import math import torch import torch.nn as nn import torch.nn.functional as F from einops import rearrange import torch.utils.checkpoint as checkpoint import numpy as np from timm.layers import DropPath, to_2tuple, trunc_normal_ from PIL import Image, ImageOps from torchvision import transforms import numpy as np import random import cv2 import os import subprocess import time import tempfile def set_random_seed(seed): random.seed(seed) np.random.seed(seed) torch.manual_seed(seed) torch.cuda.manual_seed(seed) torch.cuda.manual_seed_all(seed) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False set_random_seed(9) torch.set_float32_matmul_precision('highest') 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 PatchMerging(nn.Module): """ Patch Merging Layer Args: dim (int): Number of input channels. norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm """ def __init__(self, dim, norm_layer=nn.LayerNorm): super().__init__() self.dim = dim self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) self.norm = norm_layer(4 * dim) 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. """ B, L, C = x.shape assert L == H * W, "input feature has wrong size" x = x.view(B, H, W, C) # padding pad_input = (H % 2 == 1) or (W % 2 == 1) if pad_input: x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2)) 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 BasicLayer(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 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 """ def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): super().__init__() patch_size = to_2tuple(patch_size) self.patch_size = patch_size self.in_chans = in_chans self.embed_dim = embed_dim 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.""" # padding _, _, 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 SwinTransformer(nn.Module): """ Swin Transformer backbone. A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` - https://arxiv.org/pdf/2103.14030 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. num_heads (tuple[int]): Number of attention head of each stage. window_size (int): Window size. Default: 7. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4. qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. drop_rate (float): Dropout rate. attn_drop_rate (float): Attention dropout rate. Default: 0. drop_path_rate (float): Stochastic depth rate. Default: 0.2. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. ape (bool): If True, add absolute position embedding to the patch embedding. Default: False. 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. use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. """ def __init__(self, pretrain_img_size=224, patch_size=4, in_chans=3, embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None, drop_rate=0., attn_drop_rate=0., drop_path_rate=0.2, norm_layer=nn.LayerNorm, ape=False, patch_norm=True, out_indices=(0, 1, 2, 3), frozen_stages=-1, use_checkpoint=False): super().__init__() self.pretrain_img_size = pretrain_img_size self.num_layers = len(depths) self.embed_dim = embed_dim self.ape = ape 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) # absolute position embedding if self.ape: pretrain_img_size = to_2tuple(pretrain_img_size) patch_size = to_2tuple(patch_size) patches_resolution = [pretrain_img_size[0] // patch_size[0], pretrain_img_size[1] // patch_size[1]] self.absolute_pos_embed = nn.Parameter(torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1])) trunc_normal_(self.absolute_pos_embed, std=.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 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], 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, downsample=PatchMerging if (i_layer < self.num_layers - 1) else None, 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 >= 1 and self.ape: self.absolute_pos_embed.