# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # This source code is licensed under the license found in the # LICENSE file in the root directory of this source tree. # -------------------------------------------------------- # References: # timm: https://github.com/rwightman/pytorch-image-models/tree/master/timm # DeiT: https://github.com/facebookresearch/deit # -------------------------------------------------------- from functools import partial import torch import torch.nn as nn import numpy as np from timm.models.vision_transformer import PatchEmbed, Block class MaskedAutoencoderViT(nn.Module): """ Masked Autoencoder with VisionTransformer backbone """ def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=1024, depth=24, num_heads=16, decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16, mlp_ratio=4., norm_layer=partial(nn.LayerNorm, eps=1e-6), norm_pix_loss=False, quantile=False, quantile_head_num=9,**kwargs): super().__init__() # add for quantile forecasting self.norm_pix_loss = norm_pix_loss self.quantile = quantile self.quantile_head_num = quantile_head_num # -------------------------------------------------------------------------- # MAE encoder specifics self.patch_embed = PatchEmbed(img_size, patch_size, in_chans, embed_dim) num_patches = self.patch_embed.num_patches self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim), requires_grad=False) # fixed sin-cos embedding self.blocks = nn.ModuleList([ Block(embed_dim, num_heads, mlp_ratio, qkv_bias=True, norm_layer=norm_layer) for i in range(depth)]) self.norm = norm_layer(embed_dim) # -------------------------------------------------------------------------- # -------------------------------------------------------------------------- # MAE decoder specifics self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True) self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim)) self.decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, decoder_embed_dim), requires_grad=False) # fixed sin-cos embedding self.decoder_blocks = nn.ModuleList([ Block(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, norm_layer=norm_layer) for i in range(decoder_depth)]) self.decoder_norm = norm_layer(decoder_embed_dim) # prediction head self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True) # decoder to patch if self.quantile: # adds quantile outputs: # we need 9 outputs in total, corresponding to 10%, 20%, ..., 90% # So other 8 heads are needed to generate other quantile outputs. self.decoder_pred_quantile_list = nn.ModuleList() for i in range(self.quantile_head_num-1): self.decoder_pred_quantile_list.append(nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True)) # decoder to patch # -------------------------------------------------------------------------- self.initialize_weights() def initialize_weights(self): # initialization # initialize (and freeze) pos_embed by sin-cos embedding pos_embed = get_2d_sincos_pos_embed(self.pos_embed.shape[-1], int(self.patch_embed.num_patches**.5), cls_token=True) self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0)) decoder_pos_embed = get_2d_sincos_pos_embed(self.decoder_pos_embed.shape[-1], int(self.patch_embed.num_patches**.5), cls_token=True) self.decoder_pos_embed.data.copy_(torch.from_numpy(decoder_pos_embed).float().unsqueeze(0)) # initialize patch_embed like nn.Linear (instead of nn.Conv2d) w = self.patch_embed.proj.weight.data torch.nn.init.xavier_uniform_(w.view([w.shape[0], -1])) # timm's trunc_normal_(std=.02) is effectively normal_(std=0.02) as cutoff is too big (2.) torch.nn.init.normal_(self.cls_token, std=.02) torch.nn.init.normal_(self.mask_token, std=.02) # initialize nn.Linear and nn.LayerNorm self.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): # we use xavier_uniform following official JAX ViT: torch.nn.init.xavier_uniform_(m.weight) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) def patchify(self, imgs): """ imgs: (N, 3, H, W) x: (N, L, patch_size**2 * 3) """ p = self.patch_embed.patch_size[0] assert imgs.shape[2] == imgs.shape[3] and imgs.shape[2] % p == 0 h = w = imgs.shape[2] // p x = imgs.reshape(shape=(imgs.shape[0], 3, h, p, w, p)) x = torch.einsum('nchpwq->nhwpqc', x) x = x.reshape(shape=(imgs.shape[0], h * w, p**2 * 3)) return x def unpatchify(self, x, n_channels=3): """ x: (N, L, patch_size**2 * n_channels) imgs: (N, 3, H, W) """ p = self.patch_embed.patch_size[0] h = w = int(x.shape[1]**.5) assert h * w == x.shape[1] # x = x.reshape(shape=(x.shape[0], h, w, p, p, 3)) x = x.reshape(shape=(x.shape[0], h, w, p, p, n_channels)) x = torch.einsum('nhwpqc->nchpwq', x) # imgs = x.reshape(shape=(x.shape[0], 3, h * p, h * p)) imgs = x.reshape(shape=(x.shape[0], n_channels, h * p, h * p)) return imgs def random_masking(self, x, mask_ratio, noise=None): """ Perform per-sample random masking by per-sample shuffling. Per-sample shuffling is done by argsort random noise. x: [N, L, D], sequence noise: [N, L] """ N, L, D = x.shape # batch, length, dim len_keep = int(round(L * (1 - mask_ratio))) if noise is None: noise = torch.rand(N, L, device=x.device) # noise in [0, 1] # sort noise for each sample ids_shuffle = torch.argsort(noise, dim=1) # ascend: small is keep, large