model.Diffusion.DDPM.DDPM import DDPM
from TorchJaekwon.Model.Diffusion.DDPM.DiffusionUtil import DiffusionUtil

class DDIM(object):
    def __init__(self, ddpm_model:DDPM):
        self.ddpm_model:DDPM = ddpm_model
        self.ddpm_num_timesteps:int = ddpm_model.timesteps
        self.device:torch.device = UtilTorch.get_model_device(self.ddpm_model)

    def register_buffer(self, name, attr):
        if type(attr) == torch.Tensor:
            if attr.device != self.device:
                is_mps = self.device == "mps" or self.device == torch.device("mps")
                if is_mps and attr.dtype == torch.float64:
                    attr = attr.to(self.device, dtype=torch.float32)
                else:
                    attr = attr.to(self.device)
        setattr(self, name, attr)

    def make_schedule(
        self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0.0, verbose=True
    ):
        self.ddim_timesteps = DDIM.make_ddim_timesteps(
            ddim_discr_method=ddim_discretize,
            num_ddim_timesteps=ddim_num_steps,
            num_ddpm_timesteps=self.ddpm_num_timesteps,
            verbose=verbose,
        )
        alphas_cumprod = self.ddpm_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.device)

        self.register_buffer("betas", to_torch(self.ddpm_model.betas))
        self.register_buffer("alphas_cumprod", to_torch(alphas_cumprod))
        self.register_buffer(
            "alphas_cumprod_prev", to_torch(self.ddpm_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 = DDIM.make_ddim_sampling_parameters(
            alphacums=alphas_cumprod.cpu(),
            ddim_timesteps=self.ddim_timesteps,
            eta=ddim_eta,
            verbose=verbose,
        )
        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
        )
    
    @staticmethod
    def make_ddim_timesteps(
        ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True
    ):
        if ddim_discr_method == "uniform":
            c = num_ddpm_timesteps // num_ddim_timesteps
            ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
        elif ddim_discr_method == "quad":
            ddim_timesteps = (
                (np.linspace(0, np.sqrt(num_ddpm_timesteps * 0.8), num_ddim_timesteps)) ** 2
            ).astype(int)
        else:
            raise NotImplementedError(
                f'There is no ddim discretization method called "{ddim_discr_method}"'
            )

        # assert ddim_timesteps.shape[0] == num_ddim_timesteps
        # add one to get the final alpha values right (the ones from first scale to data during sampling)
        steps_out = ddim_timesteps + 1
        if verbose:
            print(f"Selected timesteps for ddim sampler: {steps_out}")
        return steps_out
    
    @staticmethod
    def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
        # select alphas for computing the variance schedule
        alphas = alphacums[ddim_timesteps]
        alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())

        # according the the formula provided in https://arxiv.org/abs/2010.02502
        sigmas = eta * np.sqrt(
            (1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev)
        )
        if verbose:
            print(
                f"Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}"
            )
            print(
                f"For the chosen value of eta, which is {eta}, "
                f"this results in the following sigma_t schedule for ddim sampler {sigmas}"
            )
        return sigmas, alphas, alphas_prev

    @torch.no_grad()
    def infer(
        self,
        x_shape:tuple = None,
        cond:Optional[dict] = None,
        is_cond_unpack:bool = False,
        num_steps:int = 50,
        batch_size:int = None,
        callback=None,
        normals_sequence=None,
        img_callback=None,
        quantize_x0=False,
        eta=1.0,
        mask=None,
        x0=None,
        temperature=1.0,
        noise_dropout=0.0,
        score_corrector=None,
        corrector_kwargs=None,
        verbose=False,
        x_T=None,
        log_every_t=100,
        unconditional_guidance_scale=1.0,
        dynamic_threshold=None,
        ucg_schedule=None,
    ):
        
        _, cond, additional_data_dict = self.ddpm_model.preprocess(x_start = None, cond=cond)
        if x_shape is None: x_shape = self.ddpm_model.get_x_shape(cond=cond)
        if batch_size is not None: x_shape[0] = batch_size

        self.make_schedule(ddim_num_steps=num_steps, ddim_eta=eta, verbose=verbose)

        samples, intermediates = self.ddim_sampling(
            x_shape,
            cond,
            is_cond_unpack,
            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,
            dynamic_threshold=dynamic_threshold,
            ucg_schedule=ucg_schedule,
        )
        return self.ddpm_model.postprocess(samples, additional_data_dict)

