src/sts/detectron2/modeling/backbone/fpn_swin.py

# Copyright (c) Facebook, Inc. and its affiliates.
import math
import fvcore.nn.weight_init as weight_init
import torch
import torch.nn.functional as F
from torch import nn

from detectron2.layers import Conv2d, ShapeSpec, get_norm

from .backbone import Backbone
from .build import BACKBONE_REGISTRY
from .resnet import build_resnet_backbone
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
import numpy as np

__all__ = ["build_resnet_fpn_backbone", "build_retinanet_resnet_fpn_backbone", "FPN"]

class Mlp(nn.Module):
    """ Multilayer perceptron."""

    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=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)
        x = self.fc2(x)
        x = self.drop(x)
        return x


def window_partition(x, window_size):
    """
    Args:
        x: (B, H, W, C)
        window_size (int): window size
    Returns:
        windows: (num_windows*B, window_size, window_size, C)
    """
    B, H, W, C = x.shape
    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows


def window_reverse(windows, window_size, H, W):
    """
    Args:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): Window size
        H (int): Height of image
        W (int): Width of image
    Returns:
        x: (B, H, W, C)
    """
    B = int(windows.shape[0] / (H * W / window_size / window_size))
    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
    return x


class WindowAttention(nn.Module):
    """ Window based multi-head self attention (W-MSA) module with relative position bias.
    It supports both of shifted and non-shifted window.
    Args:
        dim (int): Number of input channels.
        window_size (tuple[int]): The height and width of the window.
        num_heads (int): Number of attention heads.
        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
    """

    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):

        super().__init__()
        self.dim = dim
        self.window_size = window_size  # Wh, Ww
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        # define a parameter table of relative position bias
        self.relative_position_bias_table = nn.Parameter(
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH

        # get pair-wise relative position index for each token inside the window
        coords_h = torch.arange(self.window_size[0])
        coords_w = torch.arange(self.window_size[1])
        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.window_size[1] - 1
        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
        self.register_buffer("relative_position_index", relative_position_index)

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

        trunc_normal_(self.relative_position_bias_table, std=.02)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=None):
        """ Forward function.
        Args:
            x: input features with shape of (num_windows*B, N, C)
            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
        """
        B_, N, C = x.shape
        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)

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

        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH
        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
        attn = attn + relative_position_bias.unsqueeze(0)

        if mask is not None:
            nW = mask.shape[0]
            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
            attn = attn.view(-1, self.num_heads, N, N)
            attn = self.softmax(attn)
        else:
            attn = self.softmax(attn)

        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class SwinTransformerBlock(nn.Module):
    """ Swin Transformer Block.
    Args:
        dim (int): Number of input channels.
        num_heads (int): Number of attention heads.
        window_size (int): Window size.
        shift_size (int): Shift size for SW-MSA.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
        drop (float, optional): Dropout rate. Default: 0.0
        attn_drop (float, optional): Attention dropout rate. Default: 0.0
        drop_path (float, optional): Stochastic depth rate. Default: 0.0
        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """

    def __init__(self, dim, num_heads, window_size=7, shift_size=0,
                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
                 act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.window_size = window_size
        self.shift_size = shift_size
        self.mlp_ratio = mlp_ratio
        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"

        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention(
            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)

        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)

        self.H = None
        self.W = None

    def forward(self, x, mask_matrix):
        """ Forward function.
        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
            mask_matrix: Attention mask for cyclic shift.
        """
        B, L, C = x.shape
        H, W = self.H, self.W
        assert L == H * W, "input feature has wrong size"

        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # pad feature maps to multiples of window size
        pad_l = pad_t = 0
        pad_r = (self.window_size - W % self.window_size) % self.window_size
        pad_b = (self.window_size - H % self.window_size) % self.window_size
        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
        _, Hp, Wp, _ = x.shape

        # cyclic shift
        if self.shift_size > 0:
            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
            attn_mask = mask_matrix
        else:
            shifted_x = x
            attn_mask = None

