import pdb import torch import torch.nn as nn import torch.nn.functional as F import platform from .common import PositionalEncoding, enc_dec_mask, pad_audio from tqdm import tqdm class DiffusionSchedule(nn.Module): def __init__(self, num_steps, mode='linear', beta_1=1e-4, beta_T=0.02, s=0.008): super().__init__() if mode == 'linear': betas = torch.linspace(beta_1, beta_T, num_steps) elif mode == 'quadratic': betas = torch.linspace(beta_1 ** 0.5, beta_T ** 0.5, num_steps) ** 2 elif mode == 'sigmoid': betas = torch.sigmoid(torch.linspace(-5, 5, num_steps)) * (beta_T - beta_1) + beta_1 elif mode == 'cosine': steps = num_steps + 1 x = torch.linspace(0, num_steps, steps) alpha_bars = torch.cos(((x / num_steps) + s) / (1 + s) * torch.pi * 0.5) ** 2 alpha_bars = alpha_bars / alpha_bars[0] betas = 1 - (alpha_bars[1:] / alpha_bars[:-1]) betas = torch.clip(betas, 0.0001, 0.999) else: raise ValueError(f'Unknown diffusion schedule {mode}!') betas = torch.cat([torch.zeros(1), betas], dim=0) # Padding beta_0 = 0 alphas = 1 - betas log_alphas = torch.log(alphas) for i in range(1, log_alphas.shape[0]): # 1 to T log_alphas[i] += log_alphas[i - 1] alpha_bars = log_alphas.exp() sigmas_flex = torch.sqrt(betas) sigmas_inflex = torch.zeros_like(sigmas_flex) for i in range(1, sigmas_flex.shape[0]): sigmas_inflex[i] = ((1 - alpha_bars[i - 1]) / (1 - alpha_bars[i])) * betas[i] sigmas_inflex = torch.sqrt(sigmas_inflex) self.num_steps = num_steps self.register_buffer('betas', betas) self.register_buffer('alphas', alphas) self.register_buffer('alpha_bars', alpha_bars) self.register_buffer('sigmas_flex', sigmas_flex) self.register_buffer('sigmas_inflex', sigmas_inflex) def uniform_sample_t(self, batch_size): ts = torch.randint(1, self.num_steps + 1, (batch_size,)) return ts.tolist() def get_sigmas(self, t, flexibility=0): assert 0 <= flexibility <= 1 sigmas = self.sigmas_flex[t] * flexibility + self.sigmas_inflex[t] * (1 - flexibility) return sigmas class DitTalkingHead(nn.Module): def __init__(self, device='cuda', target="sample", architecture="decoder", motion_feat_dim=76, fps=25, n_motions=100, n_prev_motions=10, audio_model="hubert", feature_dim=512, n_diff_steps=500, diff_schedule="cosine", cfg_mode="incremental", guiding_conditions="audio,", audio_encoder_path=''): super().__init__() # Model parameters self.target = target # 预测原始图像还是预测噪声 self.architecture = architecture self.motion_feat_dim = motion_feat_dim # motion 特征维度 self.fps = fps self.n_motions = n_motions # 当前motion100个, window_length, T_w self.n_prev_motions = n_prev_motions # 前续motion 10个, T_p self.feature_dim = feature_dim # Audio encoder self.audio_model = audio_model if self.audio_model == 'wav2vec2': print("using wav2vec2 audio encoder ...") from .wav2vec2 import Wav2Vec2Model self.audio_encoder = Wav2Vec2Model.from_pretrained(audio_encoder_path) # wav2vec 2.0 weights initialization self.audio_encoder.feature_extractor._freeze_parameters() frozen_layers = [0, 1] for name, param in self.audio_encoder.named_parameters(): if name.startswith("feature_projection"): param.requires_grad = False if name.startswith("encoder.layers"): layer = int(name.split(".")