diff --git a/README.md b/README.md index 45ee15ca..d9850bdb 100644 --- a/README.md +++ b/README.md @@ -104,6 +104,7 @@ Spandrel currently supports a limited amount of network architectures. If the ar - [DCTLSA](https://github.com/zengkun301/DCTLSA) | [Models](https://github.com/zengkun301/DCTLSA/tree/main/pretrained) - [ATD](https://github.com/LabShuHangGU/Adaptive-Token-Dictionary) | [Models](https://drive.google.com/drive/folders/1D3BvTS1xBcaU1mp50k3pBzUWb7qjRvmB?usp=sharing) - [AdaCode](https://github.com/kechunl/AdaCode) | [Models](https://github.com/kechunl/AdaCode/releases/tag/v0-pretrain_models) +- [DRCT](https://github.com/ming053l/DRCT) #### Face Restoration diff --git a/libs/spandrel/spandrel/__helpers/main_registry.py b/libs/spandrel/spandrel/__helpers/main_registry.py index 4d1e5384..2866ceb6 100644 --- a/libs/spandrel/spandrel/__helpers/main_registry.py +++ b/libs/spandrel/spandrel/__helpers/main_registry.py @@ -6,6 +6,7 @@ DAT, DCTLSA, DITN, + DRCT, ESRGAN, FBCNN, GFPGAN, @@ -75,5 +76,6 @@ ArchSupport.from_architecture(DRUNet.DRUNetArch()), ArchSupport.from_architecture(DnCNN.DnCNNArch()), ArchSupport.from_architecture(IPT.IPTArch()), + ArchSupport.from_architecture(DRCT.DRCTArch()), ArchSupport.from_architecture(ESRGAN.ESRGANArch()), ) diff --git a/libs/spandrel/spandrel/architectures/DRCT/__init__.py b/libs/spandrel/spandrel/architectures/DRCT/__init__.py new file mode 100644 index 00000000..c03cb217 --- /dev/null +++ b/libs/spandrel/spandrel/architectures/DRCT/__init__.py @@ -0,0 +1,174 @@ +import math + +from typing_extensions import override + +from spandrel.util import KeyCondition, get_seq_len + +from ...__helpers.model_descriptor import ( + Architecture, + ImageModelDescriptor, + SizeRequirements, + StateDict, +) +from .arch.drct_arch import DRCT + + +def _get_upscale_pixelshuffle( + state_dict: StateDict, key_prefix: str = "upsample" +) -> int: + upscale = 1 + + for i in range(0, 10, 2): + key = f"{key_prefix}.{i}.weight" + if key not in state_dict: + break + + shape = state_dict[key].shape + num_feat = shape[1] + upscale *= math.isqrt(shape[0] // num_feat) + + return upscale + + +class DRCTArch(Architecture[DRCT]): + def __init__(self) -> None: + super().__init__( + id="DRCT", + detect=KeyCondition.has_all( + "conv_first.weight", + "conv_first.bias", + "layers.0.swin1.norm1.weight", + "layers.0.swin1.norm1.bias", + "layers.0.swin1.attn.relative_position_bias_table", + "layers.0.swin1.attn.relative_position_index", + "layers.0.swin1.attn.qkv.weight", + "layers.0.swin1.attn.proj.weight", + "layers.0.swin1.attn.proj.bias", + "layers.0.swin1.norm2.weight", + "layers.0.swin1.mlp.fc1.weight", + "layers.0.swin1.mlp.fc1.bias", + "layers.0.swin1.mlp.fc2.weight", + "layers.0.adjust1.weight", + "layers.0.swin2.norm1.weight", + "layers.0.adjust2.weight", + "layers.0.swin3.norm1.weight", + "layers.0.adjust3.weight", + "layers.0.swin4.norm1.weight", + "layers.0.adjust4.weight", + "layers.0.swin5.norm1.weight", + "layers.0.adjust5.weight", + "norm.weight", + "norm.bias", + ), + ) + + @override + def load(self, state_dict: StateDict) -> ImageModelDescriptor[DRCT]: + # Defaults + img_size = 64 + patch_size = 1 # cannot be detected + in_chans = 3 + embed_dim = 180 + depths = (6, 6, 6, 6, 6, 6) + num_heads = (6, 6, 6, 6, 6, 6) + window_size = 16 + mlp_ratio = 2.0 + qkv_bias = True + ape = False + patch_norm = True + upscale = 2 + img_range = 1.0 # cannot be deduced from state_dict + upsampler = "" + resi_connection = "1conv" + gc = 32 + + # detect + in_chans = state_dict["conv_first.weight"].shape[1] + embed_dim = state_dict["conv_first.weight"].shape[0] + + num_layers = get_seq_len(state_dict, "layers") + depths = (6,) * num_layers + num_heads = [] + for i in range(num_layers): + num_heads.append( + state_dict[f"layers.