LatentSync/latentsync/trepa/third_party/VideoMAEv2/videomaev2_finetune.py

# --------------------------------------------------------
# Based on BEiT, timm, DINO and DeiT code bases
# https://github.com/microsoft/unilm/tree/master/beit
# https://github.com/rwightman/pytorch-image-models/tree/master/timm
# https://github.com/facebookresearch/deit
# https://github.com/facebookresearch/dino
# --------------------------------------------------------'
from functools import partial

import math
import warnings
import numpy as np
import collections.abc
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as cp
from itertools import repeat


def _no_grad_trunc_normal_(tensor, mean, std, a, b):
    # Cut & paste from PyTorch official master until it's in a few official releases - RW
    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
    def norm_cdf(x):
        # Computes standard normal cumulative distribution function
        return (1.0 + math.erf(x / math.sqrt(2.0))) / 2.0

    if (mean < a - 2 * std) or (mean > b + 2 * std):
        warnings.warn(
            "mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
            "The distribution of values may be incorrect.",
            stacklevel=2,
        )

    with torch.no_grad():
        # Values are generated by using a truncated uniform distribution and
        # then using the inverse CDF for the normal distribution.
        # Get upper and lower cdf values
        l = norm_cdf((a - mean) / std)
        u = norm_cdf((b - mean) / std)

        # Uniformly fill tensor with values from [l, u], then translate to
        # [2l-1, 2u-1].
        tensor.uniform_(2 * l - 1, 2 * u - 1)

        # Use inverse cdf transform for normal distribution to get truncated
        # standard normal
        tensor.erfinv_()

        # Transform to proper mean, std
        tensor.mul_(std * math.sqrt(2.0))
        tensor.add_(mean)

        # Clamp to ensure it's in the proper range
        tensor.clamp_(min=a, max=b)
        return tensor


def trunc_normal_(tensor, mean=0.0, std=1.0, a=-2.0, b=2.0):
    r"""Fills the input Tensor with values drawn from a truncated
    normal distribution. The values are effectively drawn from the
    normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)`
    with values outside :math:`[a, b]` redrawn until they are within
    the bounds. The method used for generating the random values works
    best when :math:`a \leq \text{mean} \leq b`.
    Args:
        tensor: an n-dimensional `torch.Tensor`
        mean: the mean of the normal distribution
        std: the standard deviation of the normal distribution
        a: the minimum cutoff value
        b: the maximum cutoff value
    Examples:
        >>> w = torch.empty(3, 5)
        >>> nn.init.trunc_normal_(w)
    """
    return _no_grad_trunc_normal_(tensor, mean, std, a, b)


def _ntuple(n):
    def parse(x):
        if isinstance(x, collections.abc.Iterable):
            return x
        return tuple(repeat(x, n))

    return parse


to_2tuple = _ntuple(2)


def drop_path(x, drop_prob: float = 0.0, training: bool = False):
    """
    Adapted from timm codebase
    """
    if drop_prob == 0.0 or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output


def _cfg(url="", **kwargs):
    return {
        "url": url,
        "num_classes": 400,
        "input_size": (3, 224, 224),
        "pool_size": None,
        "crop_pct": 0.9,
        "interpolation": "bicubic",
        "mean": (0.5, 0.5, 0.5),
        "std": (0.5, 0.5, 0.5),
        **kwargs,
    }


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""

    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)

    def extra_repr(self) -> str:
        return "p={}".format(self.drop_prob)


class Mlp(nn.Module):

    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.0):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        # x = self.drop(x)
        # commit this for the original BERT implement
        x = self.fc2(x)
        x = self.drop(x)
        return x


class CosAttention(nn.Module):

    def __init__(
        self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0.0, proj_drop=0.0, attn_head_dim=None
    ):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        if attn_head_dim is not None:
            head_dim = attn_head_dim
        all_head_dim = head_dim * self.num_heads
        # self.scale = qk_scale or head_dim**-0.5
        # DO NOT RENAME [self.scale] (for no weight decay)
        if qk_scale is None:
            self.scale = nn.Parameter(torch.log(10 * torch.ones((num_heads, 1, 1))), requires_grad=True)
        else:
            self.scale = qk_scale

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

        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(all_head_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 = F.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 = F.normalize(q, dim=-1) @ F.normalize(k, dim=-1).transpose(-2, -1)

        # torch.log(torch.tensor(1. / 0.01)) = 4.6052
        logit_scale = torch.clamp(self.scale, max=4.6052).exp()

        attn = attn * logit_scale

        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B, N, -1)