[2]) if layer in frozen_layers: param.requires_grad = False elif self.audio_model == "wav2vec2_ori": from .wav2vec2 import Wav2Vec2Model self.audio_encoder = Wav2Vec2Model.from_pretrained(audio_encoder_path) # wav2vec 2.0 weights initialization self.audio_encoder.feature_extractor._freeze_parameters() elif self.audio_model == 'hubert': # 根据经验，hubert特征提取器效果更好 from .hubert import HubertModel # from hubert import HubertModel self.audio_encoder = HubertModel.from_pretrained(audio_encoder_path) self.audio_encoder.feature_extractor._freeze_parameters() # print("hubert-en: ", self.audio_encoder) frozen_layers = [0, 1] for name, param in self.audio_encoder.named_parameters(): if name.startswith("feature_projection"): param.requires_grad = False if name.startswith("encoder.layers"): layer = int(name.split(".")[2]) if layer in frozen_layers: param.requires_grad = False elif self.audio_model == 'hubert_zh': # 根据经验，hubert特征提取器效果更好 print("using hubert chinese") from .hubert import HubertModel # from hubert import HubertModel self.audio_encoder = HubertModel.from_pretrained(audio_encoder_path) self.audio_encoder.feature_extractor._freeze_parameters() frozen_layers = [0, 1] for name, param in self.audio_encoder.named_parameters(): if name.startswith("feature_projection"): param.requires_grad = False if name.startswith("encoder.layers"): layer = int(name.split(".")[2]) if layer in frozen_layers: param.requires_grad = False elif self.audio_model == 'hubert_zh_ori': # 根据经验，hubert特征提取器效果更好 print("using hubert chinese ori") from .hubert import HubertModel self.audio_encoder = HubertModel.from_pretrained(audio_encoder_path) self.audio_encoder.feature_extractor._freeze_parameters() else: raise ValueError(f'Unknown audio model {self.audio_model}!') if architecture == 'decoder': self.audio_feature_map = nn.Linear(768, feature_dim) self.start_audio_feat = nn.Parameter(torch.randn(1, self.n_prev_motions, feature_dim)) else: raise ValueError(f'Unknown architecture {architecture}!') self.start_motion_feat = nn.Parameter(torch.randn(1, self.n_prev_motions, self.motion_feat_dim)) # 1, 10, 76 # Diffusion model self.denoising_net = DenoisingNetwork(device=device, n_motions=self.n_motions, n_prev_motions=self.n_prev_motions, motion_feat_dim=self.motion_feat_dim, feature_dim=feature_dim) # diffusion schedule self.diffusion_sched = DiffusionSchedule(n_diff_steps, diff_schedule) # Classifier-free settings self.cfg_mode = cfg_mode guiding_conditions = guiding_conditions.split(',') if guiding_conditions else [] self.guiding_conditions = [cond for cond in guiding_conditions if cond in ['audio']] if 'audio' in self.guiding_conditions: audio_feat_dim = feature_dim self.null_audio_feat = nn.Parameter(torch.randn(1, 1, audio_feat_dim)) # 1, 1, 512 self.to(device) @property def device(self): return next(self.parameters()).device def forward(self, motion_feat, audio_or_feat, prev_motion_feat=None, prev_audio_feat=None, time_step=None, indicator=None): """ Args: motion_feat: (N, L, d_coef) motion coefficients or features audio_or_feat: (N, L_audio) raw audio or audio feature prev_motion_feat: (N, n_prev_motions, d_motion) previous motion coefficients or feature prev_audio_feat: (N, n_prev_motions, d_audio) previous audio features time_step: (N,) indicator: (N, L) 0/1 indicator of real (unpadded) motion coefficients Returns: motion_feat_noise: (N, L, d_motion) """ batch_size = motion_feat.shape[0] # 加载语音特征 if audio_or_feat.ndim == 2: # 原始语音 # Extract audio features assert audio_or_feat.shape[1] == 16000 * self.n_motions / self.fps, \ f'Incorrect audio length {audio_or_feat.shape[1]}' audio_feat_saved = self.extract_audio_feature(audio_or_feat) # (N, L, feature_dim) elif audio_or_feat.ndim == 3: # 语音特征 assert audio_or_feat.shape[1] == self.n_motions, f'Incorrect audio feature length {audio_or_feat.shape[1]}' audio_feat_saved = audio_or_feat else: raise ValueError(f'Incorrect audio input shape {audio_or_feat.shape}') audio_feat = audio_feat_saved.clone() # 前续motion特征 if prev_motion_feat is None: prev_motion_feat = self.start_motion_feat.expand(batch_size, -1, -1) # (N, n_prev_motions, d_motion) # 前续语音特征 if prev_audio_feat is None: # (N, n_prev_motions, feature_dim) prev_audio_feat = self.start_audio_feat.expand(batch_size, -1, -1) # Classifier-free guidance if len(self.guiding_conditions) > 0: assert len(self.guiding_conditions) <= 2, 'Only support 1 or 2 CFG conditions!' if len(self.guiding_conditions) == 1 or self.cfg_mode == 'independent': null_cond_prob = 0.5 if len(self.guiding_conditions) >= 2 else 0.1 if 'audio' in self.guiding_conditions: mask_audio = torch.rand(batch_size, device=self.device) < null_cond_prob audio_feat = torch.where(mask_audio.view(-1, 1, 1), self.null_audio_feat.expand(batch_size, self.n_motions, -1), audio_feat) else: # len(self.guiding_conditions) > 1 and self.cfg_mode == 'incremental' # full (0.45), w/o style (0.45), w/o style or audio (0.1) mask_flag = torch.rand(batch_size, device=self.device) if 'audio' in self.guiding_conditions: mask_audio = mask_flag > 0.9 audio_feat = torch.where(mask_audio.view(-1, 1, 1), self.null_audio_feat.expand(batch_size, self.n_motions, -1), audio_feat) if time_step is None: # Sample time step time_step = self.diffusion_sched.uniform_sample_t(batch_size) # (N,) # The forward diffusion process alpha_bar = self.diffusion_sched.alpha_bars[time_step] # (N,) c0 = torch.sqrt(alpha_bar).view(-1, 1, 1) # (N, 1, 1) c1 = torch.sqrt(1 - alpha_bar).view(-1, 1, 1) # (N, 1, 1) eps = torch.randn_like(motion_feat) # (N, L, d_motion) motion_feat_noisy = c0 * motion_feat + c1 * eps # The reverse diffusion process motion_feat_target = self.denoising_net(motion_feat_noisy, audio_feat, prev_motion_feat, prev_audio_feat, time_step, indicator) return eps, motion_feat_target, motion_feat.detach(), audio_feat_saved.detach() def extract_audio_feature(self, audio, frame_num=None): frame_num = frame_num or self.n_motions # # Strategy 1: resample during audio feature extraction # hidden_states = self.audio_encoder(pad_audio(audio), self.fps, frame_num=frame_num).last_hidden_state # (N, L, 768) # Strategy 2: resample after audio feature extraction (BackResample) hidden_states = self.audio_encoder(pad_audio(audio), self.fps, frame_num=frame_num * 2).last_hidden_state # (N, 2L, 768) hidden_states = hidden_states.transpose(1, 2) # (N, 768, 2L) hidden_states = F.interpolate(hidden_states, size=frame_num, align_corners=False, mode='linear') # (N, 768, L) hidden_states = hidden_states.transpose(1, 2) # (N, L, 768) audio_feat = self.audio_feature_map(hidden_states) # (N, L, feature_dim) return audio_feat @torch.no_grad() def sample(self, audio_or_feat, prev_motion_feat=None, prev_audio_feat=None, motion_at_T=None, indicator=None, cfg_mode=None, cfg_cond=None, cfg_scale=1.15, flexibility=0, dynamic_threshold=None, ret_traj=False): # Check