models_milan.py

from functools import partial

import torch
import torch.nn as nn

from timm.models.vision_transformer import PatchEmbed, Block, Mlp

from util.pos_embed import get_2d_sincos_pos_embed
import sys
import numpy as np
from random import sample
import torch.nn.functional as F


class Attention_fordecoder(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
        self.scale = qk_scale or head_dim ** -0.5

        self.q = nn.Linear(dim, dim, bias=qkv_bias)
        self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

    def forward(self, x, fix_encoding_tokens, ids_restore, ids_shuffle, ids_keep, ids_dump):

        # merge encoding tokens and x
        x_ = torch.cat([fix_encoding_tokens[:, 1:, :], x], dim=1)
        assert x_.shape[1] == ids_restore.shape[1]
        x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[-1])) # unshuffle
        x_ = torch.cat([fix_encoding_tokens[:, :1, :], x_], dim=1)

        # full sequence shape
        B, N, C = x_.shape

        # qkv
        q = self.q(x).reshape(B, x.shape[1], self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
        kv = self.kv(x_).reshape(B, N, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        k, v = kv[0], kv[1]

        # attention
        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, x.shape[1], C)

        # output projection
        x = self.proj(x)
        x = self.proj_drop(x)

        return x


class Block_fordecoder(nn.Module):
    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention_fordecoder(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def forward(self, x, fix_encoding_tokens, ids_restore, ids_shuffle, ids_keep, ids_dump):
        x = x + self.drop_path(self.attn(self.norm1(x), fix_encoding_tokens, ids_restore, ids_shuffle, ids_keep, ids_dump))
        x = x + self.drop_path(self.mlp(self.norm2(x)))
        return x


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=nn.LayerNorm, 
                 norm_pix_loss=False, use_clip=False, attn_mask=False):
        super().__init__()

        # --------------------------------------------------------------------------
        # 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, qk_scale=None, 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_fordecoder(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, qk_scale=None, norm_layer=norm_layer)
            for i in range(decoder_depth)])

        self.decoder_norm = norm_layer(decoder_embed_dim)

        if use_clip is False:
            self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True) # decoder to patch
        else:
            self.decoder_pred = nn.Linear(decoder_embed_dim, 512, bias=True) # clip output [batch x seq x 512]
        # --------------------------------------------------------------------------

        self.norm_pix_loss = norm_pix_loss

        self.use_clip = use_clip
        self.attn_mask = attn_mask

        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):
        """
        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, mask_ratio):
        """
        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 = int(L * (1 - mask_ratio))
        
        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]
        ids_dump = 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, ids_shuffle, ids_keep, ids_dump

    def attention_masking(self, x, mask_ratio, importance):
        """
        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 = int(L * (1 - mask_ratio))
        
        noise = importance.to(x.device) # large is keep, small is remove
        
        # sort noise for each sample
        ids_shuffle = torch.multinomial(noise, L, replacement=False)
        ids_restore = torch.argsort(ids_shuffle, dim=1)

        # keep the first subset
        ids_keep = ids_shuffle[:, :len_keep]
        ids_dump = 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, ids_shuffle, ids_keep, ids_dump

    def forward_encoder(self, x, mask_ratio, importance=None):
        # 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
        if self.use_clip is False:
            assert importance is None
            x, mask, ids_restore, ids_shuffle, ids_keep, ids_dump = self.random_masking(x, mask_ratio)
        elif self.use_clip:
            assert importance is not None
            if self.attn_mask:
                x, mask, ids_restore, ids_shuffle, ids_keep, ids_dump = self.attention_masking(x, mask_ratio, importance)
            else:
                x, mask, ids_restore, ids_shuffle, ids_keep, ids_dump = self.random_masking(x, mask_ratio)

        # 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, ids_shuffle, ids_keep, ids_dump

    def forward_decoder(self, x, ids_restore, ids_shuffle, ids_keep, ids_dump):
        # 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

        # fix encoding tokens
        fix_encoding_tokens = torch.cat([x[:, :1, :], torch.gather(x[:, 1:, :] , dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, x.shape[-1]))], dim=1)
        x = torch.gather(x[:, 1:, :] , dim=1, index=ids_dump.unsqueeze(-1).repeat(1, 1, x.shape[-1]))

        # apply Transformer blocks
        for blk in self.decoder_blocks:
            x = blk(x, fix_encoding_tokens, ids_restore, ids_shuffle, ids_keep, ids_dump)

        # full sequence recovery
        x_ = torch.cat([fix_encoding_tokens[:, 1:, :], x], dim=1)
        x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[-1])) # unshuffle
        x = torch.cat([fix_encoding_tokens[:, :1, :], x_], dim=1)

        x = self.decoder_norm(x)

        # predictor projection
        x = self.decoder_pred(x)

        if self.use_clip is False:
            # remove cls token
            x = x[:, 1:, :]

        return x

    def forward_loss(self, imgs, pred, mask, clip_teacher=None, cidx=None, cluster_result=None):
        if self.use_clip is False:
            """
            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
        elif self.use_clip:
            # pred: [N, L+1, 512]
            # clip_teacher: [N, L+1, 512]
            pred = pred / pred.norm(dim=2, keepdim=True)
            clip_teacher = clip_teacher / clip_teacher.norm(dim=2, keepdim=True)
            assert pred.shape == clip_teacher.shape
            loss = 2 - 2 * (pred * clip_teacher).sum(dim=-1)
            loss = loss.mean()

            return loss

    def forward(self, imgs, mask_ratio=0.75, clip_teacher=None, cidx=None, cluster_result=None):
        if self.use_clip:
            clip_teacher, importance = clip_teacher[0], clip_teacher[1]
            importance = importance[:, 0, 1:]
        else:
            clip_teacher, importance = None, None
        latent, mask, ids_restore, ids_shuffle, ids_keep, ids_dump = self.forward_encoder(imgs, mask_ratio, importance)
        pred = self.forward_decoder(latent, ids_restore, ids_shuffle, ids_keep, ids_dump)  # [N, L, p*p*3]
        loss = self.forward_loss(imgs, pred, mask, clip_teacher, cidx, cluster_result)
        return loss, pred, mask


def mae_vit_base_patch16_dec512d8b(**kwargs):
    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):
    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):
    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


def mae_vit_small_patch16_dec512d8b(**kwargs):
    model = MaskedAutoencoderViT(
        patch_size=16, embed_dim=384, depth=12, num_heads=6,
        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
mae_vit_small_patch16 = mae_vit_small_patch16_dec512d8b  # decoder: 512 dim, 8 blocks