    @torch.no_grad()
    def ddim_sampling(
        self,
        x_shape:tuple = None,
        cond:Optional[dict] = None,
        is_cond_unpack:bool = False,
        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,
        dynamic_threshold=None,
        ucg_schedule=None,
    ):
        device = self.ddpm_model.betas.device
        b = x_shape[0]
        if x_T is None:
            img = torch.randn(x_shape, device=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]
        print(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.ddpm_model.q_sample(
                    x0, ts
                )  # TODO: deterministic forward pass?
                img = img_orig * mask + (1.0 - mask) * img

            if ucg_schedule is not None:
                assert len(ucg_schedule) == len(time_range)
                unconditional_guidance_scale = ucg_schedule[i]

            outs = self.p_sample_ddim(
                img,
                ts,
                index=index,
                cond = cond,
                is_cond_unpack = is_cond_unpack,
                use_original_steps=ddim_use_original_steps,
                quantize_denoised=quantize_denoised,
                temperature=temperature,
                noise_dropout=noise_dropout,
                score_corrector=score_corrector,
                corrector_kwargs=corrector_kwargs,
                unconditional_guidance_scale=unconditional_guidance_scale,
                unconditional_conditioning=unconditional_conditioning,
                dynamic_threshold=dynamic_threshold,
            )
            img, pred_x0 = outs
            if callback:
                callback(i)
            if img_callback:
                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,
        t,
        index,
        cond:Optional[dict] = None,
        is_cond_unpack:bool = False,
        repeat_noise=False,
        use_original_steps=False,
        quantize_denoised=False,
        temperature=1.0,
        noise_dropout=0.0,
        score_corrector=None,
        corrector_kwargs=None,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
        dynamic_threshold=None,
    ):
        b, *_, device = *x.shape, x.device
        
        model_output = self.ddpm_model.apply_model(x, t, cond, is_cond_unpack, cfg_scale = self.ddpm_model.cfg_scale)

        if self.ddpm_model.model_output_type == "v_prediction":
            e_t = self.ddpm_model.predict_noise_from_v(x, t, model_output)
        else:
            e_t = model_output

        if score_corrector is not None:
            assert self.ddpm_model.parameterization == "eps", "not implemented"
            e_t = score_corrector.modify_score(
                self.ddpm_model, e_t, x, t, c, **corrector_kwargs
            )

        alphas = self.ddpm_model.alphas_cumprod if use_original_steps else self.ddim_alphas
        alphas_prev = (
            self.ddpm_model.alphas_cumprod_prev
            if use_original_steps
            else self.ddim_alphas_prev
        )
        sqrt_one_minus_alphas = (
            self.ddpm_model.sqrt_one_minus_alphas_cumprod
            if use_original_steps
            else self.ddim_sqrt_one_minus_alphas
        )
        sigmas = (
            self.ddpm_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
        if self.ddpm_model.model_output_type != "v_prediction":
            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
        else:
            pred_x0 = self.ddpm_model.predict_x_start_from_v(x, t, model_output)

        if quantize_denoised:
            pred_x0, _, *_ = self.ddpm_model.first_stage_model.quantize(pred_x0)

        if dynamic_threshold is not None:
            raise NotImplementedError()

        # direction pointing to x_t
        dir_xt = (1.0 - a_prev - sigma_t**2).sqrt() * e_t
        noise = sigma_t * DiffusionUtil.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 encode(
        self,
        x0,
        c,
        t_enc,
        use_original_steps=False,
        return_intermediates=None,
        unconditional_guidance_scale=1.0,
        unconditional_conditioning=None,
        callback=None,
    ):
        num_reference_steps = (
            self.ddpm_num_timesteps
            if use_original_steps
            else self.ddim_timesteps.shape[0]
        )

        assert t_enc <= num_reference_steps
        num_steps = t_enc

        if use_original_steps:
            alphas_next = self.alphas_cumprod[:num_steps]
            alphas = self.alphas_cumprod_prev[:num_steps]
        else:
            alphas_next = self.ddim_alphas[:num_steps]
            alphas = torch.tensor(self.ddim_alphas_prev[:num_steps])

        x_next = x0
        intermediates = []
        inter_steps = []
        for i in tqdm(range(num_steps), desc="Encoding Image"):
            t = torch.full(
                (x0.shape[0],), i, device=self.ddpm_model.device, dtype=torch.long
            )
            if unconditional_guidance_scale == 1.0:
                noise_pred = self.ddpm_model.apply_model(x_next, t, c)
            else:
                assert unconditional_conditioning is not None
                e_t_uncond, noise_pred = torch.chunk(
                    self.ddpm_model.apply_model(
                        torch.cat((x_next, x_next)),
                        torch.cat((t, t)),
                        torch.cat((unconditional_conditioning, c)),
                    ),
                    2,
                )
                noise_pred = e_t_uncond + unconditional_guidance_scale * (
                    noise_pred - e_t_uncond
                )

            xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next
            weighted_noise_pred = (
                alphas_next[i].sqrt()
                * ((1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt())
                * noise_pred
            )
            x_next = xt_weighted + weighted_noise_pred
            if (
                return_intermediates
                and i % (num_steps // return_intermediates) == 0
                and i < num_steps - 1
            ):
                intermediates.append(x_next)
                inter_steps.append(i)
            elif return_intermediates and i >= num_steps - 2:
                intermediates.append(x_next)
                inter_steps.append(i)
            if callback:
                callback(i)

        out = {"x_encoded": x_next, "intermediate_steps": inter_steps}
        if return_intermediates:
            out.update({"intermediates": intermediates})
        return x_next, out

    @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,
        callback=None,
    ):
        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]
        print(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, _ = 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,
            )
            if callback:
                callback(i)
        return x_dec