File size: 15,608 Bytes
39e9e02 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 | import numpy as np
import torch
import torch.nn as nn
import math
from functools import lru_cache
from torch.utils.checkpoint import checkpoint
def modulate(x, shift, scale=None):
if shift is None:
return x * (1 + scale)
return x * (1 + scale) + shift
class RMSNorm(nn.Module):
def __init__(self, dim: int, eps: float = 1e-5):
super().__init__()
self.eps = eps
self.weight = nn.Parameter(torch.ones(dim))
def forward(self, x: torch.Tensor) -> torch.Tensor:
output = x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
return output * self.weight
class TimestepEmbedder(nn.Module):
"""
Embeds scalar timesteps into vector representations.
"""
def __init__(self, hidden_size, frequency_embedding_size=256):
super().__init__()
self.mlp = nn.Sequential(
nn.Linear(frequency_embedding_size, hidden_size, bias=True),
nn.SiLU(),
nn.Linear(hidden_size, hidden_size, bias=True),
)
self.frequency_embedding_size = frequency_embedding_size
@staticmethod
def timestep_embedding(t: torch.Tensor, dim: int, max_period: float = 10000.0):
"""
Create sinusoidal timestep embeddings.
:param t: a 1-D Tensor of N indices, one per batch element. These may be fractional.
:param dim: the dimension of the output.
:param max_period: controls the minimum frequency of the embeddings.
:return: an (N, D) Tensor of positional embeddings.
"""
# https://github.com/openai/glide-text2im/blob/main/glide_text2im/nn.py
half = dim // 2
freqs = torch.exp(-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half).to(
device=t.device
)
args = t[:, None].float() * freqs[None]
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
if dim % 2:
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
return embedding
def forward(self, t):
t_freq = self.timestep_embedding(t, self.frequency_embedding_size)
t_emb = self.mlp(t_freq.to(self.mlp[0].weight.dtype))
return t_emb
class ResBlock(nn.Module):
def __init__(self, channels, mlp_ratio=1.0):
super().__init__()
self.channels = channels
self.intermediate_size = int(channels * mlp_ratio)
self.in_ln = nn.LayerNorm(self.channels, eps=1e-6)
self.mlp = nn.Sequential(
nn.Linear(self.channels, self.intermediate_size),
nn.SiLU(),
nn.Linear(self.intermediate_size, self.channels),
)
self.adaLN_modulation = nn.Sequential(nn.SiLU(), nn.Linear(channels, 3 * channels, bias=True))
def forward(self, x, y):
shift_mlp, scale_mlp, gate_mlp = self.adaLN_modulation(y).chunk(3, dim=-1)
h = modulate(self.in_ln(x), shift_mlp, scale_mlp)
h = self.mlp(h)
return x + gate_mlp * h
# class FinalLayer(nn.Module):
# def __init__(self, model_channels, out_channels):
# super().__init__()
# self.norm_final = nn.LayerNorm(model_channels, elementwise_affine=False, eps=1e-6)
# self.linear = nn.Linear(model_channels, out_channels, bias=True)
# self.adaLN_modulation = nn.Sequential(nn.SiLU(), nn.Linear(model_channels, 2 * model_channels, bias=True))
# def forward(self, x, c):
# shift, scale = self.adaLN_modulation(c).chunk(2, dim=-1)
# x = modulate(self.norm_final(x), shift, scale)
# x = self.linear(x)
# return x
# class SimpleMLPAdaLN(nn.Module):
# def __init__(self, input_dim, out_dim, dim=1536, layers=12, mlp_ratio=1.0):
# super().__init__()
# self.input_dim = input_dim
# self.out_dim = out_dim
# self.dim = dim
# self.layers = layers
# self.mlp_ratio = mlp_ratio
# self.time_embed = TimestepEmbedder(dim)
# self.input_proj = nn.Linear(input_dim, dim)
# res_blocks = []
# for _ in range(layers):
# res_blocks.append(ResBlock(dim, mlp_ratio))
# self.res_blocks = nn.ModuleList(res_blocks)
# self.final_layer = FinalLayer(dim, out_dim)
# self.grad_checkpointing = False
# self.initialize_weights()
# def initialize_weights(self):
# def _basic_init(module):
# if isinstance(module, nn.Linear):
# torch.nn.init.xavier_uniform_(module.weight)
# if module.bias is not None:
# nn.init.constant_(module.bias, 0)
# self.apply(_basic_init)
# # Initialize timestep embedding MLP
# nn.init.normal_(self.time_embed.mlp[0].weight, std=0.02)
# nn.init.normal_(self.time_embed.mlp[2].weight, std=0.02)
# # Zero-out adaLN modulation layers
# for block in self.res_blocks:
# nn.init.constant_(block.adaLN_modulation[-1].weight, 0)
# nn.init.constant_(block.adaLN_modulation[-1].bias, 0)
# # Zero-out output layers
# nn.init.constant_(self.final_layer.adaLN_modulation[-1].weight, 0)
# nn.init.constant_(self.final_layer.adaLN_modulation[-1].bias, 0)
# nn.init.constant_(self.final_layer.linear.weight, 0)
# nn.init.constant_(self.final_layer.linear.bias, 0)
# def forward(self, x, t):
# """
# x.shape = (bsz, input_dim)
# t.shape = (bsz,)
# """
# x = self.input_proj(x)
# t = self.time_embed(t)
# y = t
# for block in self.res_blocks:
# if self.grad_checkpointing and self.training:
# x = checkpoint(block, x, y, use_reentrant=True)
# else:
# x = block(x, y)
# return self.final_layer(x, y)
class FlowMatchingHead(nn.Module):