is remove ids_restore = torch.argsort(ids_shuffle, dim=1) # keep the first subset ids_keep = ids_shuffle[:, :len_keep] x_masked = torch.gather(x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D)) # generate the binary mask: 0 is keep, 1 is remove mask = torch.ones([N, L], device=x.device) mask[:, :len_keep] = 0 # unshuffle to get the binary mask mask = torch.gather(mask, dim=1, index=ids_restore) return x_masked, mask, ids_restore def forward_encoder(self, x, mask_ratio, noise=None,use_multiscale=True): # embed patches x = self.patch_embed(x) x = x + self.pos_embed[:, 1:, :] # x = x + self.pos_embed[:, 1:, :]+pos # masking: length -> length * mask_ratio x, mask, ids_restore = self.random_masking(x, mask_ratio, noise) # append cls token cls_token = self.cls_token + self.pos_embed[:, :1, :] cls_tokens = cls_token.expand(x.shape[0], -1, -1) x = torch.cat((cls_tokens, x), dim=1) total_depth = len(self.blocks) part_size = total_depth // 4 intermediate_features = [] # apply Transformer blocks for i, blk in enumerate(self.blocks): x = blk(x) if use_multiscale and (i + 1) % part_size == 0: intermediate_features.append(x) if use_multiscale: fused = torch.stack(intermediate_features, dim=0).sum(dim=0) # [N, L, D] fused = self.norm(fused) return fused, mask, ids_restore else: x = self.norm(x) return x, mask, ids_restore def forward_decoder(self, x, ids_restore): # embed tokens x = self.decoder_embed(x) # append mask tokens to sequence mask_tokens = self.mask_token.repeat(x.shape[0], ids_restore.shape[1] + 1 - x.shape[1], 1) x_ = torch.cat([x[:, 1:, :], mask_tokens], dim=1) # no cls token x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[2])) # unshuffle x = torch.cat([x[:, :1, :], x_], dim=1) # append cls token # add pos embed x = x + self.decoder_pos_embed # apply Transformer blocks for blk in self.decoder_blocks: x = blk(x) x = self.decoder_norm(x) # predictor projection if not self.quantile: x = self.decoder_pred(x) x = x[:, 1:, :] # remove cls token return x else: # first calculate the 50% quantile value x_mid = self.decoder_pred(x)[:, 1:, :] # [batch, ] # then calculate the other quantile values x_quantile_list = [] for i in range(self.quantile_head_num-1): x_quantile = self.decoder_pred_quantile_list[i](x)[:, 1:, :] x_quantile_list.append(x_quantile) # this will generate 9 groups of data return x_mid, x_quantile_list def forward_loss(self, imgs, pred, mask): """ imgs: [N, 3, H, W] pred: [N, L, p*p*3] mask: [N, L], 0 is keep, 1 is remove, """ target = self.patchify(imgs) if self.norm_pix_loss: mean = target.mean(dim=-1, keepdim=True) var = target.var(dim=-1, keepdim=True) target = (target - mean) / (var + 1.e-6)**.5 loss = (pred - target) ** 2 loss = loss.mean(dim=-1) # [N, L], mean loss per patch loss = (loss * mask).sum() / mask.sum() # mean loss on removed patches return loss def forward(self, imgs, mask_ratio=0.0, noise=None): latent, mask, ids_restore = self.forward_encoder(imgs, mask_ratio, noise) pred = self.forward_decoder(latent, ids_restore) # [N, L, p*p*3] # loss = self.forward_loss(imgs, pred, mask) return latent, pred, mask from transformers import PretrainedConfig class MAEConfig(PretrainedConfig): model_type = "mae_vit" def __init__( self, patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4.0, norm_layer_eps=1e-6, decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16, in_chans=3, img_size=224, mask_ratio=0.4, **kwargs ): super().__init__(**kwargs) self.patch_size = patch_size self.embed_dim = embed_dim self.depth = depth self.num_heads = num_heads self.mlp_ratio = mlp_ratio self.norm_layer_eps = norm_layer_eps self.decoder_embed_dim = decoder_embed_dim self.decoder_depth = decoder_depth self.decoder_num_heads = decoder_num_heads self.in_chans = in_chans self.img_size = img_size self.mask_ratio = mask_ratio def get_mae_base_config(): return dict( patch_size=16, embed_dim=768, depth=12, num_heads=12, decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16, mlp_ratio=4.0, norm_layer=partial(nn.LayerNorm, eps=1e-6), ) def get_mae_large_config(): config = get_mae_base_config() config.update( embed_dim=1024, depth=24, num_heads=16, ) return config def get_mae_huge_config(): config = get_mae_base_config() config.update( patch_size=14, embed_dim=1280, depth=32, num_heads=16, ) return config MAE_ARCH = { "mae_visualize_base.pth": get_mae_base_config(), "mae_visualize_large.pth": get_mae_large_config(), "mae_visualize_huge.pth": get_mae_huge_config(), } import numpy as np # -------------------------------------------------------- # 2D sine-cosine position embedding # References: # Transformer: https://github.com/tensorflow/models/blob/master/official/nlp/transformer/model_utils.py # MoCo v3: https://github.com/facebookresearch/moco-v3 # -------------------------------------------------------- def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False): """ 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) grid_w = np.arange(grid_size, dtype=np.float32) 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: pos_embed = np.concatenate([np.zeros([1, 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.float32) omega /= embed_dim / 2. omega = 1. / 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