{i}.swin1.attn.relative_position_bias_table"].shape[ + 1 + ] + ) + + mlp_ratio = state_dict["layers.0.swin1.mlp.fc1.weight"].shape[0] / embed_dim + + window_square = state_dict[ + "layers.0.swin1.attn.relative_position_bias_table" + ].shape[0] + window_size = (math.isqrt(window_square) + 1) // 2 + + if "conv_last.weight" in state_dict: + upsampler = "pixelshuffle" + upscale = _get_upscale_pixelshuffle(state_dict, "upsample") + else: + upsampler = "" + upscale = 1 + + if "conv_after_body.weight" in state_dict: + resi_connection = "1conv" + else: + resi_connection = "identity" + + qkv_bias = "layers.0.swin1.attn.qkv.bias" in state_dict + gc = state_dict["layers.0.adjust1.weight"].shape[0] + + patch_norm = "patch_embed.norm.weight" in state_dict + ape = "absolute_pos_embed" in state_dict + + if "layers.0.swin2.attn_mask" in state_dict: + img_size = ( + math.isqrt(state_dict["layers.0.swin2.attn_mask"].shape[0]) + * window_size + * patch_size + ) + else: + # we only know that the input size is <= window_size, + # so we just assume that the input size is window_size + img_size = window_size * patch_size + + model = DRCT( + img_size=img_size, + patch_size=patch_size, + in_chans=in_chans, + embed_dim=embed_dim, + depths=depths, + num_heads=num_heads, + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + ape=ape, + patch_norm=patch_norm, + upscale=upscale, + img_range=img_range, + upsampler=upsampler, + resi_connection=resi_connection, + gc=gc, + ) + + size_tag = ["large"] if len(depths) >= 10 else [] + tags = [ + *size_tag, + f"s{img_size}w{window_size}", + f"{embed_dim}dim", + f"{resi_connection}", + ] + + return ImageModelDescriptor( + model, + state_dict, + architecture=self, + purpose="Restoration" if upscale == 1 else "SR", + tags=tags, + supports_half=False, # Too much weirdness to support this at the moment + supports_bfloat16=True, + scale=upscale, + input_channels=in_chans, + output_channels=in_chans, + size_requirements=SizeRequirements(multiple_of=16), + ) diff --git a/libs/spandrel/spandrel/architectures/DRCT/arch/LICENSE b/libs/spandrel/spandrel/architectures/DRCT/arch/LICENSE new file mode 100644 index 00000000..7397ede8 --- /dev/null +++ b/libs/spandrel/spandrel/architectures/DRCT/arch/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2024 Chia-Ming Lee + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/libs/spandrel/spandrel/architectures/DRCT/arch/drct_arch.py b/libs/spandrel/spandrel/architectures/DRCT/arch/drct_arch.py new file mode 100644 index 00000000..151a818a --- /dev/null +++ b/libs/spandrel/spandrel/architectures/DRCT/arch/drct_arch.py @@ -0,0 +1,862 @@ +import math + +import torch +import torch.nn as nn + +from spandrel.util import store_hyperparameters +from spandrel.util.timm import DropPath, to_2tuple, trunc_normal_ + + +class ChannelAttention(nn.Module): + """Channel attention used in RCAN. + Args: + num_feat (int): Channel number of intermediate features. + squeeze_factor (int): Channel squeeze factor. Default: 16. + """ + + def __init__(self, num_feat, squeeze_factor=16): + super().__init__() + self.attention = nn.Sequential( + nn.AdaptiveAvgPool2d(1), + nn.Conv2d(num_feat, num_feat // squeeze_factor, 1, padding=0), + nn.ReLU(inplace=True), + nn.Conv2d(num_feat // squeeze_factor, num_feat, 1, padding=0), + nn.Sigmoid(), + ) + + def forward(self, x): + y = self.attention(x) + return x * y + + +class CAB(nn.Module): + def __init__(self, num_feat, compress_ratio=3, squeeze_factor=30): + super().__init__() + + self.cab = nn.Sequential( + nn.Conv2d(num_feat, num_feat // compress_ratio, 3, 1, 1), + nn.GELU(), + nn.Conv2d(num_feat // compress_ratio, num_feat, 3, 1, 1), + ChannelAttention(num_feat, squeeze_factor), + ) + + def forward(self, x): + return self.cab(x) + + +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) + 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): + r"""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=0.0, + proj_drop=0.0, + ): + 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( # type: ignore + 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], indexing="ij") + ) # 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=0.02) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + """ + 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) # type: ignore + ].