models_mae.py

# 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

from timm.models.vision_transformer import PatchEmbed, Block

from util.pos_embed import get_2d_sincos_pos_embed


class AttentionNoKBias(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, attn_drop=0., proj_drop=0.):
        super().__init__()
        print("using ViT self-attention without k_bias")

        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim ** -0.5

        self.qkv = nn.Linear(dim, dim * 3, bias=False)
        if qkv_bias:
            self.q_bias = nn.Parameter(torch.zeros(dim))
            self.v_bias = nn.Parameter(torch.zeros(dim))
        else:
            self.q_bias = None
            self.v_bias = None
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

    def forward(self, x):
        B, N, C = x.shape
        qkv_bias = None
        if self.q_bias is not None:
            qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias))
        # qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv = nn.functional.linear(input=x, weight=self.qkv.weight, bias=qkv_bias)
        qkv = qkv.reshape(B, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)

        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class MaskedAutoencoderViT(nn.Module):
    """ Masked Autoencoder with VisionTransformer backbone
    """
    def __init__(self, args, 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=nn.LayerNorm, norm_pix_loss=False):
        super().__init__()
        self.args = args

        if args.no_k_bias_in_vit:
            # monkey-patch `timm.models.vision_transformer.Attention`
            # to our version above
            import timm.models.vision_transformer
            timm.models.vision_transformer.Attention = AttentionNoKBias

        # --------------------------------------------------------------------------
        # 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))

        # allow using a smaller sequence length in the decoder than in the encoder
        # by downsampling the decoder input feature map with a learned conv
        assert num_patches % args.decoder_downsampling ** 2 == 0
        self.decoder_num_patches = num_patches // args.decoder_downsampling ** 2
        self.grid_size = int(num_patches ** 0.5)
        assert self.grid_size ** 2 == num_patches
        self.decoder_grid_size = self.grid_size // args.decoder_downsampling
        if args.decoder_downsampling > 1:
            self.decoder_downsample = nn.Conv2d(
                decoder_embed_dim,
                decoder_embed_dim,
                kernel_size=args.decoder_downsampling,
                stride=args.decoder_downsampling,
            )
        assert args.decoder_downsampling % args.pred_downsampling == 0
        self.decoder_out_upsampling = args.decoder_downsampling // args.pred_downsampling
        if self.decoder_out_upsampling > 1:
            self.decoder_upsample = nn.ConvTranspose2d(
                decoder_embed_dim,
                decoder_embed_dim,
                kernel_size=args.decoder_downsampling,
                stride=args.decoder_downsampling,
            )

        self.decoder_pos_embed = nn.Parameter(torch.zeros(1, self.decoder_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)
        pred_patch_size = patch_size * args.pred_downsampling
        self.decoder_pred = nn.Linear(decoder_embed_dim, pred_patch_size**2 * in_chans, bias=True) # decoder to patch
        # --------------------------------------------------------------------------

        self.norm_pix_loss = norm_pix_loss

        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], self.decoder_grid_size, 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

        p = q = self.patch_embed.patch_size[0] * self.args.pred_downsampling
        # (n, c, h, w) => (n, c, h/p, p, w/q, q) => (n, h/p, w/q, p, q, c)
        #  => (n, h/p * w/q, p*q*c)
        return imgs.view(
            imgs.size(0),
            imgs.size(1),
            imgs.size(2) // p,
            p,
            imgs.size(3) // q,
            q,
        ).permute(0, 2, 4, 3, 5, 1).flatten(3).flatten(1, 2)

    def unpatchify(self, x):
        """
        x: (N, L, patch_size**2 *3)
        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 = torch.einsum('nhwpqc->nchpwq', x)
        imgs = x.reshape(shape=(x.shape[0], 3, h * p, h * p))
        return imgs

    def random_masking(self, x, ids_keep, ids_restore):
        """
        Perform per-sample random masking by per-sample shuffling.
        Per-sample shuffling is done by argsort random noise.
        x: [N, L, D], sequence
        """
        N, L, D = x.shape  # batch, length, dim
        len_keep = ids_keep.size(-1)
        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

    def forward_encoder(self, x, ids_keep, ids_restore):
        # embed patches
        x = self.patch_embed(x)