def __init__(self, input_dim, out_dim, dim=1536, layers=12, mlp_ratio=1.0):
super(FlowMatchingHead, self).__init__()
self.net = SimpleMLPAdaLN(input_dim=input_dim, out_dim=out_dim, dim=dim, layers=layers, mlp_ratio=mlp_ratio)
@property
def dtype(self):
return self.net.input_proj.weight.dtype
@property
def device(self):
return self.net.input_proj.weight.device
def forward(self, x, t):
x = self.net(x, t)
return x
def precompute_freqs_cis_2d(dim: int, height: int, width:int, theta: float = 10000.0, scale=16.0):
# assert H * H == end
# flat_patch_pos = torch.linspace(-1, 1, end) # N = end
x_pos = torch.linspace(0, scale, width)
y_pos = torch.linspace(0, scale, height)
y_pos, x_pos = torch.meshgrid(y_pos, x_pos, indexing="ij")
y_pos = y_pos.reshape(-1)
x_pos = x_pos.reshape(-1)
freqs = 1.0 / (theta ** (torch.arange(0, dim, 4)[: (dim // 4)].float() / dim)) # Hc/4
x_freqs = torch.outer(x_pos, freqs).float() # N Hc/4
y_freqs = torch.outer(y_pos, freqs).float() # N Hc/4
x_cis = torch.polar(torch.ones_like(x_freqs), x_freqs)
y_cis = torch.polar(torch.ones_like(y_freqs), y_freqs)
freqs_cis = torch.cat([x_cis.unsqueeze(dim=-1), y_cis.unsqueeze(dim=-1)], dim=-1) # N,Hc/4,2
freqs_cis = freqs_cis.reshape(height*width, -1)
return freqs_cis
class NerfEmbedder(nn.Module):
def __init__(self, in_channels, hidden_size_input, max_freqs):
super().__init__()
self.max_freqs = max_freqs
self.hidden_size_input = hidden_size_input
self.embedder = nn.Sequential(
nn.Linear(in_channels+max_freqs**2, hidden_size_input, bias=True),
)
@lru_cache
def fetch_pos(self, patch_size, device, dtype):
pos = precompute_freqs_cis_2d(self.max_freqs ** 2 * 2, patch_size, patch_size).real
pos = pos[None, :, :].to(device=device, dtype=dtype)
return pos
def forward(self, inputs):
B, P2, C = inputs.shape
patch_size = int(P2 ** 0.5)
device = inputs.device
dtype = inputs.dtype
dct = self.fetch_pos(patch_size, device, dtype)
dct = dct.repeat(B, 1, 1)
inputs = torch.cat([inputs, dct], dim=-1)
inputs = self.embedder(inputs)
return inputs
class SimpleMLPAdaLN(nn.Module):
"""
The MLP for Diffusion Loss.
:param in_channels: channels in the input Tensor.
:param model_channels: base channel count for the model.
:param out_channels: channels in the output Tensor.
:param z_channels: channels in the condition.
:param num_res_blocks: number of residual blocks per downsample.
"""
def __init__(
self,
in_channels,
model_channels,
out_channels,
z_channels,
num_res_blocks,
patch_size,
grad_checkpointing=False
):
super().__init__()
self.in_channels = in_channels
self.model_channels = model_channels
self.out_channels = out_channels
self.num_res_blocks = num_res_blocks
self.grad_checkpointing = grad_checkpointing
self.patch_size = patch_size
self.cond_embed = nn.Linear(z_channels, patch_size**2*model_channels)
self.input_proj = nn.Linear(in_channels, model_channels)
res_blocks = []
for i in range(num_res_blocks):
res_blocks.append(ResBlock(
model_channels,
))
self.res_blocks = nn.ModuleList(res_blocks)
self.final_layer = FinalLayer(model_channels, out_channels)
self.initialize_weights()
def initialize_weights(self):
def _basic_init(module):
if isinstance(module, nn.Linear):
torch.nn.init.xavier_uniform_(module.weight)
if module.bias is not None:
nn.init.constant_(module.bias, 0)
self.apply(_basic_init)
# Zero-out adaLN modulation layers
for block in self.res_blocks:
nn.init.constant_(block.adaLN_modulation[-1].weight, 0)
nn.init.constant_(block.adaLN_modulation[-1].bias, 0)
# Zero-out output layers
nn.init.constant_(self.final_layer.linear.weight, 0)
nn.init.constant_(self.final_layer.linear.bias, 0)
def forward(self, x, c):
"""
Apply the model to an input batch.
