# -------------------------------------------------------- # FocalNet for Semantic Segmentation # Copyright (c) 2022 Microsoft # Licensed under The MIT License [see LICENSE for details] # Written by Jianwei Yang # -------------------------------------------------------- import math import time import numpy as np import logging import torch import torch.nn as nn import torch.nn.functional as F import torch.utils.checkpoint as checkpoint from timm.models.layers import DropPath, to_2tuple, trunc_normal_ from detectron2.utils.file_io import PathManager from detectron2.modeling import BACKBONE_REGISTRY, Backbone, ShapeSpec from .registry import register_backbone logger = logging.getLogger(__name__) 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 class FocalModulation(nn.Module): """ Focal Modulation Args: dim (int): Number of input channels. proj_drop (float, optional): Dropout ratio of output. Default: 0.0 focal_level (int): Number of focal levels focal_window (int): Focal window size at focal level 1 focal_factor (int, default=2): Step to increase the focal window use_postln (bool, default=False): Whether use post-modulation layernorm """ def __init__(self, dim, proj_drop=0., focal_level=2, focal_window=7, focal_factor=2, use_postln=False, use_postln_in_modulation=False, scaling_modulator=False): super().__init__() self.dim = dim # specific args for focalv3 self.focal_level = focal_level self.focal_window = focal_window self.focal_factor = focal_factor self.use_postln_in_modulation = use_postln_in_modulation self.scaling_modulator = scaling_modulator self.f = nn.Linear(dim, 2*dim+(self.focal_level+1), bias=True) self.h = nn.Conv2d(dim, dim, kernel_size=1, stride=1, padding=0, groups=1, bias=True) self.act = nn.GELU() self.proj = nn.Linear(dim, dim) self.proj_drop = nn.Dropout(proj_drop) self.focal_layers = nn.ModuleList() if self.use_postln_in_modulation: self.ln = nn.LayerNorm(dim) for k in range(self.focal_level): kernel_size = self.focal_factor*k + self.focal_window self.focal_layers.append( nn.Sequential( nn.Conv2d(dim, dim, kernel_size=kernel_size, stride=1, groups=dim, padding=kernel_size//2, bias=False), nn.GELU(), ) ) def forward(self, x): """ Forward function. Args: x: input features with shape of (B, H, W, C) """ B, nH, nW, C = x.shape x = self.f(x) x = x.permute(0, 3, 1, 2).contiguous() q, ctx, gates = torch.split(x, (C, C, self.focal_level+1), 1) ctx_all = 0 for l in range(self.focal_level): ctx = self.focal_layers[l](ctx) ctx_all = ctx_all + ctx*gates[:, l:l+1] ctx_global = self.act(ctx.mean(2, keepdim=True).mean(3, keepdim=True)) ctx_all = ctx_all + ctx_global*gates[:,self.focal_level:] if self.scaling_modulator: ctx_all = ctx_all / (self.focal_level + 1) x_out = q * self.h(ctx_all) x_out = x_out.permute(0, 2, 3, 1).contiguous() if self.use_postln_in_modulation: x_out = self.ln(x_out) x_out = self.proj(x_out) x_out = self.proj_drop(x_out) return x_out class FocalModulationBlock(nn.Module): """ Focal Modulation Block. Args: dim (int): Number of input channels. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. drop (float, optional): 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 focal_level (int): number of focal levels focal_window (int): focal kernel size at level 1 """ def __init__(self, dim, mlp_ratio=4., drop=0., drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, focal_level=2, focal_window=9, use_postln=False, use_postln_in_modulation=False, scaling_modulator=False, use_layerscale=False, layerscale_value=1e-4): super().__init__() self.dim = dim self.mlp_ratio = mlp_ratio self.focal_window = focal_window self.focal_level = focal_level self.use_postln = use_postln self.use_layerscale = use_layerscale self.norm1 = norm_layer(dim) self.modulation = FocalModulation( dim, focal_window=self.focal_window, focal_level=self.focal_level, proj_drop=drop, use_postln_in_modulation=use_postln_in_modulation, scaling_modulator=scaling_modulator ) 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 self.gamma_1 = 1.0 self.gamma_2 = 1.0 if self.use_layerscale: self.gamma_1 = nn.Parameter(layerscale_value * torch.ones((dim)), requires_grad=True) self.gamma_2 = nn.Parameter(layerscale_value * torch.ones((dim)), requires_grad=True) def forward(self, x): """ Forward function. Args: x: Input feature, tensor size (B, H*W, C). H, W: Spatial resolution of the input feature. """ B, L, C = x.shape H, W = self.H, self.W assert L == H * W, "input feature has wrong size" shortcut = x if not self.use_postln: x = self.norm1(x) x = x.view(B, H, W, C) # FM x = self.modulation(x).view(B, H * W, C) if self.use_postln: x = self.norm1(x) # FFN x = shortcut + self.drop_path(self.gamma_1 * x) if self.use_postln: x = x + self.drop_path(self.gamma_2 * self.norm2(self.mlp(x))) else: x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x))) return x class BasicLayer(nn.Module): """ A basic focal modulation layer for one stage. Args: dim (int): Number of feature channels depth (int): Depths of this stage. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4. drop (float, optional): 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 focal_level (int): Number of focal levels focal_window (int): Focal window size at focal level 1 use_conv_embed (bool): Use overlapped convolution for patch embedding or now. Default: False use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False """ def __init__(self, dim, depth, mlp_ratio=4., drop=0., drop_path=0., norm_layer=nn.LayerNorm, downsample=None, focal_window=9, focal_level=2, use_conv_embed=False, use_postln=False, use_postln_in_modulation=False, scaling_modulator=False, use_layerscale=False, use_checkpoint=False ): super().__init__() self.depth = depth self.use_checkpoint = use_checkpoint # build blocks self.blocks = nn.ModuleList([ FocalModulationBlock( dim=dim, mlp_ratio=mlp_ratio, drop=drop, drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path, focal_window=focal_window, focal_level=focal_level, use_postln=use_postln, use_postln_in_modulation=use_postln_in_modulation, scaling_modulator=scaling_modulator, use_layerscale=use_layerscale, norm_layer=norm_layer) for i in range(depth)]) # patch merging layer if downsample is not None: self.downsample = downsample( patch_size=2, in_chans=dim, embed_dim=2*dim, use_conv_embed=use_conv_embed, norm_layer=norm_layer, is_stem=False ) 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. """ for blk in self.blocks: blk.H, blk.W = H, W if self.use_checkpoint: x = checkpoint.checkpoint(blk, x) else: x = blk(x) if self.downsample is not None: x_reshaped = x.transpose(1, 2).view(x.shape[0], x.shape[-1], H, W) x_down = self.downsample(x_reshaped) x_down = x_down.flatten(2).transpose(1, 2) 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 PatchEmbed(nn.Module): """ Image to Patch Embedding Args: 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 use_conv_embed (bool): Whether use overlapped convolution for patch embedding. Default: False is_stem (bool): Is the stem block or not. """ def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None, use_conv_embed=False, is_stem=False): super().__init__() patch_size = to_2tuple(patch_size) self.patch_size = patch_size self.in_chans = in_chans self.embed_dim = embed_dim if use_conv_embed: # if we choose to use conv embedding, then we treat the stem and non-stem differently if is_stem: kernel_size = 7; padding = 2; stride = 4 else: kernel_size = 3; padding = 1; stride = 2 self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=kernel_size, stride=stride, padding=padding) else: self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) if norm_layer is not None: self.norm = norm_layer(embed_dim) else: self.norm = None def forward(self, x): """Forward function.""" _, _, H, W = x.size() if W % self.patch_size[1] != 0: x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1])) if H % self.patch_size[0] != 0: x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0])) x = self.proj(x) # B C Wh Ww if self.norm is not None: Wh, Ww = x.size(2), x.size(3) x = x.flatten(2).transpose(1, 2) x = self.norm(x) x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww) return x class FocalNet(nn.Module): """ FocalNet backbone. Args: pretrain_img_size (int): Input image size for training the pretrained model, used in absolute postion embedding. Default 224. patch_size (int | tuple(int)): Patch size. Default: 4. in_chans (int): Number of input image channels. Default: 3. embed_dim (int): Number of linear projection output channels. Default: 96. depths (tuple[int]): Depths of each Swin Transformer stage. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4. drop_rate (float): Dropout rate. drop_path_rate (float): Stochastic depth rate. Default: 0.2. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. patch_norm (bool): If True, add normalization after patch embedding. Default: True. out_indices (Sequence[int]): Output from which stages. frozen_stages (int): Stages to be frozen (stop grad and set eval mode). -1 means not freezing any parameters. focal_levels (Sequence[int]): Number of focal levels at four stages focal_windows (Sequence[int]): Focal window sizes at first focal level at four stages use_conv_embed (bool): Whether use overlapped convolution for patch embedding use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. """ def __init__(self, pretrain_img_size=1600, patch_size=4, in_chans=3, embed_dim=96, depths=[2, 2, 6, 2], mlp_ratio=4., drop_rate=0., drop_path_rate=0.2, norm_layer=nn.LayerNorm, patch_norm=True, out_indices=[0, 1, 2, 3], frozen_stages=-1, focal_levels=[2,2,2,2], focal_windows=[9,9,9,9], use_conv_embed=False, use_postln=False, use_postln_in_modulation=False, scaling_modulator=False, use_layerscale=False, use_checkpoint=False, ): super().__init__() self.pretrain_img_size = pretrain_img_size self.num_layers = len(depths) self.embed_dim = embed_dim self.patch_norm = patch_norm self.out_indices = out_indices self.frozen_stages = frozen_stages # split image into non-overlapping patches self.patch_embed = PatchEmbed( patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim, norm_layer=norm_layer if self.patch_norm else None, use_conv_embed=use_conv_embed, is_stem=True) 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 layers self.layers = nn.ModuleList() for i_layer in range(self.num_layers): layer = BasicLayer( dim=int(embed_dim * 2 ** i_layer), depth=depths[i_layer], mlp_ratio=mlp_ratio, drop=drop_rate, drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], norm_layer=norm_layer, downsample=PatchEmbed if (i_layer < self.num_layers - 1) else None, focal_window=focal_windows[i_layer], focal_level=focal_levels[i_layer], use_conv_embed=use_conv_embed, use_postln=use_postln, use_postln_in_modulation=use_postln_in_modulation, scaling_modulator=scaling_modulator, use_layerscale=use_layerscale, use_checkpoint=use_checkpoint) self.layers.append(layer) num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)] self.num_features = num_features # add a norm layer for each output for i_layer in out_indices: layer = norm_layer(num_features[i_layer]) layer_name = f'norm{i_layer}' self.add_module(layer_name, layer) self._freeze_stages() def _freeze_stages(self): if self.frozen_stages >= 0: self.patch_embed.eval() for param in self.patch_embed.parameters(): param.requires_grad = False if self.frozen_stages >= 2: self.pos_drop.eval() for i in range(0, self.frozen_stages - 1): m = self.layers[i] m.eval() for param in m.parameters(): param.requires_grad = False def init_weights(self, pretrained=None): """Initialize the weights in backbone. Args: pretrained (str, optional): Path to pre-trained weights. Defaults to None. """ def _init_weights(m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) if isinstance(pretrained, str): self.apply(_init_weights) logger = get_root_logger() load_checkpoint(self, pretrained, strict=False, logger=logger) elif pretrained is None: self.apply(_init_weights) else: raise TypeError('pretrained must be a str or None') def load_weights(self, pretrained_dict=None, pretrained_layers=[], verbose=True): model_dict = self.state_dict() missed_dict = [k for k in model_dict.keys() if k not in pretrained_dict] logger.info(f'=> Missed keys {missed_dict}') unexpected_dict = [k for k in pretrained_dict.keys() if k not in model_dict] logger.info(f'=> Unexpected keys {unexpected_dict}') 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: # if verbose: # logger.info(f'=> init {k} from {pretrained}') if ('pool_layers' in k) or ('focal_layers' in k) and v.size() != model_dict[k].size(): table_pretrained = v table_current = model_dict[k] fsize1 = table_pretrained.shape[2] fsize2 = table_current.shape[2] # NOTE: different from interpolation used in self-attention, we use padding or clipping for focal conv if fsize1 < fsize2: table_pretrained_resized = torch.zeros(table_current.shape) table_pretrained_resized[:, :, (fsize2-fsize1)//2:-(fsize2-fsize1)//2, (fsize2-fsize1)//2:-(fsize2-fsize1)//2] = table_pretrained v = table_pretrained_resized elif fsize1 > fsize2: table_pretrained_resized = table_pretrained[:, :, (fsize1-fsize2)//2:-(fsize1-fsize2)//2, (fsize1-fsize2)//2:-(fsize1-fsize2)//2] v = table_pretrained_resized if ("modulation.f" in k or "pre_conv" in k): table_pretrained = v table_current = model_dict[k] if table_pretrained.shape != table_current.shape: if len(table_pretrained.shape) == 2: dim = table_pretrained.shape[1] assert table_current.shape[1] == dim L1 = table_pretrained.shape[0] L2 = table_current.shape[0] if L1 < L2: table_pretrained_resized = torch.zeros(table_current.shape) # copy for linear project table_pretrained_resized[:2*dim] = table_pretrained[:2*dim] # copy for global token gating table_pretrained_resized[-1] = table_pretrained[-1] # copy for first multiple focal levels table_pretrained_resized[2*dim:2*dim+(L1-2*dim-1)] = table_pretrained[2*dim:-1] # reassign pretrained weights v = table_pretrained_resized elif L1 > L2: raise NotImplementedError elif len(table_pretrained.shape) == 1: dim = table_pretrained.shape[0] L1 = table_pretrained.shape[0] L2 = table_current.shape[0] if L1 < L2: table_pretrained_resized = torch.zeros(table_current.shape) # copy for linear project table_pretrained_resized[:dim] = table_pretrained[:dim] # copy for global token gating table_pretrained_resized[-1] = table_pretrained[-1] # copy for first multiple focal levels # table_pretrained_resized[dim:2*dim+(L1-2*dim-1)] = table_pretrained[2*dim:-1] # reassign pretrained weights v = table_pretrained_resized elif L1 > L2: raise NotImplementedError need_init_state_dict[k] = v self.load_state_dict(need_init_state_dict, strict=False) def forward(self, x): """Forward function.""" tic = time.time() x = self.patch_embed(x) Wh, Ww = x.size(2), x.size(3) x = x.flatten(2).transpose(1, 2) x = self.pos_drop(x) outs = {} for i in range(self.num_layers): layer = self.layers[i] x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww) if i in self.out_indices: norm_layer = getattr(self, f'norm{i}') x_out = norm_layer(x_out) out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous() outs["res{}".format(i + 2)] = out if len(self.out_indices) == 0: outs["res5"] = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous() toc = time.time() return outs def train(self, mode=True): """Convert the model into training mode while keep layers freezed.""" super(FocalNet, self).train(mode) self._freeze_stages() class D2FocalNet(FocalNet, Backbone): def __init__(self, cfg, input_shape): pretrain_img_size = cfg['BACKBONE']['FOCAL']['PRETRAIN_IMG_SIZE'] patch_size = cfg['BACKBONE']['FOCAL']['PATCH_SIZE'] in_chans = 3 embed_dim = cfg['BACKBONE']['FOCAL']['EMBED_DIM'] depths = cfg['BACKBONE']['FOCAL']['DEPTHS'] mlp_ratio = cfg['BACKBONE']['FOCAL']['MLP_RATIO'] drop_rate = cfg['BACKBONE']['FOCAL']['DROP_RATE'] drop_path_rate = cfg['BACKBONE']['FOCAL']['DROP_PATH_RATE'] norm_layer = nn.LayerNorm patch_norm = cfg['BACKBONE']['FOCAL']['PATCH_NORM'] use_checkpoint = cfg['BACKBONE']['FOCAL']['USE_CHECKPOINT'] out_indices = cfg['BACKBONE']['FOCAL']['OUT_INDICES'] scaling_modulator = cfg['BACKBONE']['FOCAL'].get('SCALING_MODULATOR', False) super().__init__( pretrain_img_size, patch_size, in_chans, embed_dim, depths, mlp_ratio, drop_rate, drop_path_rate, norm_layer, patch_norm, out_indices, focal_levels=cfg['BACKBONE']['FOCAL']['FOCAL_LEVELS'], focal_windows=cfg['BACKBONE']['FOCAL']['FOCAL_WINDOWS'], use_conv_embed=cfg['BACKBONE']['FOCAL']['USE_CONV_EMBED'], use_postln=cfg['BACKBONE']['FOCAL']['USE_POSTLN'], use_postln_in_modulation=cfg['BACKBONE']['FOCAL']['USE_POSTLN_IN_MODULATION'], scaling_modulator=scaling_modulator, use_layerscale=cfg['BACKBONE']['FOCAL']['USE_LAYERSCALE'], use_checkpoint=use_checkpoint, ) self._out_features = cfg['BACKBONE']['FOCAL']['OUT_FEATURES'] self._out_feature_strides = { "res2": 4, "res3": 8, "res4": 16, "res5": 32, } self._out_feature_channels = { "res2": self.num_features[0], "res3": self.num_features[1], "res4": self.num_features[2], "res5": self.num_features[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 } @property def size_divisibility(self): return 32 @register_backbone def get_focal_backbone(cfg): focal = D2FocalNet(cfg['MODEL'], 224) if cfg['MODEL']['BACKBONE']['LOAD_PRETRAINED'] is True: filename = cfg['MODEL']['BACKBONE']['PRETRAINED'] logger.info(f'=> init from {filename}') with PathManager.open(filename, "rb") as f: ckpt = torch.load(f)['model'] focal.load_weights(ckpt, cfg['MODEL']['BACKBONE']['FOCAL'].get('PRETRAINED_LAYERS', ['*']), cfg['VERBOSE']) return focal