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 + + def extra_repr(self) -> str: + return f"dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}" + + +class RDG(nn.Module): + def __init__( + self, + dim, + input_resolution, + num_heads, + window_size, + mlp_ratio, + qkv_bias, + qk_scale, + drop, + attn_drop, + drop_path, + norm_layer, + gc, + patch_size, + img_size, + ): + super().__init__() + + self.swin1 = SwinTransformerBlock( + dim=dim, + input_resolution=input_resolution, + num_heads=num_heads, + window_size=window_size, + shift_size=0, # For first block + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[0] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, # type: ignore + ) + self.adjust1 = nn.Conv2d(dim, gc, 1) + + self.swin2 = SwinTransformerBlock( + dim + gc, + input_resolution=input_resolution, + num_heads=num_heads - ((dim + gc) % num_heads), + window_size=window_size, + shift_size=window_size // 2, # For first block + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[0] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, # type: ignore + ) + self.adjust2 = nn.Conv2d(dim + gc, gc, 1) + + self.swin3 = SwinTransformerBlock( + dim + 2 * gc, + input_resolution=input_resolution, + num_heads=num_heads - ((dim + 2 * gc) % num_heads), + window_size=window_size, + shift_size=0, # For first block + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[0] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, # type: ignore + ) + self.adjust3 = nn.Conv2d(dim + gc * 2, gc, 1) + + self.swin4 = SwinTransformerBlock( + dim + 3 * gc, + input_resolution=input_resolution, + num_heads=num_heads - ((dim + 3 * gc) % num_heads), + window_size=window_size, + shift_size=window_size // 2, # For first block + mlp_ratio=1, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[0] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, # type: ignore + ) + self.adjust4 = nn.Conv2d(dim + gc * 3, gc, 1) + + self.swin5 = SwinTransformerBlock( + dim + 4 * gc, + input_resolution=input_resolution, + num_heads=num_heads - ((dim + 4 * gc) % num_heads), + window_size=window_size, + shift_size=0, # For first block + mlp_ratio=1, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[0] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, # type: ignore + ) + self.adjust5 = nn.Conv2d(dim + gc * 4, dim, 1) + + self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True) + + self.pe = PatchEmbed( + img_size=img_size, + patch_size=patch_size, + in_chans=0, + embed_dim=dim, + norm_layer=None, + ) + + self.pue = PatchUnEmbed( + img_size=img_size, + patch_size=patch_size, + in_chans=0, + embed_dim=dim, + norm_layer=None, + ) + + def forward(self, x, xsize): + x1 = self.pe(self.lrelu(self.adjust1(self.pue(self.swin1(x, xsize), xsize)))) + x2 = self.pe( + self.lrelu( + self.adjust2(self.pue(self.swin2(torch.cat((x, x1), -1), xsize), xsize)) + ) + ) + x3 = self.pe( + self.lrelu( + self.adjust3( + self.pue(self.swin3(torch.cat((x, x1, x2), -1), xsize), xsize) + ) + ) + ) + x4 = self.pe( + self.lrelu( + self.adjust4( + self.pue(self.swin4(torch.cat((x, x1, x2, x3), -1), xsize), xsize) + ) + ) + ) + x5 = self.pe( + self.adjust5( + self.pue(self.swin5(torch.cat((x, x1, x2, x3, x4), -1), xsize), xsize) + ) + ) + + return x5 * 0.2 + x + + +class SwinTransformerBlock(nn.Module): + r"""Swin Transformer Block. + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resulotion. + 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, + input_resolution, + num_heads, + window_size=7, + shift_size=0, + mlp_ratio=4.0, + qkv_bias=True, + qk_scale=None, + drop=0.0, + attn_drop=0.0, + drop_path=0.0, + act_layer=nn.GELU, + norm_layer=nn.LayerNorm, + ): + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + if min(self.input_resolution) <= self.window_size: + # if window size is larger than input resolution, we don't partition windows + self.shift_size = 0 + self.window_size = min(self.input_resolution) + 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 > 0.0 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 