transformer.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.


import torch
import torch.nn as nn


class ModLN(nn.Module):
    """
    Modulation with adaLN.
    
    References:
    DiT: https://github.com/facebookresearch/DiT/blob/main/models.py#L101
    """
    def __init__(self, inner_dim: int, mod_dim: int, eps: float):
        super().__init__()
        self.norm = nn.LayerNorm(inner_dim, eps=eps)
        self.mlp = nn.Sequential(
            nn.SiLU(),
            nn.Linear(mod_dim, inner_dim * 2),
        )

    @staticmethod
    def modulate(x, shift, scale):
        # x: [N, L, D]
        # shift, scale: [N, D]
        return x * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1)

    def forward(self, x, cond):
        shift, scale = self.mlp(cond).chunk(2, dim=-1)  # [N, D]
        return self.modulate(self.norm(x), shift, scale)  # [N, L, D]


class ConditionModulationBlock(nn.Module):
    """
    Transformer block that takes in a cross-attention condition and another modulation vector applied to sub-blocks.
    """
    # use attention from torch.nn.MultiHeadAttention
    # Block contains a cross-attention layer, a self-attention layer, and a MLP
    def __init__(self, inner_dim: int, cond_dim: int, mod_dim: int, num_heads: int, eps: float,
                 attn_drop: float = 0., attn_bias: bool = False,
                 mlp_ratio: float = 4., mlp_drop: float = 0.):
        super().__init__()
        self.norm1 = ModLN(inner_dim, mod_dim, eps)
        self.cross_attn = nn.MultiheadAttention(
            embed_dim=inner_dim, num_heads=num_heads, kdim=cond_dim, vdim=cond_dim,
            dropout=attn_drop, bias=attn_bias, batch_first=True)
        self.norm2 = ModLN(inner_dim, mod_dim, eps)
        self.self_attn = nn.MultiheadAttention(
            embed_dim=inner_dim, num_heads=num_heads,
            dropout=attn_drop, bias=attn_bias, batch_first=True)
        self.norm3 = ModLN(inner_dim, mod_dim, eps)
        self.mlp = nn.Sequential(
            nn.Linear(inner_dim, int(inner_dim * mlp_ratio)),
            nn.GELU(),
            nn.Dropout(mlp_drop),
            nn.Linear(int(inner_dim * mlp_ratio), inner_dim),
            nn.Dropout(mlp_drop),
        )

    def forward(self, x, cond, mod):
        # x: [N, L, D]
        # cond: [N, L_cond, D_cond]
        # mod: [N, D_mod]
        x = x + self.cross_attn(self.norm1(x, mod), cond, cond, need_weights=False)[0]
        before_sa = self.norm2(x, mod)
        x = x + self.self_attn(before_sa, before_sa, before_sa, need_weights=False)[0]
        x = x + self.mlp(self.norm3(x, mod))
        return x


class TriplaneTransformer(nn.Module):
    """
    Transformer with condition and modulation that generates a triplane representation.
    
    Reference:
    Timm: https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/vision_transformer.py#L486
    """
    def __init__(self, inner_dim: int, image_feat_dim: int, camera_embed_dim: int,
                 triplane_low_res: int, triplane_high_res: int, triplane_dim: int,
                 num_layers: int, num_heads: int,
                 eps: float = 1e-6):
        super().__init__()

        # attributes
        self.triplane_low_res = triplane_low_res
        self.triplane_high_res = triplane_high_res
        self.triplane_dim = triplane_dim

        # modules
        # initialize pos_embed with 1/sqrt(dim) * N(0, 1)
        self.pos_embed = nn.Parameter(torch.randn(1, 3*triplane_low_res**2, inner_dim) * (1. / inner_dim) ** 0.5)
        self.layers = nn.ModuleList([
            ConditionModulationBlock(
                inner_dim=inner_dim, cond_dim=image_feat_dim, mod_dim=camera_embed_dim, num_heads=num_heads, eps=eps)
            for _ in range(num_layers)
        ])
        self.norm = nn.LayerNorm(inner_dim, eps=eps)
        self.deconv = nn.ConvTranspose2d(inner_dim, triplane_dim, kernel_size=2, stride=2, padding=0)

    def forward(self, image_feats, camera_embeddings):
        # image_feats: [N, L_cond, D_cond]
        # camera_embeddings: [N, D_mod]

        assert image_feats.shape[0] == camera_embeddings.shape[0], \
            f"Mismatched batch size: {image_feats.shape[0]} vs {camera_embeddings.shape[0]}"

        N = image_feats.shape[0]
        H = W = self.triplane_low_res
        L = 3 * H * W

        x = self.pos_embed.repeat(N, 1, 1)  # [N, L, D]
        for layer in self.layers:
            x = layer(x, image_feats, camera_embeddings)
        x = self.norm(x)

        # separate each plane and apply deconv
        x = x.view(N, 3, H, W, -1)
        x = torch.einsum('nihwd->indhw', x)  # [3, N, D, H, W]
        x = x.contiguous().view(3*N, -1, H, W)  # [3*N, D, H, W]
        x = self.deconv(x)  # [3*N, D', H', W']
        x = x.view(3, N, *x.shape[-3:])  # [3, N, D', H', W']
        x = torch.einsum('indhw->nidhw', x)  # [N, 3, D', H', W']
        x = x.contiguous()

        assert self.triplane_high_res == x.shape[-2], \
            f"Output triplane resolution does not match with expected: {x.shape[-2]} vs {self.triplane_high_res}"
        assert self.triplane_dim == x.shape[-3], \
            f"Output triplane dimension does not match with expected: {x.shape[-3]} vs {self.triplane_dim}"

        return x