models_mae_before.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.layers.helpers import to_2tuple

from timm.models.vision_transformer import PatchEmbed, Block

from util.pos_embed import get_2d_sincos_pos_embed
from PASS import RunningMode, MaskMode, MaskGate, m_cfg

class PatchEmbed_(PatchEmbed):
# class PatchEmbed(nn.Module):
    """ Image to Patch Embedding
    """
    def __init__(self, img_size=256, patch_size=16,  in_chans=3,  embed_dim=768,
                 running_mode=RunningMode.GatePreTrain):
        super().__init__()
        img_size = to_2tuple(img_size)
        # patch_size = to_2tuple(patch_size)
        num_patches = (img_size[1] // patch_size) * (img_size[0] // patch_size)
        self.img_size = img_size
        self.patch_size = patch_size
        self.num_patches = num_patches

        if self.num_patches == -1:
            self.gate = None
        else:
            # patch_size = 4
            self.gate = MaskGate(patch_size, self.num_patches, in_chans, embed_dim, running_mode)
        self.running_mode = running_mode
        self.previous_x = None

        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)

    def forward(self, x):
        B, C, H, W = x.shape
        # print(self.gate.convmixer)
        #
        # # print(self.gate.convmixer[0])
        # print(self.gate.convmixer._modules['0'].Conv2d)
        #
        # print(self.gate.convmixer._modules['0'].Conv2d.weight.data)

        # FIXME look at relaxing size constraints
        assert H == self.img_size[0] and W == self.img_size[1], \
            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        if self.previous_x is None:
            out = x
            if m_cfg.mask_mode == MaskMode.Anchor or self.running_mode == RunningMode.Test:
                self.previous_x = x.clone().detach()
        else:
            # out = self.gate(x, self.previous_x)
            if m_cfg.mask_mode == MaskMode.Anchor:
                self.previous_x = x.clone().detach()
            elif self.gate is not None:
                x = self.gate(x, self.previous_x)
                if self.running_mode == RunningMode.Test:
                    self.previous_x = x.clone().detach()

        x = self.proj(x).flatten(2).transpose(1, 2)

        return x
# struct
class MaskedAutoencoderViT(nn.Module):
    """ Masked Autoencoder with VisionTransformer backbone
    """
    def __init__(self, img_size=256, 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, running_mode=RunningMode.GatePreTrain, batch_size=16):
        super().__init__()
        self.running_mode = running_mode
        # --------------------------------------------------------------------------
        # MAE encoder specifics
        # self, img_size = 224, patch_size = 16, patch_num = 196, in_chans = 3, embed_dim = 768,
        # running_mode = RunningMode.GatePreTrain)
        if running_mode == RunningMode.GatePreTrain:
            self.patch_embed = PatchEmbed_(img_size, patch_size, in_chans, embed_dim, running_mode)
        if running_mode == RunningMode.BackboneTrain:
            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.batch_size = batch_size
        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(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)
        self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True) # decoder to patch
        # --------------------------------------------------------------------------

        self.norm_pix_loss = norm_pix_loss
        self.window = 0
        self.window_period = 1

        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]
        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
    # here
    def forward_encoder(self, x):
        # 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, ids_restore = 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)
        # if x.shape != torch.Size([4, 257, 768]):

        if x.shape != torch.Size([12, 257, 1024]):
            print(x.shape,"EMMMMMMMMMMMMMMMMMMMMMMMMM")
        # apply Transformer blocks
        for blk in self.blocks:
            x = blk(x)
        x = self.norm(x)

        return x

    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
        x = self.decoder_pred(x)

        # remove cls token
        x = x[:, 1:, :]

        return x

    def forward_loss(self, imgs, pred):
        """
        imgs: [N, 3, H, W]
        pred: [N, L, p*p*3]
        mask: [N, L], 0 is keep, 1 is remove, 
        """
        # if self.running_mode == RunningMode.GatePreTrain:
        #     loss = nn.TripletMarginLoss(margin=200, swap=False, reduction='mean')
        # else:
        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):
        if self.running_mode == RunningMode.GatePreTrain:
            for n, p in self.patch_embed.named_parameters():
                if 'gate' in n:
                    p.requires_grad = True
                else:
                    p.requires_grad = False
            m_cfg.mask_mode = MaskMode.Positive
            dy = self.forward_encoder(imgs)
            m_cfg.mask_mode = MaskMode.Negative
            dx = self.forward_encoder(imgs)
            m_cfg.mask_mode = MaskMode.Anchor
            y = self.forward_encoder(imgs)
            dy = torch.reshape(dy, (self.batch_size, -1))
            dx = torch.reshape(dx, (self.batch_size, -1))
            y = torch.reshape(y, (self.batch_size, -1))
            forward_loss = nn.TripletMarginLoss(margin=200, swap=False, reduction='mean')
            tri_loss = forward_loss(y, dy, dx)
            pos_dist = torch.dist(y, dy)
            neg_dist = torch.dist(y, dx)
            pos_neg = pos_dist.detach() - neg_dist.detach()
            # loss_stats = {
            #     'loss': tri_loss, 'pos_dist': pos_dist, 'neg_dist': neg_dist, 'pos-neg': pos_neg
            # }

            return tri_loss
        else :
            if self.running_mode == RunningMode.FineTuning:
                for n, p in self.patch_embed.named_parameters():
                    if 'gate' in n:
                        p.requires_grad = False
                    else:
                        p.requires_grad = True
            if self.window == 0:
                m_cfg.mask_mode = MaskMode.Anchor
                outputs = self.forward_encoder(imgs)
                self.window = self.window_period
            else:
                if self.running_mode == RunningMode.FineTuning or self.running_mode == RunningMode.Test:
                    m_cfg.mask_mode = MaskMode.Positive
                outputs = self.forward_encoder(imgs)
                self.window = self.window - 1

            # loss, loss_stats = self.loss.forward(outputs=outputs, batch=batch)

            # latent = self.forward_encoder(imgs)
            pred = self.forward_decoder(outputs)  # [N, L, p*p*3]
            loss = self.forward_loss(imgs, pred)
            print("LOSS PRED")
            return loss, pred


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


# 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