        # partition windows
        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C

        # W-MSA/SW-MSA
        attn_windows = self.attn(x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C

        # merge windows
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C

        # reverse cyclic shift
        if self.shift_size > 0:
            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
        else:
            x = shifted_x

        if pad_r > 0 or pad_b > 0:
            x = x[:, :H, :W, :].contiguous()

        x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        return x

class swin_layer(nn.Module):
    """ A basic Swin Transformer layer for one stage.
    Args:
        dim (int): Number of feature channels
        depth (int): Depths of this stage.
        num_heads (int): Number of attention head.
        window_size (int): Local window size. Default: 7.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
        drop (float, optional): Dropout rate. Default: 0.0
        attn_drop (float, optional): Attention dropout rate. Default: 0.0
        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
    """

    def __init__(self,
                 dim,
                 depth,
                 num_heads,
                 window_size=7,
                 mlp_ratio=4.,
                 qkv_bias=True,
                 qk_scale=None,
                 drop=0.,
                 attn_drop=0.,
                 drop_path=0.,
                 norm_layer=nn.LayerNorm,
                 downsample=None,
                 use_checkpoint=False):
        super().__init__()
        self.window_size = window_size
        self.shift_size = window_size // 2
        self.depth = depth
        self.use_checkpoint = use_checkpoint

        # build blocks
        self.blocks = nn.ModuleList([
            SwinTransformerBlock(
                dim=dim,
                num_heads=num_heads,
                window_size=window_size,
                shift_size=0 if (i % 2 == 0) else window_size // 2,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                drop=drop,
                attn_drop=attn_drop,
                drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
                norm_layer=norm_layer)
            for i in range(depth)])

        # patch merging layer
        if downsample is not None:
            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
        else:
            self.downsample = None

    def forward(self, x, H, W):
        """ Forward function.
        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
        """

        # calculate attention mask for SW-MSA
        Hp = int(np.ceil(H / self.window_size)) * self.window_size
        Wp = int(np.ceil(W / self.window_size)) * self.window_size
        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
        h_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        w_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        cnt = 0
        for h in h_slices:
            for w in w_slices:
                img_mask[:, h, w, :] = cnt
                cnt += 1

        mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))

        for blk in self.blocks:
            blk.H, blk.W = H, W
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x, attn_mask)
            else:
                x = blk(x, attn_mask)
        if self.downsample is not None:
            x_down = self.downsample(x, H, W)
            Wh, Ww = (H + 1) // 2, (W + 1) // 2
            return x, H, W, x_down, Wh, Ww
        else:
            return x, H, W, x, H, W

class FPN(Backbone):
    """
    This module implements :paper:`FPN`.
    It creates pyramid features built on top of some input feature maps.
    """

    _fuse_type: torch.jit.Final[str]

    def __init__(
        self, bottom_up, in_features, out_channels, norm="", top_block=None, fuse_type="sum"
    ):
        """
        Args:
            bottom_up (Backbone): module representing the bottom up subnetwork.
                Must be a subclass of :class:`Backbone`. The multi-scale feature
                maps generated by the bottom up network, and listed in `in_features`,
                are used to generate FPN levels.
            in_features (list[str]): names of the input feature maps coming
                from the backbone to which FPN is attached. For example, if the
                backbone produces ["res2", "res3", "res4"], any *contiguous* sublist
                of these may be used; order must be from high to low resolution.
            out_channels (int): number of channels in the output feature maps.
            norm (str): the normalization to use.
            top_block (nn.Module or None): if provided, an extra operation will
                be performed on the output of the last (smallest resolution)
                FPN output, and the result will extend the result list. The top_block
                further downsamples the feature map. It must have an attribute
                "num_levels", meaning the number of extra FPN levels added by
                this block, and "in_feature", which is a string representing
                its input feature (e.g., p5).
            fuse_type (str): types for fusing the top down features and the lateral
                ones. It can be "sum" (default), which sums up element-wise; or "avg",
                which takes the element-wise mean of the two.
        """
        super(FPN, self).__init__()
        assert isinstance(bottom_up, Backbone)
        assert in_features, in_features