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 forward(self, x): x = self.patch_embed(x) Wh, Ww = x.size(2), x.size(3) if self.ape: # interpolate the position embedding to the corresponding size absolute_pos_embed = F.interpolate(self.absolute_pos_embed, size=(Wh, Ww), mode='bicubic') x = (x + absolute_pos_embed) # B Wh*Ww C outs = [x.contiguous()] x = x.flatten(2).transpose(1, 2) x = self.pos_drop(x) 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.append(out) return tuple(outs) def get_activation_fn(activation): """Return an activation function given a string""" if activation == "gelu": return F.gelu raise RuntimeError(F"activation should be gelu, not {activation}.") def make_cbr(in_dim, out_dim): return nn.Sequential(nn.Conv2d(in_dim, out_dim, kernel_size=3, padding=1), nn.InstanceNorm2d(out_dim), nn.GELU()) def make_cbg(in_dim, out_dim): return nn.Sequential(nn.Conv2d(in_dim, out_dim, kernel_size=3, padding=1), nn.InstanceNorm2d(out_dim), nn.GELU()) def rescale_to(x, scale_factor: float = 2, interpolation='nearest'): return F.interpolate(x, scale_factor=scale_factor, mode=interpolation) def resize_as(x, y, interpolation='bilinear'): return F.interpolate(x, size=y.shape[-2:], mode=interpolation) def image2patches(x): """b c (hg h) (wg w) -> (hg wg b) c h w""" x = rearrange(x, 'b c (hg h) (wg w) -> (hg wg b) c h w', hg=2, wg=2 ) return x def patches2image(x): """(hg wg b) c h w -> b c (hg h) (wg w)""" x = rearrange(x, '(hg wg b) c h w -> b c (hg h) (wg w)', hg=2, wg=2) return x class PositionEmbeddingSine: def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None): super().__init__() self.num_pos_feats = num_pos_feats self.temperature = temperature self.normalize = normalize if scale is not None and normalize is False: raise ValueError("normalize should be True if scale is passed") if scale is None: scale = 2 * math.pi self.scale = scale self.dim_t = torch.arange(0, self.num_pos_feats, dtype=torch.float32) def __call__(self, b, h, w): device = self.dim_t.device mask = torch.zeros([b, h, w], dtype=torch.bool, device=device) assert mask is not None not_mask = ~mask y_embed = not_mask.cumsum(dim=1, dtype=torch.float32) x_embed = not_mask.cumsum(dim=2, dtype=torch.float32) if self.normalize: eps = 1e-6 y_embed = (y_embed - 0.5) / (y_embed[:, -1:, :] + eps) * self.scale x_embed = (x_embed - 0.5) / (x_embed[:, :, -1:] + eps) * self.scale dim_t = self.temperature ** (2 * (self.dim_t.to(device) // 2) / self.num_pos_feats) pos_x = x_embed[:, :, :, None] / dim_t pos_y = y_embed[:, :, :, None] / dim_t pos_x = torch.stack((pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4).flatten(3) pos_y = torch.stack((pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4).flatten(3) return torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2) class PositionEmbeddingSine: def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None): super().__init__() self.num_pos_feats = num_pos_feats self.temperature = temperature self.normalize = normalize if scale is not None and normalize is False: raise ValueError("normalize should be True if scale is passed") if scale is None: scale = 2 * math.pi self.scale = scale self.dim_t = torch.arange(0, self.num_pos_feats, dtype=torch.float32) def __call__(self, b, h, w): device = self.dim_t.device mask = torch.zeros([b, h, w], dtype=torch.bool, device=device) assert mask is not None not_mask = ~mask y_embed = not_mask.cumsum(dim=1, dtype=torch.float32) x_embed = not_mask.cumsum(dim=2, dtype=torch.float32) if self.normalize: eps = 1e-6 y_embed = (y_embed - 0.5) / (y_embed[:, -1:, :] + eps) * self.scale x_embed = (x_embed - 0.5) / (x_embed[:, :, -1:] + eps) * self.scale dim_t = self.temperature ** (2 * (self.dim_t.to(device) // 2) / self.num_pos_feats) pos_x = x_embed[:, :, :, None] / dim_t pos_y = y_embed[:, :, :, None] / dim_t pos_x = torch.stack((pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4).flatten(3) pos_y = torch.stack((pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4).flatten(3) return torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2) class MCLM(nn.Module): def __init__(self, d_model, num_heads, pool_ratios=[1, 4, 8]): super(MCLM, self).