modeling/vision/backbone/vit.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 os

import logging
import torch
import torch.nn as nn
import torch.nn.functional as F

from typing import Optional, Tuple, Type
from functools import partial

from .common import LayerNorm2d, MLPBlock

from detectron2.utils.file_io import PathManager
from detectron2.modeling import BACKBONE_REGISTRY, Backbone, ShapeSpec

from .build import register_backbone

logger = logging.getLogger(__name__)

# This class and its supporting functions below lightly adapted from the ViTDet backbone available at: https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/backbone/vit.py # noqa
class ImageEncoderViT(nn.Module):
    def __init__(
        self,
        img_size: int = 1024,
        patch_size: int = 16,
        in_chans: int = 3,
        embed_dim: int = 768,
        depth: int = 12,
        num_heads: int = 12,
        mlp_ratio: float = 4.0,
        out_chans: int = 256,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Module] = nn.LayerNorm,
        act_layer: Type[nn.Module] = nn.GELU,
        use_abs_pos: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        global_attn_indexes: Tuple[int, ...] = (),
    ) -> None:
        """
        Args:
            img_size (int): Input image size.
            patch_size (int): Patch size.
            in_chans (int): Number of input image channels.
            embed_dim (int): Patch embedding dimension.
            depth (int): Depth of ViT.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            norm_layer (nn.Module): Normalization layer.
            act_layer (nn.Module): Activation layer.
            use_abs_pos (bool): If True, use absolute positional embeddings.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks.
            global_attn_indexes (list): Indexes for blocks using global attention.
        """
        super().__init__()
        self.img_size = img_size

        self.patch_embed = PatchEmbed(
            kernel_size=(patch_size, patch_size),
            stride=(patch_size, patch_size),
            in_chans=in_chans,
            embed_dim=embed_dim,
        )

        self.pos_embed: Optional[nn.Parameter] = None
        if use_abs_pos:
            # Initialize absolute positional embedding with pretrain image size.
            self.pos_embed = nn.Parameter(
                torch.zeros(1, img_size // patch_size, img_size // patch_size, embed_dim)
            )

        self.blocks = nn.ModuleList()
        for i in range(depth):
            block = Block(
                dim=embed_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                norm_layer=norm_layer,
                act_layer=act_layer,
                use_rel_pos=use_rel_pos,
                rel_pos_zero_init=rel_pos_zero_init,
                window_size=window_size if i not in global_attn_indexes else 0,
                input_size=(img_size // patch_size, img_size // patch_size),
            )
            self.blocks.append(block)

        self.neck = nn.Sequential(
            nn.Conv2d(
                embed_dim,
                out_chans,
                kernel_size=1,
                bias=False,
            ),
            LayerNorm2d(out_chans),
            nn.Conv2d(
                out_chans,
                out_chans,
                kernel_size=3,
                padding=1,
                bias=False,
            ),
            LayerNorm2d(out_chans),
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.patch_embed(x)
        if self.pos_embed is not None:
            x = x + self.pos_embed

        for blk in self.blocks:
            x = blk(x)

        x = self.neck(x.permute(0, 3, 1, 2))

        return x


class Block(nn.Module):
    """Transformer blocks with support of window attention and residual propagation blocks"""

    def __init__(
        self,
        dim: int,
        num_heads: int,
        mlp_ratio: float = 4.0,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Module] = nn.LayerNorm,
        act_layer: Type[nn.Module] = nn.GELU,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            norm_layer (nn.Module): Normalization layer.
            act_layer (nn.Module): Activation layer.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks. If it equals 0, then
                use global attention.
            input_size (tuple(int, int) or None): Input resolution for calculating the relative
                positional parameter size.
        """
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads,
            qkv_bias=qkv_bias,
            use_rel_pos=use_rel_pos,
            rel_pos_zero_init=rel_pos_zero_init,
            input_size=input_size if window_size == 0 else (window_size, window_size),
        )

        self.norm2 = norm_layer(dim)
        self.mlp = MLPBlock(embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer)

        self.window_size = window_size

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        shortcut = x
        x = self.norm1(x)
        # Window partition
        if self.window_size > 0:
            H, W = x.shape[1], x.shape[2]
            x, pad_hw = window_partition(x, self.window_size)

        x = self.attn(x)
        # Reverse window partition
        if self.window_size > 0:
            x = window_unpartition(x, self.window_size, pad_hw, (H, W))

        x = shortcut + x
        x = x + self.mlp(self.norm2(x))

        return x


class Attention(nn.Module):
    """Multi-head Attention block with relative position embeddings."""

