"""SAMPLING ONLY.""" import torch import numpy as np from tqdm import tqdm from functools import partial from .diffusion_utils import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like class DDIMSampler(object): def __init__(self, model, schedule="linear", **kwargs): super().__init__() self.model = model self.ddpm_num_timesteps = model.num_timesteps self.schedule = schedule def register_buffer(self, name, attr): if type(attr) == torch.Tensor: if attr.device != torch.device("cuda"): attr = attr.to(torch.device("cuda")) setattr(self, name, attr) def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True): self.ddim_timesteps = 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.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. - alphas_cumprod.cpu()))) self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu()))) self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu()))) self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / 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,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. - 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, steps, shape, x_info, c_info, eta=0., temperature=1., noise_dropout=0., verbose=True, log_every_t=100,): self.make_schedule(ddim_num_steps=steps, ddim_eta=eta, verbose=verbose) print(f'Data shape for DDIM sampling is {shape}, eta {eta}') samples, intermediates = self.ddim_sampling( shape, x_info=x_info, c_info=c_info, noise_dropout=noise_dropout, temperature=temperature, log_every_t=log_every_t,) return samples, intermediates @torch.no_grad() def ddim_sampling(self, shape, x_info, c_info, noise_dropout=0., temperature=1., log_every_t=100,): device = self.model.device dtype = c_info['conditioning'].dtype bs = shape[0] timesteps = self.ddim_timesteps if ('xt' in x_info) and (x_info['xt'] is not None): xt = x_info['xt'].astype(dtype).to(device) x_info['x'] = xt elif ('x0' in x_info) and (x_info['x0'] is not None): x0 = x_info['x0'].type(dtype).to(device) ts = timesteps[x_info['x0_forward_timesteps']].repeat(bs) ts = torch.Tensor(ts).long().to(device) timesteps = timesteps[:x_info['x0_forward_timesteps']] x0_nz = self.model.q_sample(x0, ts) x_info['x'] = x0_nz else: x_info['x'] = torch.randn(shape, device=device, dtype=dtype) intermediates = {'pred_xt': [], 'pred_x0': []} time_range = np.flip(timesteps) total_steps = timesteps.shape[0] iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps) for i, step in enumerate(iterator): index = total_steps - i - 1 ts = torch.full((bs,), step, device=device, dtype=torch.long) outs = self.p_sample_ddim( x_info, c_info, ts, index, noise_dropout=noise_dropout, temperature=temperature,) pred_xt, pred_x0 = outs x_info['x'] = pred_xt if index % log_every_t == 0 or index == total_steps - 1: intermediates['pred_xt'].append(pred_xt) intermediates['pred_x0'].append(pred_x0) return pred_xt, intermediates @torch.no_grad() def p_sample_ddim(self, x_info, c_info, t, index, repeat_noise=False, use_original_steps=False, noise_dropout=0., temperature=1.,): x = x_info['x'] unconditional_guidance_scale = c_info['unconditional_guidance_scale'] b, *_, device = *x.shape, x.device if unconditional_guidance_scale == 1.: c_info['c'] = c_info['conditioning'] e_t = self.model.apply_model(x_info, t, c_info) else: x_in = torch.cat([x] * 2) t_in = torch.cat([t] * 2) c_in = torch.cat([c_info['unconditional_conditioning'], c_info['conditioning']]) x_info['x'] = x_in c_info['c'] = c_in e_t_uncond, e_t = self.model.apply_model(x_info, t_in, c_info).chunk(2) e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_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 extended_shape = [b] + [1]*(len(e_t.shape)-1) a_t = torch.full(extended_shape, alphas[index], device=device, dtype=x.dtype) a_prev = torch.full(extended_shape, alphas_prev[index], device=device, dtype=x.dtype) sigma_t = torch.full(extended_shape, sigmas[index], device=device, dtype=x.dtype) sqrt_one_minus_at = torch.full(extended_shape, sqrt_one_minus_alphas[index], device=device, dtype=x.dtype) # current prediction for x_0 pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t noise = sigma_t * noise_like(x, repeat_noise) * temperature if noise_dropout > 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 sample_multicontext(self, steps, shape, x_info, c_info_list, eta=0., temperature=1., noise_dropout=0., verbose=True, log_every_t=100,): self.make_schedule(ddim_num_steps=steps, ddim_eta=eta, verbose=verbose) print(f'Data shape for DDIM sampling is {shape}, eta {eta}') samples, intermediates = self.ddim_sampling_multicontext( shape, x_info=x_info, c_info_list=c_info_list, noise_dropout=noise_dropout, temperature=temperature, log_every_t=log_every_t,) return samples, intermediates @torch.no_grad() def ddim_sampling_multicontext(self, shape, x_info, c_info_list, noise_dropout=0., temperature=1., log_every_t=100,): device = self.model.device dtype = c_info_list[0]['conditioning'].dtype bs = shape[0] timesteps = self.ddim_timesteps if ('xt' in x_info) and (x_info['xt'] is not None): xt = x_info['xt'].astype(dtype).to(device) x_info['x'] = xt elif ('x0' in x_info) and (x_info['x0'] is not None): x0 = x_info['x0'].type(dtype).to(device) ts = timesteps[x_info['x0_forward_timesteps']].repeat(bs) ts = torch.Tensor(ts).long().to(device) timesteps = timesteps[:x_info['x0_forward_timesteps']] x0_nz = self.model.q_sample(x0, ts) x_info['x'] = x0_nz else: x_info['x'] = torch.randn(shape, device=device, dtype=dtype) intermediates = {'pred_xt': [], 'pred_x0': []} time_range = np.flip(timesteps) total_steps = timesteps.shape[0] iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps) for i, step in enumerate(iterator): index = total_steps - i - 1 ts = torch.full((bs,), step, device=device, dtype=torch.long) outs = self.p_sample_ddim_multicontext( x_info, c_info_list, ts, index, noise_dropout=noise_dropout, temperature=temperature,) pred_xt, pred_x0 = outs x_info['x'] = pred_xt if index % log_every_t == 0 or index == total_steps - 1: intermediates['pred_xt'].append(pred_xt) intermediates['pred_x0'].append(pred_x0) return pred_xt, intermediates @torch.no_grad() def p_sample_ddim_multicontext( self, x_info, c_info_list, t, index, repeat_noise=False, use_original_steps=False, noise_dropout=0., temperature=1.,): x = x_info['x'] b, *_, device = *x.shape, x.device unconditional_guidance_scale = None for c_info in c_info_list: if unconditional_guidance_scale is None: unconditional_guidance_scale = c_info['unconditional_guidance_scale'] else: assert unconditional_guidance_scale==c_info['unconditional_guidance_scale'], \ "A different unconditional guidance scale between different context is not allowed!" if unconditional_guidance_scale == 1.: c_info['c'] = c_info['conditioning'] else: c_in = torch.cat([c_info['unconditional_conditioning'], c_info['conditioning']]) c_info['c'] = c_in if unconditional_guidance_scale == 1.: e_t = self.model.apply_model_multicontext(x_info, t, c_info_list) else: x_in = torch.cat([x] * 2) t_in = torch.cat([t] * 2) x_info['x'] = x_in e_t_uncond, e_t = self.model.apply_model_multicontext(x_info, t_in, c_info_list).chunk(2) e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_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 extended_shape = [b] + [1]*(len(e_t.shape)-1) a_t = torch.full(extended_shape, alphas[index], device=device, dtype=x.dtype) a_prev = torch.full(extended_shape, alphas_prev[index], device=device, dtype=x.dtype) sigma_t = torch.full(extended_shape, sigmas[index], device=device, dtype=x.dtype) sqrt_one_minus_at = torch.full(extended_shape, sqrt_one_minus_alphas[index], device=device, dtype=x.dtype) # current prediction for x_0 pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t noise = sigma_t * noise_like(x, repeat_noise) * temperature if noise_dropout > 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