self.shift_size > 0: + attn_mask = self.calculate_mask(self.input_resolution) + else: + attn_mask = None + + self.register_buffer("attn_mask", attn_mask) + + def calculate_mask(self, x_size): + # calculate attention mask for SW-MSA + H, W = x_size + img_mask = torch.zeros((1, H, W, 1)) # 1 H W 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, -100.0).masked_fill( + attn_mask == 0, 0.0 + ) + + return attn_mask + + def forward(self, x, x_size): + H, W = x_size + B, _L, C = x.shape + # assert L == H * W, "input feature has wrong size" + + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + # cyclic shift + if self.shift_size > 0: + shifted_x = torch.roll( + x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2) + ) + else: + shifted_x = x + + # 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 (to be compatible for testing on images whose shapes are the multiple of window size + if self.input_resolution == x_size: + attn_windows = self.attn( + x_windows, mask=self.attn_mask + ) # nW*B, window_size*window_size, C + else: + attn_windows = self.attn( + x_windows, mask=self.calculate_mask(x_size).to(x.device) + ) + + # merge windows + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, H, W) # 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 + 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 + + def extra_repr(self) -> str: + return ( + f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " + f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}" + ) + + +class PatchMerging(nn.Module): + r"""Patch Merging Layer. + + Args: + input_resolution (tuple[int]): Resolution of input feature. + dim (int): Number of input channels. + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, input_resolution, dim: int, norm_layer=nn.LayerNorm): + super().__init__() + self.input_resolution = input_resolution + self.dim = dim + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + self.norm = norm_layer(4 * dim) + + def forward(self, x): + """ + x: b, h*w, c + """ + h, w = self.input_resolution + b, seq_len, c = x.shape + assert seq_len == h * w, "input feature has wrong size" + assert h % 2 == 0 and w % 2 == 0, f"x size ({h}*{w}) are not even." + + x = x.view(b, h, w, c) + + x0 = x[:, 0::2, 0::2, :] # b h/2 w/2 c + x1 = x[:, 1::2, 0::2, :] # b h/2 w/2 c + x2 = x[:, 0::2, 1::2, :] # b h/2 w/2 c + x3 = x[:, 1::2, 1::2, :] # b h/2 w/2 c + x = torch.cat([x0, x1, x2, x3], -1) # b h/2 w/2 4*c + x = x.view(b, -1, 4 * c) # b h/2*w/2 4*c + + x = self.norm(x) + x = self.reduction(x) + + return x + + +class PatchEmbed(nn.Module): + r"""Image to Patch Embedding + + Args: + img_size (int): Image size. Default: 224. + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__( + self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None + ): + super().__init__() + img_size = to_2tuple(img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [ + img_size[0] // patch_size[0], + img_size[1] // patch_size[1], + ] + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = patches_resolution[0] * patches_resolution[1] + + self.in_chans = in_chans + self.embed_dim = embed_dim + + if norm_layer is not None: + self.norm = norm_layer(embed_dim) + else: + self.norm = None + + def forward(self, x): + x = x.flatten(2).transpose(1, 2) # structured as [B, num_patches, C] + if self.norm is not None: + x = self.norm(x) # 归一化 + return x + + +class PatchUnEmbed(nn.Module): + r"""Image to Patch Unembedding + + args: + img_size (int): image size, default 224*224. + patch_size (int): Patch token sizd, default 4*4. + in_chans (int): num image channels, default 3. + embed_dim (int): num channels for linear projection output, default 96 + norm_layer (nn.Module, optional): normalization layer, default None. + """ + + def __init__( + self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None + ): + super().