    def __init__(
        self,
        dim: int,
        num_heads: int = 8,
        qkv_bias: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads.
            qkv_bias (bool):  If True, add a learnable bias to query, key, value.
            rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            input_size (tuple(int, int) or None): Input resolution for calculating the relative
                positional parameter size.
        """
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.proj = nn.Linear(dim, dim)

        self.use_rel_pos = use_rel_pos
        if self.use_rel_pos:
            assert (
                input_size is not None
            ), "Input size must be provided if using relative positional encoding."
            # initialize relative positional embeddings
            self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
            self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        B, H, W, _ = x.shape
        # qkv with shape (3, B, nHead, H * W, C)
        qkv = self.qkv(x).reshape(B, H * W, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        # q, k, v with shape (B * nHead, H * W, C)
        q, k, v = qkv.reshape(3, B * self.num_heads, H * W, -1).unbind(0)

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

        if self.use_rel_pos:
            attn = add_decomposed_rel_pos(attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W))

        attn = attn.softmax(dim=-1)
        x = (attn @ v).view(B, self.num_heads, H, W, -1).permute(0, 2, 3, 1, 4).reshape(B, H, W, -1)
        x = self.proj(x)

        return x


def window_partition(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, Tuple[int, int]]:
    """
    Partition into non-overlapping windows with padding if needed.
    Args:
        x (tensor): input tokens with [B, H, W, C].
        window_size (int): window size.

    Returns:
        windows: windows after partition with [B * num_windows, window_size, window_size, C].
        (Hp, Wp): padded height and width before partition
    """
    B, H, W, C = x.shape

    pad_h = (window_size - H % window_size) % window_size
    pad_w = (window_size - W % window_size) % window_size
    if pad_h > 0 or pad_w > 0:
        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h))
    Hp, Wp = H + pad_h, W + pad_w

    x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows, (Hp, Wp)


def window_unpartition(
    windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
) -> torch.Tensor:
    """
    Window unpartition into original sequences and removing padding.
    Args:
        windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
        window_size (int): window size.
        pad_hw (Tuple): padded height and width (Hp, Wp).
        hw (Tuple): original height and width (H, W) before padding.

    Returns:
        x: unpartitioned sequences with [B, H, W, C].
    """
    Hp, Wp = pad_hw
    H, W = hw
    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
    x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1)
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1)

    if Hp > H or Wp > W:
        x = x[:, :H, :W, :].contiguous()
    return x


def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor:
    """
    Get relative positional embeddings according to the relative positions of
        query and key sizes.
    Args:
        q_size (int): size of query q.
        k_size (int): size of key k.
        rel_pos (Tensor): relative position embeddings (L, C).

    Returns:
        Extracted positional embeddings according to relative positions.
    """
    max_rel_dist = int(2 * max(q_size, k_size) - 1)
    # Interpolate rel pos if needed.
    if rel_pos.shape[0] != max_rel_dist:
        # Interpolate rel pos.
        rel_pos_resized = F.interpolate(
            rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
            size=max_rel_dist,
            mode="linear",
        )
        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
    else:
        rel_pos_resized = rel_pos

    # Scale the coords with short length if shapes for q and k are different.
    q_coords = torch.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
    k_coords = torch.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)

    return rel_pos_resized[relative_coords.long()]


def add_decomposed_rel_pos(
    attn: torch.Tensor,
    q: torch.Tensor,
    rel_pos_h: torch.Tensor,
    rel_pos_w: torch.Tensor,
    q_size: Tuple[int, int],
    k_size: Tuple[int, int],
) -> torch.Tensor:
    """
    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
    Args:
        attn (Tensor): attention map.
        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).