:param x: an [N x C] Tensor of inputs.
:param t: a 1-D batch of timesteps.
:param c: conditioning from AR transformer.
:return: an [N x C] Tensor of outputs.
"""
x = self.input_proj(x)
c = self.cond_embed(c)
y = c.reshape(-1, self.patch_size**2, self.model_channels)
for block in self.res_blocks:
x = block(x, y)
return self.final_layer(x)
class FinalLayer(nn.Module):
"""
The final layer adopted from DiT.
"""
def __init__(self, model_channels, out_channels):
super().__init__()
self.norm_final = nn.LayerNorm(model_channels, elementwise_affine=False, eps=1e-6)
self.linear = nn.Linear(model_channels, out_channels, bias=True)
def forward(self, x):
x = self.norm_final(x)
x = self.linear(x)
return x
#################################################################################
# Sine/Cosine Positional Embedding Functions #
#################################################################################
# https://github.com/facebookresearch/mae/blob/main/util/pos_embed.py
def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False, extra_tokens=0, pe_interpolation=1.0):
"""
grid_size: int of the grid height and width
return:
pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
"""
grid_h = np.arange(grid_size, dtype=np.float32) / pe_interpolation
grid_w = np.arange(grid_size, dtype=np.float32) / pe_interpolation
grid = np.meshgrid(grid_w, grid_h) # here w goes first
grid = np.stack(grid, axis=0)
grid = grid.reshape([2, 1, grid_size, grid_size])
pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
if cls_token and extra_tokens > 0:
pos_embed = np.concatenate([np.zeros([extra_tokens, embed_dim]), pos_embed], axis=0)
return pos_embed
def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
assert embed_dim % 2 == 0
# use half of dimensions to encode grid_h
emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0]) # (H*W, D/2)
emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1]) # (H*W, D/2)
emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
return emb
def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
"""
embed_dim: output dimension for each position
pos: a list of positions to be encoded: size (M,)
out: (M, D)
"""
assert embed_dim % 2 == 0
omega = np.arange(embed_dim // 2, dtype=np.float64)
omega /= embed_dim / 2.0
omega = 1.0 / 10000**omega # (D/2,)
pos = pos.reshape(-1) # (M,)
out = np.einsum("m,d->md", pos, omega) # (M, D/2), outer product
emb_sin = np.sin(out) # (M, D/2)
emb_cos = np.cos(out) # (M, D/2)
emb = np.concatenate([emb_sin, emb_cos], axis=1) # (M, D)
return emb
# --------------------------------------------------------
# Interpolate position embeddings for high-resolution
# References:
# DeiT: https://github.com/facebookresearch/deit
# --------------------------------------------------------
def interpolate_pos_embed(model_path, pe_key: str = "gen_pos_embed", new_len: int = 4096):
state_dict = torch.load(model_path, map_location="cpu")
pos_embed_1d = state_dict[pe_key]
_, ori_len, embed_dim = pos_embed_1d.shape
ori_size = int(ori_len**0.5)
new_size = int(new_len**0.5)
if ori_size != new_size:
logger.info("Position interpolate from %dx%d to %dx%d" % (ori_size, ori_size, new_size, new_size))
pos_embed_2d = pos_embed_1d.reshape(-1, ori_size, ori_size, embed_dim).permute(0, 3, 1, 2)
pos_embed_2d = torch.nn.functional.interpolate(
pos_embed_2d, size=(new_size, new_size), mode="bicubic", align_corners=False
)
pos_embed_1d = pos_embed_2d.permute(0, 2, 3, 1).flatten(1, 2)
state_dict[pe_key] = pos_embed_1d
torch.save(state_dict, model_path)
class PositionEmbedding(nn.Module):
def __init__(self, max_num_patch_per_side, hidden_size):
super().__init__()
self.max_num_patch_per_side = max_num_patch_per_side
self.hidden_size = hidden_size
self.pos_embed = nn.Parameter(
torch.zeros(max_num_patch_per_side ** 2, hidden_size),
requires_grad=False
)
self._init_weights()
def _init_weights(self):
# Initialize (and freeze) pos_embed by sin-cos embedding:
pos_embed = get_2d_sincos_pos_embed(self.hidden_size, self.max_num_patch_per_side)
self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float())
def forward(self, position_ids):
return self.pos_embed[position_ids]
class PostConvSmoother(nn.Module):
def __init__(self, in_channels=3, hidden_channels=64):
super().__init__()
self.net = nn.Sequential(
nn.Conv2d(in_channels, hidden_channels, kernel_size=3, padding=1),
nn.SiLU(),
nn.Conv2d(hidden_channels, in_channels, kernel_size=3, padding=1)
)
nn.init.zeros_(self.net[2].weight)
nn.init.zeros_(self.net[2].bias)
def forward(self, x):
return x + self.net(x) |