"""SAMPLING ONLY.""" import torch import numpy as np from tqdm import tqdm from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, extract_into_tensor import torch.nn.functional as F import cv2 # Gaussian blur def gaussian_blur_2d(img, kernel_size, sigma): ksize_half = (kernel_size - 1) * 0.5 x = torch.linspace(-ksize_half, ksize_half, steps=kernel_size) pdf = torch.exp(-0.5 * (x / sigma).pow(2)) x_kernel = pdf / pdf.sum() x_kernel = x_kernel.to(device=img.device, dtype=img.dtype) kernel2d = torch.mm(x_kernel[:, None], x_kernel[None, :]) kernel2d = kernel2d.expand(img.shape[-3], 1, kernel2d.shape[0], kernel2d.shape[1]) padding = [kernel_size // 2, kernel_size // 2, kernel_size // 2, kernel_size // 2] img = F.pad(img, padding, mode="reflect") img = F.conv2d(img, kernel2d, groups=img.shape[-3]) return img # processes and stores attention probabilities class CrossAttnStoreProcessor: def __init__(self): self.attention_probs = None def __call__( self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None, ): batch_size, sequence_length, _ = hidden_states.shape attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size) query = attn.to_q(hidden_states) if encoder_hidden_states is None: encoder_hidden_states = hidden_states elif attn.norm_cross: encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states) key = attn.to_k(encoder_hidden_states) value = attn.to_v(encoder_hidden_states) query = attn.head_to_batch_dim(query) key = attn.head_to_batch_dim(key) value = attn.head_to_batch_dim(value) self.attention_probs = attn.get_attention_scores(query, key, attention_mask) hidden_states = torch.bmm(self.attention_probs, value) hidden_states = attn.batch_to_head_dim(hidden_states) # linear proj hidden_states = attn.to_out[0](hidden_states) # dropout hidden_states = attn.to_out[1](hidden_states) return hidden_states 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, S, batch_size, shape, conditioning=None, callback=None, normals_sequence=None, img_callback=None, quantize_x0=False, eta=0., mask=None, masked_image_latents=None, x0=None, temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, verbose=True, x_T=None, log_every_t=100, unconditional_guidance_scale=1., sag_scale=0.75, SAG_influence_step=600, noise = None, unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... dynamic_threshold=None, ucg_schedule=None, **kwargs ): if conditioning is not None: if isinstance(conditioning, dict): ctmp = conditioning[list(conditioning.keys())[0]] while isinstance(ctmp, list): ctmp = ctmp[0] cbs = ctmp.shape[0] if cbs != batch_size: print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") elif isinstance(conditioning, list): for ctmp in conditioning: if ctmp.shape[0] != batch_size: print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") else: if conditioning.shape[0] != batch_size: print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}") self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose) # sampling C, H, W = shape size = (batch_size, C, H, W) print(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,masked_image_latents=masked_image_latents, 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, sag_scale = sag_scale, SAG_influence_step = SAG_influence_step, noise = noise, unconditional_conditioning=unconditional_conditioning, dynamic_threshold=dynamic_threshold, ucg_schedule=ucg_schedule ) return samples, intermediates def add_noise(self, original_samples: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor, ) -> torch.FloatTensor: betas = torch.linspace(0.00085, 0.0120, 1000, dtype=torch.float32) alphas = 1.0 - betas alphas_cumprod = torch.cumprod(alphas, dim=0) alphas_cumprod = alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype) timesteps = timesteps.to(original_samples.device) sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5 sqrt_alpha_prod = sqrt_alpha_prod.flatten() while len(sqrt_alpha_prod.shape) < len(original_samples.shape): sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1) sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5 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) noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise return noisy_samples # def add_noise( # self, # original_samples: torch.FloatTensor, # noise: torch.FloatTensor, # timesteps: torch.FloatTensor, # sigma_t, # ) -> torch.FloatTensor: # # Make sure sigmas and timesteps have the same device and dtype as original_samples # sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype) # if original_samples.device.type == "mps" and torch.is_floating_point(timesteps): # # mps does not support float64 # schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32) # timesteps = timesteps.to(original_samples.device, dtype=torch.float32) # else: # schedule_timesteps = self.timesteps.to(original_samples.device) # timesteps = timesteps.to(original_samples.device) # step_indices = [(schedule_timesteps == t).nonzero().item() for t in timesteps] # sigma = sigmas[step_indices].flatten() # while len(sigma.shape) < len(original_samples.shape): # sigma = sigma.unsqueeze(-1) # # print(sigma_t) # noisy_samples = original_samples + noise * sigma_t # return noisy_samples def sag_masking(self, original_latents,model_output,x, attn_map, map_size, t, eps): # Same masking process as in SAG paper: https://arxiv.org/pdf/2210.00939.pdf bh, hw1, hw2 = attn_map.shape b, latent_channel, latent_h, latent_w = original_latents.shape h = 4 #self.unet.config.attention_head_dim if isinstance(h, list): h = h[-1] # print(attn_map.shape) # print(original_latents.shape) # print(map_size) # Produce attention mask attn_map = attn_map.reshape(b, h, hw1, hw2) attn_mask = attn_map.mean(1, keepdim=False).sum(1, keepdim=False) > 1.0 # print(attn_mask.shape) attn_mask = ( attn_mask.reshape(b, map_size[0], map_size[1]) .unsqueeze(1) .repeat(1, latent_channel, 1, 1) .type(attn_map.dtype) ) attn_mask = F.interpolate(attn_mask, (latent_h, latent_w)) # print(attn_mask.shape) # cv2.imwrite("attn_mask.png",attn_mask) # Blur according to the self-attention mask degraded_latents = gaussian_blur_2d(original_latents, kernel_size=9, sigma=1.0) # degraded_latents = self.add_noise(degraded_latents, noise=eps, timesteps=t)#,sigma_t=sigma_t) degraded_latents = degraded_latents * attn_mask + original_latents * (1 - attn_mask) #x#original_latents # degraded_latents = self.model.get_x_t_from_start_and_t(degraded_latents,t,model_output) # print(original_latents.shape) # print(eps.shape) # Noise it again to match the noise level # print("t",t) # degraded_latents = self.add_noise(degraded_latents, noise=eps, timesteps=t)#,sigma_t=sigma_t) return degraded_latents def pred_epsilon(self, sample, model_output, timestep): alpha_prod_t = timestep beta_prod_t = 1 - alpha_prod_t # print(self.model.parameterization)#eps if self.model.parameterization == "eps": pred_eps = model_output elif self.model.parameterization == "sample": pred_eps = (sample - (alpha_prod_t**0.5) * model_output) / (beta_prod_t**0.5) elif self.model.parameterization == "v": pred_eps = (beta_prod_t**0.5) * sample + (alpha_prod_t**0.5) * model_output else: raise ValueError( f"prediction_type given as {self.scheduler.config.prediction_type} must be one of `eps`, `sample`," " or `v`" ) return pred_eps @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,masked_image_latents=None, x0=None, img_callback=None, log_every_t=100, temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, unconditional_guidance_scale=1.,sag_scale = 0.75, SAG_influence_step=600, sag_enable = True, noise = None, unconditional_conditioning=None, dynamic_threshold=None, ucg_schedule=None): device = self.model.betas.device b = shape[0] if x_T is None: img = torch.randn(shape, device=device) else: img = x_T # timesteps =100 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] # timesteps=timesteps[:-3] # print("timesteps",timesteps) 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): print(step) if step > SAG_influence_step: sag_enable_t=True else: sag_enable_t=False 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. - 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,mask,masked_image_latents, cond, ts, index=index, 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, sag_scale = sag_scale, sag_enable=sag_enable_t, noise =noise, 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,mask,masked_image_latents, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False, temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, unconditional_guidance_scale=1.,sag_scale = 0.75, sag_enable=True, noise=None, unconditional_conditioning=None, dynamic_threshold=None): b, *_, device = *x.shape, x.device # map_size = None # def get_map_size(module, input, output): # nonlocal map_size # map_size = output.shape[-2:] # store_processor = CrossAttnStoreProcessor() # for name, param in self.model.model.diffusion_model.named_parameters(): # print(name) # self.model.control_model.middle_block[1].transformer_blocks[0].attn1.processor = store_processor # print(self.model.model.diffusion_model.middle_block[1].transformer_blocks[0].attn1) # self.model.model.diffusion_model.middle_block[1].transformer_blocks[0].attn1 = store_processor # with self.model.model.diffusion_model.middle_block[1].register_forward_hook(get_map_size): if unconditional_conditioning is None or unconditional_guidance_scale == 1.: model_output = self.model.apply_model(x,mask,masked_image_latents, t, c) else: model_t = self.model.apply_model(x,mask,masked_image_latents, t, c) model_uncond = self.model.apply_model(x,mask,masked_image_latents, t, unconditional_conditioning) model_output = model_uncond + unconditional_guidance_scale * (model_t - model_uncond) if self.model.parameterization == "v": e_t = self.model.predict_eps_from_z_and_v(x, t, model_output) else: e_t = model_output if score_corrector is not None: assert self.model.parameterization == "eps", 'not implemented' e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs) 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 if self.model.parameterization != "v": pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() else: pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output) if quantize_denoised: pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) if dynamic_threshold is not None: raise NotImplementedError() if sag_enable == True: uncond_attn, cond_attn = self.model.model.diffusion_model.middle_block[1].transformer_blocks[0].attn1.attention_probs.chunk(2) # self-attention-based degrading of latents map_size = self.model.model.diffusion_model.middle_block[1].map_size degraded_latents = self.sag_masking( pred_x0,model_output,x,uncond_attn, map_size, t, eps = noise, #self.pred_epsilon(x, model_uncond, self.model.alphas_cumprod[t]),#noise ) if unconditional_conditioning is None or unconditional_guidance_scale == 1.: degraded_model_output = self.model.apply_model(degraded_latents,mask,masked_image_latents, t, c) else: degraded_model_t = self.model.apply_model(degraded_latents,mask,masked_image_latents, t, c) degraded_model_uncond = self.model.apply_model(degraded_latents,mask,masked_image_latents, t, unconditional_conditioning) degraded_model_output = degraded_model_uncond + unconditional_guidance_scale * (degraded_model_t - degraded_model_uncond) # print("sag_scale",sag_scale) model_output += sag_scale * (model_output - degraded_model_output) # model_output = (1-sag_scale) * model_output + sag_scale * degraded_model_output # current prediction for x_0 if self.model.parameterization != "v": pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() else: pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output) if quantize_denoised: pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) if dynamic_threshold is not None: raise NotImplementedError() # direction pointing to x_t dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t noise = sigma_t * noise_like(x.shape, device, 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 encode(self, x0, c, t_enc, use_original_steps=False, return_intermediates=None, unconditional_guidance_scale=1.0, unconditional_conditioning=None, callback=None): timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps num_reference_steps = 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],), timesteps[i], device=self.model.device, dtype=torch.long) if unconditional_guidance_scale == 1.: noise_pred = self.model.apply_model(x_next, t, c) else: assert unconditional_conditioning is not None e_t_uncond, noise_pred = torch.chunk( self.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