__init__() + img_size = to_2tuple(img_size) # 图像的大小，默认为 224*224 + patch_size = to_2tuple(patch_size) # Patch token 的大小，默认为 4*4 + patches_resolution = [ + img_size[0] // patch_size[0], + img_size[1] // patch_size[1], + ] # patch 的分辨率 + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = ( + patches_resolution[0] * patches_resolution[1] + ) # patch 的个数，num_patches + + self.in_chans = in_chans # 输入图像的通道数 + self.embed_dim = embed_dim # 线性 projection 输出的通道数 + + def forward(self, x, x_size): + B, _HW, _C = x.shape # 输入 x 的结构 + x = x.transpose(1, 2).view( + B, -1, x_size[0], x_size[1] + ) # 输出结构为 [B, Ph*Pw, C] + return x + + +class Upsample(nn.Sequential): + """Upsample module. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + """ + + def __init__(self, scale, num_feat): + m = [] + if (scale & (scale - 1)) == 0: # scale = 2^n + for _ in range(int(math.log2(scale))): + m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(2)) + elif scale == 3: + m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(3)) + else: + raise ValueError( + f"scale {scale} is not supported. " "Supported scales: 2^n and 3." + ) + super().__init__(*m) + + +@store_hyperparameters() +class DRCT(nn.Module): + hyperparameters = {} + + def __init__( + self, + img_size=64, + patch_size=1, + in_chans=3, + embed_dim=180, + depths=(6, 6, 6, 6, 6, 6), + num_heads=(6, 6, 6, 6, 6, 6), + window_size=16, + mlp_ratio=2.0, + qkv_bias=True, + qk_scale=None, + drop_rate=0.0, + attn_drop_rate=0.0, + drop_path_rate=0.1, + norm_layer=nn.LayerNorm, + ape=False, + patch_norm=True, + upscale=1, + img_range=1.0, + upsampler="", + resi_connection="1conv", + gc=32, + ): + super().__init__() + + self.window_size = window_size + self.shift_size = window_size // 2 + + num_in_ch = in_chans + num_out_ch = in_chans + num_feat = 64 + self.img_range = img_range + if in_chans == 3: + rgb_mean = (0.4488, 0.4371, 0.4040) + self.mean = torch.Tensor(rgb_mean).view(1, 3, 1, 1) + else: + self.mean = torch.zeros(1, 1, 1, 1) + self.upscale = upscale + self.upsampler = upsampler + + # ------------------------- 1, shallow feature extraction ------------------------- # + self.conv_first = nn.Conv2d(num_in_ch, embed_dim, 3, 1, 1) + + # ------------------------- 2, deep feature extraction ------------------------- # + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.ape = ape + self.patch_norm = patch_norm + self.num_features = embed_dim + self.mlp_ratio = mlp_ratio + + # split image into non-overlapping patches + self.patch_embed = PatchEmbed( + img_size=img_size, + patch_size=patch_size, + in_chans=embed_dim, + embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None, + ) + num_patches = self.patch_embed.num_patches + patches_resolution = self.patch_embed.patches_resolution + self.patches_resolution = patches_resolution + + # merge non-overlapping patches into image + self.patch_unembed = PatchUnEmbed( + img_size=img_size, + patch_size=patch_size, + in_chans=embed_dim, + embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None, + ) + + # absolute position embedding + if self.ape: + self.absolute_pos_embed = nn.Parameter( # type: ignore + torch.zeros(1, num_patches, embed_dim) + ) + trunc_normal_(self.absolute_pos_embed, std=0.02) + + self.pos_drop = nn.Dropout(p=drop_rate) + + # stochastic depth + dpr = [ + x.item() for x in torch.linspace(0, drop_path_rate, sum(depths)) + ] # stochastic depth decay rule + + # build + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = RDG( + dim=embed_dim, + input_resolution=(patches_resolution[0], patches_resolution[1]), + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop_rate, + attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]) : sum(depths[: i_layer + 1])], + norm_layer=norm_layer, + gc=gc, + img_size=img_size, + patch_size=patch_size, + ) + + self.layers.append(layer) + self.norm = norm_layer(self.num_features) + + # build the last conv layer in deep feature extraction + if resi_connection == "1conv": + self.conv_after_body = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1) + elif resi_connection == "identity": + self.conv_after_body = nn.Identity() + + # ------------------------- 3, high quality image reconstruction ------------------------- # + if self.upsampler == "pixelshuffle": + # for classical SR + self.conv_before_upsample = nn.Sequential( + nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True) + ) + self.upsample = Upsample(upscale, num_feat) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + + self.apply(self._init_weights) + + 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: # type: ignore + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) + + @torch.jit.ignore # type: ignore + def no_weight_decay(self): + return {"absolute_pos_embed"} + + @torch.jit.ignore # type: ignore + def no_weight_decay_keywords(self): + return {"relative_position_bias_table"} + + def forward_features(self, x): + x_size = (x.shape[2], x.shape[3]) + + x = self.patch_embed(x) + if self.ape: + x = x + self.absolute_pos_embed + x = self.pos_drop(x) + + for layer in self.layers: + x = layer(x, x_size) + + x = self.norm(x) # b seq_len c + x = self.patch_unembed(x, x_size) + + return x + + def forward(self, x): + self.mean = self.mean.type_as(x) + x = (x - self.mean) * self.img_range + + if self.upsampler == "pixelshuffle": + # for classical SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + x = self.conv_last(self.upsample(x)) + + x = x / self.img_range + self.mean + + return x diff --git a/tests/__snapshots__/test_DRCT.ambr b/tests/__snapshots__/test_DRCT.ambr new file mode 100644 index 00000000..e8511909 --- /dev/null +++ b/tests/__snapshots__/test_DRCT.ambr @@ -0,0 +1,23 @@ +# serializer version: 1 +# name: test_community_model + ImageModelDescriptor( + architecture=DRCTArch( + id='DRCT', + name='DRCT', + ), + input_channels=3, + output_channels=3, + purpose='SR', + scale=4, + size_requirements=SizeRequirements(minimum=0, multiple_of=16, square=False), + supports_bfloat16=True, + supports_half=False, + tags=list([ + 'large', + 's64w16', + '180dim', + '1conv', + ]), + tiling=, + ) +# --- diff --git a/tests/images/outputs/16x16/4xRealWebPhoto_v4_drct-l.png b/tests/images/outputs/16x16/4xRealWebPhoto_v4_drct-l.png new file mode 100644 index 00000000..94fe60f8 Binary files /dev/null and b/tests/images/outputs/16x16/4xRealWebPhoto_v4_drct-l.png differ diff --git a/tests/images/outputs/32x32/4xRealWebPhoto_v4_drct-l.png b/tests/images/outputs/32x32/4xRealWebPhoto_v4_drct-l.png new file mode 100644 index 00000000..5de78f87 Binary files /dev/null and b/tests/images/outputs/32x32/4xRealWebPhoto_v4_drct-l.png differ diff --git a/tests/test_DRCT.py b/tests/test_DRCT.py new file mode 100644 index 00000000..4912b328 --- /dev/null +++ b/tests/test_DRCT.py @@ -0,0 +1,53 @@ +from spandrel.architectures.DRCT import DRCT, DRCTArch + +from .util import ( + ModelFile, + TestImage, + assert_image_inference, + assert_loads_correctly, + assert_size_requirements, + disallowed_props, + skip_if_unchanged, +) + +skip_if_unchanged(__file__) + + +def test_load(): + assert_loads_correctly( + DRCTArch(), + lambda: DRCT(), + lambda: DRCT(in_chans=4, embed_dim=60), + lambda: DRCT(upsampler="pixelshuffle", upscale=2, resi_connection="1conv"), + lambda: DRCT(upsampler="pixelshuffle", upscale=4, resi_connection="1conv"), + lambda: DRCT(upsampler="", upscale=1, resi_connection="identity"), + lambda: DRCT(qkv_bias=False), + lambda: DRCT(gc=16), + lambda: DRCT(ape=True, patch_norm=False), + lambda: DRCT(mlp_ratio=4.0), + lambda: DRCT(window_size=8), + lambda: DRCT(depths=[6, 6, 6, 6], num_heads=[6, 4, 6, 3]), + lambda: DRCT(img_size=32), + lambda: DRCT(img_size=16), + ) + + +def test_size_requirements(): + file = ModelFile.from_url( + "https://github.com/Phhofm/models/releases/download/4xRealWebPhoto_v4_drct-l/4xRealWebPhoto_v4_drct-l.pth" + ) + assert_size_requirements(file.load_model()) + + +def test_community_model(snapshot): + file = ModelFile.from_url( + "https://github.com/Phhofm/models/releases/download/4xRealWebPhoto_v4_drct-l/4xRealWebPhoto_v4_drct-l.pth", + ) + model = file.load_model() + assert model == snapshot(exclude=disallowed_props) + assert isinstance(model.model, DRCT) + assert_image_inference( + file, + model, + [TestImage.SR_16, TestImage.SR_32], + )