        # Feature map strides and channels from the bottom up network (e.g. ResNet)
        input_shapes = bottom_up.output_shape()
        strides = [input_shapes[f].stride for f in in_features]
        in_channels_per_feature = [input_shapes[f].channels for f in in_features]

        _assert_strides_are_log2_contiguous(strides)
        lateral_convs = []
        output_convs = []

        use_bias = norm == ""
        for idx, in_channels in enumerate(in_channels_per_feature):
            lateral_norm = get_norm(norm, out_channels)
            output_norm = get_norm(norm, out_channels)

            lateral_conv = Conv2d(
                in_channels, out_channels, kernel_size=1, bias=use_bias, norm=lateral_norm
            )
            # output_conv = Conv2d(
            #     out_channels,
            #     out_channels,
            #     kernel_size=3,
            #     stride=1,
            #     padding=1,
            #     bias=use_bias,
            #     norm=output_norm,
            # )
            output_conv = swin_layer( dim=out_channels,
                                        depth=1,
                                        num_heads=2,
                                        window_size=7)
            self.out_channels = out_channels
            weight_init.c2_xavier_fill(lateral_conv)
            # weight_init.c2_xavier_fill(output_conv)
            stage = int(math.log2(strides[idx]))
            self.add_module("fpn_lateral{}".format(stage), lateral_conv)
            self.add_module("fpn_output{}".format(stage), output_conv)

            lateral_convs.append(lateral_conv)
            output_convs.append(output_conv)
        # Place convs into top-down order (from low to high resolution)
        # to make the top-down computation in forward clearer.
        self.lateral_convs = lateral_convs[::-1]
        self.output_convs = output_convs[::-1]
        self.top_block = top_block
        self.in_features = tuple(in_features)
        self.bottom_up = bottom_up
        # Return feature names are "p<stage>", like ["p2", "p3", ..., "p6"]
        self._out_feature_strides = {"p{}".format(int(math.log2(s))): s for s in strides}
        # top block output feature maps.
        if self.top_block is not None:
            for s in range(stage, stage + self.top_block.num_levels):
                self._out_feature_strides["p{}".format(s + 1)] = 2 ** (s + 1)

        self._out_features = list(self._out_feature_strides.keys())
        self._out_feature_channels = {k: out_channels for k in self._out_features}
        self._size_divisibility = strides[-1]
        assert fuse_type in {"avg", "sum"}
        self._fuse_type = fuse_type

    @property
    def size_divisibility(self):
        return self._size_divisibility

    def forward(self, x):
        """
        Args:
            input (dict[str->Tensor]): mapping feature map name (e.g., "res5") to
                feature map tensor for each feature level in high to low resolution order.

        Returns:
            dict[str->Tensor]:
                mapping from feature map name to FPN feature map tensor
                in high to low resolution order. Returned feature names follow the FPN
                paper convention: "p<stage>", where stage has stride = 2 ** stage e.g.,
                ["p2", "p3", ..., "p6"].
        """
        bottom_up_features = self.bottom_up(x)
        results = []
        prev_features = self.lateral_convs[0](bottom_up_features[self.in_features[-1]])
        B, C, Wh, Ww = prev_features.size()
        prev_features = prev_features.flatten(2).transpose(1, 2)
        x_out, H, W, x, Wh, Ww = self.output_convs[0](prev_features, Wh, Ww)
        prev_features = x_out.transpose(1, 2).view(-1, self.out_channels, Wh, Ww)
        results.append(prev_features)
        # Reverse feature maps into top-down order (from low to high resolution)
        for idx, (lateral_conv, output_conv) in enumerate(
            zip(self.lateral_convs, self.output_convs)
        ):
            # Slicing of ModuleList is not supported https://github.com/pytorch/pytorch/issues/47336
            # Therefore we loop over all modules but skip the first one
            if idx > 0:
                features = self.in_features[-idx - 1]
                features = bottom_up_features[features]
                top_down_features = F.interpolate(prev_features, scale_factor=2.0, mode="nearest")
                lateral_features = lateral_conv(features)
                prev_features = lateral_features + top_down_features
                if self._fuse_type == "avg":
                    prev_features /= 2
                B, C, Wh, Ww = prev_features.size()
                prev_features = prev_features.flatten(2).transpose(1, 2)
                x_out, H, W, x, Wh, Ww = self.output_convs[0](prev_features, Wh, Ww)
                prev_features = x_out.transpose(1, 2).view(-1, self.out_channels, Wh, Ww)
                results.insert(0, prev_features)