        # add pos embed w/o cls token
        x = x + self.pos_embed[:, 1:, :]

        # masking: length -> length * mask_ratio
        x, mask = self.random_masking(x, ids_keep, ids_restore)

        # 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)

        # apply Transformer blocks
        for blk in self.blocks:
            x = blk(x)
        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
        if self.args.decoder_downsampling != 1:
            x_ = x_.view(x_.size(0), self.grid_size, self.grid_size, x_.size(2)).permute(0, 3, 1, 2)  # NHWC => NCHW
            x_ = self.decoder_downsample(x_)
            x_ = x_.flatten(start_dim=2).permute(0, 2, 1)  # NCHW => NHWC
        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)
        # remove cls token
        x = x[:, 1:, :]
        if self.decoder_out_upsampling != 1:
            x = x.view(x.size(0), self.decoder_grid_size, self.decoder_grid_size, x.size(2)).permute(0, 3, 1, 2)  # NHWC => NCHW
            x = self.decoder_upsample(x)
            x = x.flatten(start_dim=2).permute(0, 2, 1)  # NCHW => NHWC

        x = self.decoder_norm(x)

        # predictor projection
        x = self.decoder_pred(x)

        return x

    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

        if self.args.pred_downsampling > 1:
            mask = nn.functional.avg_pool2d(
                mask.view(mask.size(0), 1, self.grid_size, self.grid_size),
                kernel_size=self.args.pred_downsampling,
                stride=self.args.pred_downsampling,
            ).flatten(1)
        loss = (loss * mask).sum() / mask.sum()  # mean loss on removed patches
        return loss

    def forward(self, imgs, ids_keep, ids_restore):
        latent, mask, ids_restore = self.forward_encoder(imgs, ids_keep, ids_restore)
        pred = self.forward_decoder(latent, ids_restore)  # [N, L, p*p*3]
        loss = self.forward_loss(imgs, pred, mask)
        return loss, pred, mask


def mae_vit_base_patch16_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 16
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_base_patch4_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 4
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_base_patch8_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 8
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_base_patch24_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 24
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_base_patch32_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 32
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_base_patch64_dec384d12h8b(**kwargs):
    assert kwargs["patch_size"] == 64
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=768, depth=12, num_heads=12,
        decoder_embed_dim=384, decoder_num_heads=12, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch16_dec512d16h8b(**kwargs):
    assert kwargs["patch_size"] == 16
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=1024, depth=24, num_heads=16,
        decoder_embed_dim=512, decoder_num_heads=16, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch4_dec512d16h8b(**kwargs):
    assert kwargs["patch_size"] == 4
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=1024, depth=24, num_heads=16,
        decoder_embed_dim=512, decoder_num_heads=16, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch8_dec512d16h8b(**kwargs):
    assert kwargs["patch_size"] == 8
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=1024, depth=24, num_heads=16,
        decoder_embed_dim=512, decoder_num_heads=16, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch24_dec512d16h8b(**kwargs):
    assert kwargs["patch_size"] == 24
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=1024, depth=24, num_heads=16,
        decoder_embed_dim=512, decoder_num_heads=16, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch32_dec512d16h8b(**kwargs):
    assert kwargs["patch_size"] == 32
    kwargs.pop("decoder_embed_dim", None)
    kwargs.pop("decoder_depth", None)
    model = MaskedAutoencoderViT(
        embed_dim=1024, depth=24, num_heads=16,
        decoder_embed_dim=512, decoder_num_heads=16, decoder_depth=8,
        mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


# Below are the original MAE models

def mae_vit_base_patch16_dec512d8b(**kwargs):
    assert kwargs["patch_size"] == 16
    model = MaskedAutoencoderViT(
        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, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
    return model


def mae_vit_large_patch16_dec512d8b(**kwargs):
    assert kwargs["patch_size"] == 16
    model = MaskedAutoencoderViT(
        patch_size=16, 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), **kwargs)
    return model


def mae_vit_huge_patch14_dec512d8b(**kwargs):
    assert kwargs["patch_size"] == 14
    model = MaskedAutoencoderViT(
        patch_size=14, embed_dim=1280, depth=32, 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), **kwargs)
    return model


# set recommended archs
mae_vit_base_patch16 = mae_vit_base_patch16_dec512d8b  # decoder: 512 dim, 8 blocks
mae_vit_large_patch16 = mae_vit_large_patch16_dec512d8b  # decoder: 512 dim, 8 blocks
mae_vit_huge_patch14 = mae_vit_huge_patch14_dec512d8b  # decoder: 512 dim, 8 blocks