and convert inputs batch_size = audio_or_feat.shape[0] # Check CFG conditions if cfg_mode is None: # Use default CFG mode cfg_mode = self.cfg_mode if cfg_cond is None: # Use default CFG conditions cfg_cond = self.guiding_conditions cfg_cond = [c for c in cfg_cond if c in ['audio', ]] if not isinstance(cfg_scale, list): cfg_scale = [cfg_scale] * len(cfg_cond) # sort cfg_cond and cfg_scale if len(cfg_cond) > 0: cfg_cond, cfg_scale = zip(*sorted(zip(cfg_cond, cfg_scale), key=lambda x: ['audio', ].index(x[0]))) else: cfg_cond, cfg_scale = [], [] if audio_or_feat.ndim == 2: # Extract audio features assert audio_or_feat.shape[1] == 16000 * self.n_motions / self.fps, \ f'Incorrect audio length {audio_or_feat.shape[1]}' audio_feat = self.extract_audio_feature(audio_or_feat) # (N, L, feature_dim) elif audio_or_feat.ndim == 3: assert audio_or_feat.shape[1] == self.n_motions, f'Incorrect audio feature length {audio_or_feat.shape[1]}' audio_feat = audio_or_feat else: raise ValueError(f'Incorrect audio input shape {audio_or_feat.shape}') if prev_motion_feat is None: prev_motion_feat = self.start_motion_feat.expand(batch_size, -1, -1) # (N, n_prev_motions, d_motion) if prev_audio_feat is None: # (N, n_prev_motions, feature_dim) prev_audio_feat = self.start_audio_feat.expand(batch_size, -1, -1) if motion_at_T is None: motion_at_T = torch.randn((batch_size, self.n_motions, self.motion_feat_dim)).to(self.device) # Prepare input for the reverse diffusion process (including optional classifier-free guidance) if 'audio' in cfg_cond: audio_feat_null = self.null_audio_feat.expand(batch_size, self.n_motions, -1) else: audio_feat_null = audio_feat audio_feat_in = [audio_feat_null] for cond in cfg_cond: if cond == 'audio': audio_feat_in.append(audio_feat) n_entries = len(audio_feat_in) audio_feat_in = torch.cat(audio_feat_in, dim=0) prev_motion_feat_in = torch.cat([prev_motion_feat] * n_entries, dim=0) prev_audio_feat_in = torch.cat([prev_audio_feat] * n_entries, dim=0) indicator_in = torch.cat([indicator] * n_entries, dim=0) if indicator is not None else None traj = {self.diffusion_sched.num_steps: motion_at_T} for t in tqdm(range(self.diffusion_sched.num_steps, 0, -1)): if t > 1: z = torch.randn_like(motion_at_T) else: z = torch.zeros_like(motion_at_T) alpha = self.diffusion_sched.alphas[t] alpha_bar = self.diffusion_sched.alpha_bars[t] alpha_bar_prev = self.diffusion_sched.alpha_bars[t - 1] sigma = self.diffusion_sched.get_sigmas(t, flexibility) motion_at_t = traj[t] motion_in = torch.cat([motion_at_t] * n_entries, dim=0) step_in = torch.tensor([t] * batch_size, device=self.device) step_in = torch.cat([step_in] * n_entries, dim=0) results = self.denoising_net(motion_in, audio_feat_in, prev_motion_feat_in, prev_audio_feat_in, step_in, indicator_in) # Apply thresholding if specified if dynamic_threshold: dt_ratio, dt_min, dt_max = dynamic_threshold abs_results = results[:, -self.n_motions:].reshape(batch_size * n_entries, -1).abs() s = torch.quantile(abs_results, dt_ratio, dim=1) s = torch.clamp(s, min=dt_min, max=dt_max) s = s[..., None, None] results = torch.clamp(results, min=-s, max=s) results = results.chunk(n_entries) # Unconditional target (CFG) or the conditional target (non-CFG) target_theta = results[0][:, -self.n_motions:] # Classifier-free Guidance (optional) for i in range(0, n_entries - 1): if cfg_mode == 'independent': target_theta += cfg_scale[i] * ( results[i + 1][:, -self.n_motions:] - results[0][:, -self.n_motions:]) elif cfg_mode == 'incremental': target_theta += cfg_scale[i] * ( results[i + 1][:, -self.n_motions:] - results[i][:, -self.n_motions:]) else: raise NotImplementedError(f'Unknown cfg_mode {cfg_mode}') if self.target == 'noise': c0 = 1 / torch.sqrt(alpha) c1 = (1 - alpha) / torch.sqrt(1 - alpha_bar) motion_next = c0 * (motion_at_t - c1 * target_theta) + sigma * z elif self.target == 'sample': c0 = (1 - alpha_bar_prev) * torch.sqrt(alpha) / (1 - alpha_bar) c1 = (1 - alpha) * torch.sqrt(alpha_bar_prev) / (1 - alpha_bar) motion_next = c0 * motion_at_t + c1 * target_theta + sigma * z else: raise ValueError('Unknown target type: {}'.format(self.target)) traj[t - 1] = motion_next.detach() # Stop gradient and save trajectory. traj[t] = traj[t].cpu() # Move previous output to CPU memory. if not ret_traj: del traj[t] if ret_traj: return traj, motion_at_T, audio_feat else: return traj[0], motion_at_T, audio_feat class DenoisingNetwork(nn.Module): def __init__(self, device='cuda', motion_feat_dim=76, use_indicator=None, architecture="decoder", feature_dim=512, n_heads=8, n_layers=8, mlp_ratio=4, align_mask_width=1, no_use_learnable_pe=True, n_prev_motions=10, n_motions=100, n_diff_steps=500, ): super().__init__() # Model parameters self.motion_feat_dim = motion_feat_dim self.use_indicator = use_indicator # Transformer self.architecture = architecture self.feature_dim = feature_dim self.n_heads = n_heads self.n_layers = n_layers self.mlp_ratio = mlp_ratio self.align_mask_width = align_mask_width self.use_learnable_pe = not no_use_learnable_pe # sequence length self.n_prev_motions = n_prev_motions self.n_motions = n_motions # Temporal embedding for the diffusion time step self.TE = PositionalEncoding(self.feature_dim, max_len=n_diff_steps + 1) self.diff_step_map = nn.Sequential( nn.Linear(self.feature_dim, self.feature_dim), nn.GELU(), nn.Linear(self.feature_dim, self.feature_dim) ) if self.use_learnable_pe: # Learnable positional encoding self.PE = nn.Parameter(torch.randn(1, 1 + self.n_prev_motions + self.n_motions, self.feature_dim)) else: self.PE = PositionalEncoding(self.feature_dim) # Transformer decoder if self.architecture == 'decoder': self.feature_proj = nn.Linear(self.motion_feat_dim + (1 if self.use_indicator else 0), self.feature_dim) decoder_layer = nn.TransformerDecoderLayer( d_model=self.feature_dim, nhead=self.n_heads, dim_feedforward=self.mlp_ratio * self.feature_dim, activation='gelu', batch_first=True ) self.transformer = nn.TransformerDecoder(decoder_layer, num_layers=self.n_layers) if self.align_mask_width > 0: motion_len = self.n_prev_motions + self.n_motions alignment_mask = enc_dec_mask(motion_len, motion_len, frame_width=1, expansion=self.align_mask_width - 1) # print(f"alignment_mask: ", alignment_mask.shape) # alignment_mask = F.pad(alignment_mask, (0, 0, 1, 0), value=False) self.register_buffer('alignment_mask', alignment_mask) else: self.alignment_mask = None else: raise ValueError(f'Unknown architecture: {self.architecture}') # Motion decoder self.motion_dec = nn.Sequential( nn.Linear(self.feature_dim, self.feature_dim // 2), nn.GELU(), nn.Linear(self.feature_dim // 2, self.motion_feat_dim), # nn.Tanh() # 增加了一个tanh # nn.Softmax() ) self.to(device) @property def device(self): return next(self.parameters()).device def forward(self, motion_feat, audio_feat, prev_motion_feat, prev_audio_feat, step, indicator=None): """ Args: motion_feat: (N, L, d_motion). Noisy motion feature audio_feat: (N, L, feature_dim) prev_motion_feat: (N, L_p, d_motion). Padded previous motion coefficients or feature prev_audio_feat: (N, L_p, d_audio). Padded previous motion coefficients or feature step: (N,) indicator: (N, L). 0/1 indicator for the real (unpadded) motion feature Returns: motion_feat_target: (N, L_p + L, d_motion) """ motion_feat = motion_feat.to(audio_feat.dtype) # Diffusion time step embedding diff_step_embedding = self.diff_step_map(self.TE.pe[0, step]).unsqueeze(1) # (N, 1, diff_step_dim) if indicator is not None: indicator = torch.cat([torch.zeros((indicator.shape[0], self.n_prev_motions), device=indicator.device), indicator], dim=1) # (N, L_p + L) indicator = indicator.unsqueeze(-1) # (N, L_p + L, 1) # Concat features and embeddings if self.architecture == 'decoder': # print("prev_motion_feat: ", prev_motion_feat.shape, "motion_feat: ", motion_feat.shape) feats_in = torch.cat([prev_motion_feat, motion_feat], dim=1) # (N, L_p + L, d_motion) else: raise ValueError(f'Unknown architecture: {self.architecture}') if self.use_indicator: feats_in = torch.cat([feats_in, indicator], dim=-1) # (N, L_p + L, d_motion + d_audio + 1) feats_in = self.feature_proj(feats_in) # (N, L_p + L, feature_dim) # feats_in = torch.cat([person_feat, feats_in], dim=1) # (N, 1 + L_p + L, feature_dim) if self.use_learnable_pe: # feats_in = feats_in + self.PE feats_in = feats_in + self.PE + diff_step_embedding else: # feats_in = self.PE(feats_in) feats_in = self.PE(feats_in) + diff_step_embedding # Transformer if self.architecture == 'decoder': audio_feat_in = torch.cat([prev_audio_feat, audio_feat], dim=1) # (N, L_p + L, d_audio) # print(f"feats_in: {feats_in.shape}, audio_feat_in: {audio_feat_in.shape}, memory_mask: {self.alignment_mask.shape}") feat_out = self.transformer(feats_in, audio_feat_in, memory_mask=self.alignment_mask) else: raise ValueError(f'Unknown architecture: {self.architecture}') # Decode predicted motion feature noise / sample # motion_feat_target = self.motion_dec(feat_out[:, 1:]) # (N, L_p + L, d_motion) motion_feat_target = self.motion_dec(feat_out) # (N, L_p + L, d_motion) return motion_feat_target if __name__ == "__main__": device = "cuda" motion_feat_dim = 76 n_motions = 100 # L n_prev_motions = 10 # L_p L_audio = int(16000 * n_motions / 25) # 64000 d_audio = 768 N = 5 feature_dim = 512 motion_feat = torch.ones((N, n_motions, motion_feat_dim)).to(device) prev_motion_feat = torch.ones((N, n_prev_motions, motion_feat_dim)).to(device) audio_or_feat = torch.ones((N, L_audio)).to(device) prev_audio_feat = torch.ones((N, n_prev_motions, d_audio)).to(device) time_step = torch.ones(N, dtype=torch.long).to(device) model = DitTalkingHead().to(device) z = model(motion_feat, audio_or_feat, prev_motion_feat=None, prev_audio_feat=None, time_step=None, indicator=None) traj, motion_at_T, audio_feat = z[0], z[1], z[2] print(motion_at_T.shape, audio_feat.shape)