model.py

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


class DropPath(nn.Module):
    def __init__(self, drop_prob: float = 0.0):
        super().__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        if self.drop_prob == 0.0 or not self.training:
            return x

        keep_prob = 1.0 - self.drop_prob
        shape = (x.shape[0],) + (1,) * (x.ndim - 1)
        random_tensor = torch.rand(shape, dtype=x.dtype, device=x.device)
        random_tensor = torch.floor(random_tensor + keep_prob)
        return x * random_tensor / keep_prob


def window_partition(x, window_size):
    # x: [B, H, W, 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()
    windows = windows.view(-1, window_size, window_size, C)
    return windows


def window_reverse(windows, window_size, H, W):
    num_windows = (H // window_size) * (W // window_size)
    B = windows.shape[0] // num_windows

    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()
    x = x.view(B, H, W, -1)
    return x


class Mlp(nn.Module):
    def __init__(self, dim, mlp_ratio=4.0):
        super().__init__()
        hidden = int(dim * mlp_ratio)
        self.fc1 = nn.Linear(dim, hidden)
        self.act = nn.GELU()
        self.fc2 = nn.Linear(hidden, dim)

    def forward(self, x):
        return self.fc2(self.act(self.fc1(x)))


class WindowAttention(nn.Module):
    def __init__(self, dim, window_size, num_heads):
        super().__init__()
        assert dim % num_heads == 0, "dim должен делиться на num_heads"

        self.dim = dim
        self.window_size = window_size
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim ** -0.5

        self.relative_position_bias_table = nn.Parameter(
            torch.zeros((2 * window_size - 1) * (2 * window_size - 1), num_heads)
        )
        nn.init.trunc_normal_(self.relative_position_bias_table, std=0.02)

        coords_h = torch.arange(window_size)
        coords_w = torch.arange(window_size)
        coords = torch.stack(torch.meshgrid(coords_h, coords_w, indexing="ij"))
        coords_flatten = torch.flatten(coords, 1)

        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()

        relative_coords[:, :, 0] += window_size - 1
        relative_coords[:, :, 1] += window_size - 1
        relative_coords[:, :, 0] *= 2 * window_size - 1

        relative_position_index = relative_coords.sum(-1)
        self.register_buffer("relative_position_index", relative_position_index)

        self.qkv = nn.Linear(dim, dim * 3)
        self.proj = nn.Linear(dim, dim)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=None):
        # x: [B_windows, N, C]
        B_, N, C = x.shape

        qkv = self.qkv(x)
        qkv = qkv.reshape(B_, N, 3, self.num_heads, C // self.num_heads)
        qkv = qkv.permute(2, 0, 3, 1, 4)

        q, k, v = qkv.unbind(0)

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

        relative_position_bias = self.relative_position_bias_table[
            self.relative_position_index.view(-1)
        ].view(
            self.window_size * self.window_size,
            self.window_size * self.window_size,
            -1,
        )
        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
        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)
            attn = attn + mask.unsqueeze(1).unsqueeze(0)
            attn = attn.view(-1, self.num_heads, N, N)

        attn = self.softmax(attn)

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


class SwinBlock(nn.Module):
    def __init__(
        self,
        dim,
        num_heads,
        window_size,
        shift_size=0,
        mlp_ratio=4.0,
        drop_path=0.0,
    ):
        super().__init__()
        self.window_size = window_size
        self.shift_size = shift_size

        self.norm1 = nn.LayerNorm(dim)
        self.attn = WindowAttention(
            dim=dim,
            window_size=window_size,
            num_heads=num_heads,
        )
        self.drop_path1 = DropPath(drop_path)
        self.drop_path2 = DropPath(drop_path)

        self.norm2 = nn.LayerNorm(dim)
        self.mlp = Mlp(dim=dim, mlp_ratio=mlp_ratio)

    def _make_mask(self, H, W, shift_size, device):
        img_mask = torch.zeros((1, H, W, 1), device=device)

        h_slices = (
            slice(0, -self.window_size),
            slice(-self.window_size, -shift_size),
            slice(-shift_size, None),
        )
        w_slices = (
            slice(0, -self.window_size),
            slice(-self.window_size, -shift_size),
            slice(-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)
        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, -100.0)
        attn_mask = attn_mask.masked_fill(attn_mask == 0, 0.0)
        return attn_mask

    def forward(self, x, H, W):
        # x: [B, H*W, C]
        B, L, C = x.shape

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

        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, 0, pad_r, 0, pad_b))

        _, Hp, Wp, _ = x.shape

        shift_size = self.shift_size if min(Hp, Wp) > self.window_size else 0

        if shift_size > 0:
            shifted = torch.roll(x, shifts=(-shift_size, -shift_size), dims=(1, 2))
            attn_mask = self._make_mask(Hp, Wp, shift_size, x.device)
        else:
            shifted = x
            attn_mask = None

        x_windows = window_partition(shifted, self.window_size)
        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)

        attn_windows = self.attn(x_windows, mask=attn_mask)
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)

        shifted = window_reverse(attn_windows, self.window_size, Hp, Wp)

        if shift_size > 0:
            x = torch.roll(shifted, shifts=(shift_size, shift_size), dims=(1, 2))
        else:
            x = shifted

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

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

        x = shortcut + self.drop_path1(x)
        x = x + self.drop_path2(self.mlp(self.norm2(x)))

        return x


class UpscalerTransformer(nn.Module):
    """
    Лёгкая трансформерная сеть для апскейла x2.

    Вход:  [B, 3, 128, 128]
    Выход: [B, 3, 256, 256]
    """

    def __init__(
        self,
        dim=256,
        depth=8,
        num_heads=8,
        window_size=8,
        mlp_ratio=4.0,
    ):
        super().__init__()
        assert dim % num_heads == 0, "dim должен делиться на num_heads"

        self.conv_first = nn.Conv2d(3, dim, kernel_size=3, padding=1)

        self.layers = nn.ModuleList(
            [
                SwinBlock(
                    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,
                )
                for i in range(depth)
            ]
        )

        self.norm = nn.LayerNorm(dim)
        self.conv_feat = nn.Conv2d(dim, dim, kernel_size=3, padding=1)

        # PixelShuffle x2:
        # conv делает 3 * 2 * 2 = 12 каналов,
        # PixelShuffle собирает из них RGB x2 размера.
        self.up = nn.Sequential(
            nn.Conv2d(dim, 3 * 4, kernel_size=3, padding=1),
            nn.PixelShuffle(2),
        )

        self.conv_last = nn.Conv2d(3, 3, kernel_size=3, padding=1)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            nn.init.trunc_normal_(m.weight, std=0.02)
            if m.bias is not None:
                nn.init.constant_(m.bias, 0.0)

        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0.0)
            nn.init.constant_(m.weight, 1.0)

        elif isinstance(m, nn.Conv2d):
            nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
            if m.bias is not None:
                nn.init.constant_(m.bias, 0.0)

    def forward(self, x):
        base = F.interpolate(x, scale_factor=2, mode="bicubic", align_corners=False)

        f = self.conv_first(x)
        shortcut = f

        B, C, H, W = f.shape
        f = f.flatten(2).transpose(1, 2)

        for layer in self.layers:
            f = layer(f, H, W)

        f = self.norm(f)
        f = f.transpose(1, 2).reshape(B, C, H, W)

        f = self.conv_feat(f + shortcut)

        out = self.up(f)
        out = self.conv_last(out)

        return base + out