    Returns:
        attn (Tensor): attention map with added relative positional embeddings.
    """
    q_h, q_w = q_size
    k_h, k_w = k_size
    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
    Rw = get_rel_pos(q_w, k_w, rel_pos_w)

    B, _, dim = q.shape
    r_q = q.reshape(B, q_h, q_w, dim)
    rel_h = torch.einsum("bhwc,hkc->bhwk", r_q, Rh)
    rel_w = torch.einsum("bhwc,wkc->bhwk", r_q, Rw)

    attn = (
        attn.view(B, q_h, q_w, k_h, k_w) + rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]
    ).view(B, q_h * q_w, k_h * k_w)

    return attn


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

    def __init__(
        self,
        kernel_size: Tuple[int, int] = (16, 16),
        stride: Tuple[int, int] = (16, 16),
        padding: Tuple[int, int] = (0, 0),
        in_chans: int = 3,
        embed_dim: int = 768,
    ) -> None:
        """
        Args:
            kernel_size (Tuple): kernel size of the projection layer.
            stride (Tuple): stride of the projection layer.
            padding (Tuple): padding size of the projection layer.
            in_chans (int): Number of input image channels.
            embed_dim (int): Patch embedding dimension.
        """
        super().__init__()

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

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.proj(x)
        # B C H W -> B H W C
        x = x.permute(0, 2, 3, 1)
        return x

class SimpleFPN(nn.Module):
    def __init__(self, in_dim=768, out_dims=[128, 256, 512, 1024]):
        super().__init__()
        self.down_4_chan = max(out_dims[0]*2, in_dim // 2)
        self.down_4 = nn.Sequential(
            nn.ConvTranspose2d(in_dim, self.down_4_chan, 2, stride=2),
            nn.GroupNorm(1, self.down_4_chan),
            nn.GELU(),
            nn.ConvTranspose2d(self.down_4_chan, self.down_4_chan // 2, 2, stride=2),
            nn.GroupNorm(1, self.down_4_chan // 2),
            nn.Conv2d(self.down_4_chan // 2, out_dims[0], 1),
            nn.GroupNorm(1, out_dims[0]),
            nn.GELU()
        )
        self.down_8_chan = max(out_dims[1], in_dim // 2)
        self.down_8 = nn.Sequential(
            nn.ConvTranspose2d(in_dim, self.down_8_chan, 2, stride=2),
            nn.GroupNorm(1, self.down_8_chan),
            nn.Conv2d(self.down_8_chan, out_dims[1], 1),
            nn.GroupNorm(1, out_dims[1]),
            nn.GELU()
        )
        self.down_16 = nn.Sequential(
            nn.Conv2d(in_dim, out_dims[2], 1),
            nn.GroupNorm(1, out_dims[2]),
            nn.GELU()
        )
        self.down_32_chan = max(out_dims[3], in_dim * 2)
        self.down_32 = nn.Sequential(
            nn.Conv2d(in_dim, self.down_32_chan, 2, stride=2),
            nn.GroupNorm(1, self.down_32_chan),
            nn.Conv2d(self.down_32_chan, out_dims[3], 1),
            nn.GroupNorm(1, out_dims[3]),
            nn.GELU()
        )

        self.init_weights()

    def init_weights(self):
        # TODO
        pass

    def forward(self, x):
        x_down_4 = self.down_4(x)
        x_down_8 = self.down_8(x)
        x_down_16 = self.down_16(x)
        x_down_32 = self.down_32(x)

        return {
            'res2': x_down_4, 
            'res3': x_down_8, 
            'res4': x_down_16, 
            'res5': x_down_32
        }


class D2ViT(ImageEncoderViT, Backbone):
    def __init__(self, cfg, input_shape):    
        size = cfg['BACKBONE']['VIT']['SIZE']
        if size == "base":
            encoder_depth = 12
            encoder_embed_dim = 768
            encoder_num_heads = 12
            encoder_global_attn_indexes = [2, 5, 8, 11]
            neck_in_dim=768
            neck_out_dims=[128, 256, 512, 1024]
        elif size == "large":
            encoder_embed_dim = 1024
            encoder_depth = 24
            encoder_num_heads = 16
            encoder_global_attn_indexes = [5, 11, 17, 23]
            neck_in_dim=1024
            neck_out_dims=[128, 256, 512, 1024]            
        elif size == "huge":
            encoder_embed_dim = 1280
            encoder_depth = 32
            encoder_num_heads = 16
            encoder_global_attn_indexes = [7, 15, 23, 31]
            neck_in_dim=1280
            neck_out_dims=[128, 256, 512, 1024]    