        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class Attention(nn.Module):

    def __init__(
        self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0.0, proj_drop=0.0, attn_head_dim=None
    ):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        if attn_head_dim is not None:
            head_dim = attn_head_dim
        all_head_dim = head_dim * self.num_heads
        self.scale = qk_scale or head_dim**-0.5

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

        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(all_head_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 = F.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)

        # Use PyTorch native implementation of FlashAttention-2
        attn = F.scaled_dot_product_attention(q, k, v)

        x = attn.transpose(1, 2).reshape(B, N, -1)

        # Deprecated attn implementation, which consumes much more VRAM
        # q = q * self.scale
        # attn = q @ k.transpose(-2, -1)
        # attn = attn.softmax(dim=-1)
        # attn = self.attn_drop(attn)
        # x = (attn @ v).transpose(1, 2).reshape(B, N, -1)

        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class Block(nn.Module):

    def __init__(
        self,
        dim,
        num_heads,
        mlp_ratio=4.0,
        qkv_bias=False,
        qk_scale=None,
        drop=0.0,
        attn_drop=0.0,
        drop_path=0.0,
        init_values=None,
        act_layer=nn.GELU,
        norm_layer=nn.LayerNorm,
        attn_head_dim=None,
        cos_attn=False,
    ):
        super().__init__()
        self.norm1 = norm_layer(dim)
        if cos_attn:
            self.attn = CosAttention(
                dim,
                num_heads=num_heads,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                attn_drop=attn_drop,
                proj_drop=drop,
                attn_head_dim=attn_head_dim,
            )
        else:
            self.attn = Attention(
                dim,
                num_heads=num_heads,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                attn_drop=attn_drop,
                proj_drop=drop,
                attn_head_dim=attn_head_dim,
            )
        # 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 > 0.0 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)

        if init_values > 0:
            self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)), requires_grad=True)
            self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)), requires_grad=True)
        else:
            self.gamma_1, self.gamma_2 = None, None

    def forward(self, x):
        if self.gamma_1 is None:
            x = x + self.drop_path(self.attn(self.norm1(x)))
            x = x + self.drop_path(self.mlp(self.norm2(x)))
        else:
            x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x)))
            x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))
        return x


class PatchEmbed(nn.Module):
    """Image to Patch Embedding"""

    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, num_frames=16, tubelet_size=2):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        num_spatial_patches = (img_size[0] // patch_size[0]) * (img_size[1] // patch_size[1])
        num_patches = num_spatial_patches * (num_frames // tubelet_size)

        self.img_size = img_size
        self.tubelet_size = tubelet_size
        self.patch_size = patch_size
        self.num_patches = num_patches
        self.proj = nn.Conv3d(
            in_channels=in_chans,
            out_channels=embed_dim,
            kernel_size=(self.tubelet_size, patch_size[0], patch_size[1]),
            stride=(self.tubelet_size, patch_size[0], patch_size[1]),
        )

    def forward(self, x, **kwargs):
        B, C, T, H, W = x.shape
        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]})."
        # b, c, l -> b, l, c
        # [1, 1408, 8, 16, 16] -> [1, 1408, 2048] -> [1, 2048, 1408]
        x = self.proj(x).flatten(2).transpose(1, 2)
        return x


# sin-cos position encoding
# https://github.com/jadore801120/attention-is-all-you-need-pytorch/blob/master/transformer/Models.py#L31
def get_sinusoid_encoding_table(n_position, d_hid):
    """Sinusoid position encoding table"""

    # TODO: make it with torch instead of numpy
    def get_position_angle_vec(position):
        return [position / np.power(10000, 2 * (hid_j // 2) / d_hid) for hid_j in range(d_hid)]

    sinusoid_table = np.array([get_position_angle_vec(pos_i) for pos_i in range(n_position)])
    sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2])  # dim 2i
    sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2])  # dim 2i+1

    return torch.tensor(sinusoid_table, dtype=torch.float, requires_grad=False).unsqueeze(0)


class VisionTransformer(nn.Module):
    """Vision Transformer with support for patch or hybrid CNN input stage"""