        if self.top_block is not None:
            if self.top_block.in_feature in bottom_up_features:
                top_block_in_feature = bottom_up_features[self.top_block.in_feature]
            else:
                top_block_in_feature = results[self._out_features.index(self.top_block.in_feature)]
            results.extend(self.top_block(top_block_in_feature))
        assert len(self._out_features) == len(results)
        return {f: res for f, res in zip(self._out_features, results)}

    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 _assert_strides_are_log2_contiguous(strides):
    """
    Assert that each stride is 2x times its preceding stride, i.e. "contiguous in log2".
    """
    for i, stride in enumerate(strides[1:], 1):
        assert stride == 2 * strides[i - 1], "Strides {} {} are not log2 contiguous".format(
            stride, strides[i - 1]
        )


class LastLevelMaxPool(nn.Module):
    """
    This module is used in the original FPN to generate a downsampled
    P6 feature from P5.
    """

    def __init__(self):
        super().__init__()
        self.num_levels = 1
        self.in_feature = "p5"

    def forward(self, x):
        return [F.max_pool2d(x, kernel_size=1, stride=2, padding=0)]


class LastLevelP6P7(nn.Module):
    """
    This module is used in RetinaNet to generate extra layers, P6 and P7 from
    C5 feature.
    """

    def __init__(self, in_channels, out_channels, in_feature="res5"):
        super().__init__()
        self.num_levels = 2
        self.in_feature = in_feature
        self.p6 = nn.Conv2d(in_channels, out_channels, 3, 2, 1)
        self.p7 = nn.Conv2d(out_channels, out_channels, 3, 2, 1)
        for module in [self.p6, self.p7]:
            weight_init.c2_xavier_fill(module)

    def forward(self, c5):
        p6 = self.p6(c5)
        p7 = self.p7(F.relu(p6))
        return [p6, p7]


@BACKBONE_REGISTRY.register()
def build_resnet_fpn_backbone(cfg, input_shape: ShapeSpec):
    """
    Args:
        cfg: a detectron2 CfgNode

    Returns:
        backbone (Backbone): backbone module, must be a subclass of :class:`Backbone`.
    """
    bottom_up = build_resnet_backbone(cfg, input_shape)
    in_features = cfg.MODEL.FPN.IN_FEATURES
    out_channels = cfg.MODEL.FPN.OUT_CHANNELS
    backbone = FPN(
        bottom_up=bottom_up,
        in_features=in_features,
        out_channels=out_channels,
        norm=cfg.MODEL.FPN.NORM,
        top_block=LastLevelMaxPool(),
        fuse_type=cfg.MODEL.FPN.FUSE_TYPE,
    )
    return backbone


@BACKBONE_REGISTRY.register()
def build_retinanet_resnet_fpn_backbone(cfg, input_shape: ShapeSpec):
    """
    Args:
        cfg: a detectron2 CfgNode

    Returns:
        backbone (Backbone): backbone module, must be a subclass of :class:`Backbone`.
    """
    bottom_up = build_resnet_backbone(cfg, input_shape)
    in_features = cfg.MODEL.FPN.IN_FEATURES
    out_channels = cfg.MODEL.FPN.OUT_CHANNELS
    in_channels_p6p7 = bottom_up.output_shape()["res5"].channels
    backbone = FPN(
        bottom_up=bottom_up,
        in_features=in_features,
        out_channels=out_channels,
        norm=cfg.MODEL.FPN.NORM,
        top_block=LastLevelP6P7(in_channels_p6p7, out_channels),
        fuse_type=cfg.MODEL.FPN.FUSE_TYPE,
    )
    return backbone