        prompt_embed_dim = 256
        image_size = 1024
        vit_patch_size = 16
        image_embedding_size = image_size // vit_patch_size

        super().__init__(
            depth=encoder_depth,
            embed_dim=encoder_embed_dim,
            img_size=image_size,
            mlp_ratio=4,
            norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
            num_heads=encoder_num_heads,
            patch_size=vit_patch_size,
            qkv_bias=True,
            use_rel_pos=True,
            global_attn_indexes=encoder_global_attn_indexes,
            window_size=14,
            out_chans=prompt_embed_dim,
        )

        self.neck = SimpleFPN(in_dim=neck_in_dim, out_dims=neck_out_dims)

        self._out_features = cfg['BACKBONE']['VIT']['OUT_FEATURES']

        self._out_feature_strides = {
            "res2": 4,
            "res3": 8,
            "res4": 16,
            "res5": 32,
        }
        self._out_feature_channels = {
            "res2": neck_out_dims[0],
            "res3": neck_out_dims[1],
            "res4": neck_out_dims[2],
            "res5": neck_out_dims[3],
        }

    def forward(self, x):
        """
        Args:
            x: Tensor of shape (N,C,H,W). H, W must be a multiple of ``self.size_divisibility``.
        Returns:
            dict[str->Tensor]: names and the corresponding features
        """
        assert (
            x.dim() == 4
        ), f"SwinTransformer takes an input of shape (N, C, H, W). Got {x.shape} instead!"
        outputs = {}
        y = super().forward(x)
        for k in y.keys():
            if k in self._out_features:
                outputs[k] = y[k]
        return outputs

    def output_shape(self):
        return {
            name: ShapeSpec(
                channels=self._out_feature_channels[name], stride=self._out_feature_strides[name]
            )
            for name in self._out_features
        }

    def load_weights(self, pretrained_dict=None, pretrained_layers=[], verbose=True):
        model_dict = self.state_dict()
        pretrained_dict = pretrained_dict['model'] if 'model' in pretrained_dict else pretrained_dict
        pretrained_dict = {k.replace('image_encoder.', ''):v for k,v in pretrained_dict.items()}
        pretrained_dict = {
            k: v for k, v in pretrained_dict.items()
            if k in model_dict.keys()
        }
        need_init_state_dict = {}
        for k, v in pretrained_dict.items():
            need_init = (
                    (
                            k.split('.')[0] in pretrained_layers
                            or pretrained_layers[0] == '*'
                    )
                    and 'relative_position_index' not in k
                    and 'attn_mask' not in k
            )
            if need_init:
                need_init_state_dict[k] = v
        logger.info(f'=> loaded keys {need_init_state_dict.keys()}')    
        unloaded_keys = set(model_dict.keys()) - set(need_init_state_dict.keys())
        logger.info(f'=> unloaded keys {unloaded_keys}')        
        self.load_state_dict(need_init_state_dict, strict=False)

    @property
    def size_divisibility(self):
        return 32

@register_backbone
def get_vit_backbone(cfg):
    vit = D2ViT(cfg['MODEL'], 224)    

    if cfg['MODEL']['BACKBONE']['LOAD_PRETRAINED'] is True:
        filename = cfg['MODEL']['BACKBONE']['PRETRAINED']
        assert os.path.isfile(filename), f"=> no checkpoint found at '{filename}'"
        logger.info(f'=> init from {filename}')
        with PathManager.open(filename, "rb") as f:
            ckpt = torch.load(f)
        vit.load_weights(ckpt, cfg['MODEL']['BACKBONE']['VIT'].get('PRETRAINED_LAYERS', ['*']), cfg['VERBOSE'])

    return vit