    def __init__(
        self,
        img_size=224,
        patch_size=16,
        in_chans=3,
        num_classes=1000,
        embed_dim=768,
        depth=12,
        num_heads=12,
        mlp_ratio=4.0,
        qkv_bias=False,
        qk_scale=None,
        drop_rate=0.0,
        attn_drop_rate=0.0,
        drop_path_rate=0.0,
        head_drop_rate=0.0,
        norm_layer=nn.LayerNorm,
        init_values=0.0,
        use_learnable_pos_emb=False,
        init_scale=0.0,
        all_frames=16,
        tubelet_size=2,
        use_mean_pooling=True,
        with_cp=False,
        cos_attn=False,
    ):
        super().__init__()
        self.num_classes = num_classes
        # num_features for consistency with other models
        self.num_features = self.embed_dim = embed_dim
        self.tubelet_size = tubelet_size
        self.patch_embed = PatchEmbed(
            img_size=img_size,
            patch_size=patch_size,
            in_chans=in_chans,
            embed_dim=embed_dim,
            num_frames=all_frames,
            tubelet_size=tubelet_size,
        )
        num_patches = self.patch_embed.num_patches
        self.with_cp = with_cp

        if use_learnable_pos_emb:
            self.pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
        else:
            # sine-cosine positional embeddings is on the way
            self.pos_embed = get_sinusoid_encoding_table(num_patches, embed_dim)

        self.pos_drop = nn.Dropout(p=drop_rate)

        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
        self.blocks = nn.ModuleList(
            [
                Block(
                    dim=embed_dim,
                    num_heads=num_heads,
                    mlp_ratio=mlp_ratio,
                    qkv_bias=qkv_bias,
                    qk_scale=qk_scale,
                    drop=drop_rate,
                    attn_drop=attn_drop_rate,
                    drop_path=dpr[i],
                    norm_layer=norm_layer,
                    init_values=init_values,
                    cos_attn=cos_attn,
                )
                for i in range(depth)
            ]
        )
        self.norm = nn.Identity() if use_mean_pooling else norm_layer(embed_dim)
        self.fc_norm = norm_layer(embed_dim) if use_mean_pooling else None
        self.head_dropout = nn.Dropout(head_drop_rate)
        self.head = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()

        if use_learnable_pos_emb:
            trunc_normal_(self.pos_embed, std=0.02)

        self.apply(self._init_weights)

        self.head.weight.data.mul_(init_scale)
        self.head.bias.data.mul_(init_scale)
        self.num_frames = all_frames

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=0.02)
            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 get_num_layers(self):
        return len(self.blocks)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {"pos_embed", "cls_token"}

    def get_classifier(self):
        return self.head

    def reset_classifier(self, num_classes, global_pool=""):
        self.num_classes = num_classes
        self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()

    def interpolate_pos_encoding(self, t):
        T = 8
        t0 = t // self.tubelet_size
        if T == t0:
            return self.pos_embed
        dim = self.pos_embed.shape[-1]
        patch_pos_embed = self.pos_embed.permute(0, 2, 1).reshape(1, dim, 8, 16, 16)
        # we add a small number to avoid floating point error in the interpolation
        # see discussion at https://github.com/facebookresearch/dino/issues/8
        t0 = t0 + 0.1
        patch_pos_embed = nn.functional.interpolate(
            patch_pos_embed,
            scale_factor=(t0 / T, 1, 1),
            mode="trilinear",
        )
        assert int(t0) == patch_pos_embed.shape[-3]
        patch_pos_embed = patch_pos_embed.reshape(1, dim, -1).permute(0, 2, 1)
        return patch_pos_embed

    def forward_features(self, x):
        # [1, 3, 16, 224, 224]
        B = x.size(0)
        T = x.size(2)

        # [1, 2048, 1408]
        x = self.patch_embed(x)

        if self.pos_embed is not None:
            x = x + self.interpolate_pos_encoding(T).expand(B, -1, -1).type_as(x).to(x.device).clone().detach()
        x = self.pos_drop(x)

        for blk in self.blocks:
            if self.with_cp:
                x = cp.checkpoint(blk, x, use_reentrant=False)
            else:
                x = blk(x)

        # return self.fc_norm(x)

        if self.fc_norm is not None:
            return self.fc_norm(x.mean(1))
        else:
            return self.norm(x[:, 0])

    def forward(self, x):
        x = self.forward_features(x)
        x = self.head_dropout(x)
        x = self.head(x)
        return x


def vit_giant_patch14_224(pretrained=False, **kwargs):
    model = VisionTransformer(
        patch_size=14,
        embed_dim=1408,
        depth=40,
        num_heads=16,
        mlp_ratio=48 / 11,
        qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),
        **kwargs,
    )
    model.default_cfg = _cfg()
    return model