__init__() self.attention = nn.ModuleList([ nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1) ]) self.linear1 = nn.Linear(d_model, d_model * 2) self.linear2 = nn.Linear(d_model * 2, d_model) self.linear3 = nn.Linear(d_model, d_model * 2) self.linear4 = nn.Linear(d_model * 2, d_model) self.norm1 = nn.LayerNorm(d_model) self.norm2 = nn.LayerNorm(d_model) self.dropout = nn.Dropout(0.1) self.dropout1 = nn.Dropout(0.1) self.dropout2 = nn.Dropout(0.1) self.activation = get_activation_fn('gelu') self.pool_ratios = pool_ratios self.p_poses = [] self.g_pos = None self.positional_encoding = PositionEmbeddingSine(num_pos_feats=d_model // 2, normalize=True) def forward(self, l, g): """ l: 4,c,h,w g: 1,c,h,w """ self.p_poses = [] self.g_pos = None b, c, h, w = l.size() # 4,c,h,w -> 1,c,2h,2w concated_locs = rearrange(l, '(hg wg b) c h w -> b c (hg h) (wg w)', hg=2, wg=2) pools = [] for pool_ratio in self.pool_ratios: # b,c,h,w tgt_hw = (round(h / pool_ratio), round(w / pool_ratio)) pool = F.adaptive_avg_pool2d(concated_locs, tgt_hw) pools.append(rearrange(pool, 'b c h w -> (h w) b c')) if self.g_pos is None: pos_emb = self.positional_encoding(pool.shape[0], pool.shape[2], pool.shape[3]) pos_emb = rearrange(pos_emb, 'b c h w -> (h w) b c') self.p_poses.append(pos_emb) pools = torch.cat(pools, 0) if self.g_pos is None: self.p_poses = torch.cat(self.p_poses, dim=0) pos_emb = self.positional_encoding(g.shape[0], g.shape[2], g.shape[3]) self.g_pos = rearrange(pos_emb, 'b c h w -> (h w) b c') device = pools.device self.p_poses = self.p_poses.to(device) self.g_pos = self.g_pos.to(device) # attention between glb (q) & multisensory concated-locs (k,v) g_hw_b_c = rearrange(g, 'b c h w -> (h w) b c') g_hw_b_c = g_hw_b_c + self.dropout1(self.attention[0](g_hw_b_c + self.g_pos, pools + self.p_poses, pools)[0]) g_hw_b_c = self.norm1(g_hw_b_c) g_hw_b_c = g_hw_b_c + self.dropout2(self.linear2(self.dropout(self.activation(self.linear1(g_hw_b_c)).clone()))) g_hw_b_c = self.norm2(g_hw_b_c) # attention between origin locs (q) & freashed glb (k,v) l_hw_b_c = rearrange(l, "b c h w -> (h w) b c") _g_hw_b_c = rearrange(g_hw_b_c, '(h w) b c -> h w b c', h=h, w=w) _g_hw_b_c = rearrange(_g_hw_b_c, "(ng h) (nw w) b c -> (h w) (ng nw b) c", ng=2, nw=2) outputs_re = [] for i, (_l, _g) in enumerate(zip(l_hw_b_c.chunk(4, dim=1), _g_hw_b_c.chunk(4, dim=1))): outputs_re.append(self.attention[i + 1](_l, _g, _g)[0]) # (h w) 1 c outputs_re = torch.cat(outputs_re, 1) # (h w) 4 c l_hw_b_c = l_hw_b_c + self.dropout1(outputs_re) l_hw_b_c = self.norm1(l_hw_b_c) l_hw_b_c = l_hw_b_c + self.dropout2(self.linear4(self.dropout(self.activation(self.linear3(l_hw_b_c)).clone()))) l_hw_b_c = self.norm2(l_hw_b_c) l = torch.cat((l_hw_b_c, g_hw_b_c), 1) # hw,b(5),c return rearrange(l, "(h w) b c -> b c h w", h=h, w=w) ## (5,c,h*w) class MCRM(nn.Module): def __init__(self, d_model, num_heads, pool_ratios=[4, 8, 16], h=None): super(MCRM, self).__init__() self.attention = nn.ModuleList([ nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1), nn.MultiheadAttention(d_model, num_heads, dropout=0.1) ]) self.linear3 = nn.Linear(d_model, d_model * 2) self.linear4 = nn.Linear(d_model * 2, d_model) self.norm1 = nn.LayerNorm(d_model) self.norm2 = nn.LayerNorm(d_model) self.dropout = nn.Dropout(0.1) self.dropout1 = nn.Dropout(0.1) self.dropout2 = nn.Dropout(0.1) self.sigmoid = nn.Sigmoid() self.activation = get_activation_fn('gelu') self.sal_conv = nn.Conv2d(d_model, 1, 1) self.pool_ratios = pool_ratios def forward(self, x): device = x.device b, c, h, w = x.size() loc, glb = x.split([4, 1], dim=0) # 4,c,h,w; 1,c,h,w patched_glb = rearrange(glb, 'b c (hg h) (wg w) -> (hg wg b) c h w', hg=2, wg=2) token_attention_map = self.sigmoid(self.sal_conv(glb)) token_attention_map = F.interpolate(token_attention_map, size=patches2image(loc).shape[-2:], mode='nearest') loc = loc * rearrange(token_attention_map, 'b c (hg h) (wg w) -> (hg wg b) c h w', hg=2, wg=2) pools = [] for pool_ratio in self.pool_ratios: tgt_hw = (round(h / pool_ratio), round(w / pool_ratio)) pool = F.adaptive_avg_pool2d(patched_glb, tgt_hw) pools.append(rearrange(pool, 'nl c h w -> nl c (h w)')) # nl(4),c,hw pools = rearrange(torch.cat(pools, 2), "nl c nphw -> nl nphw 1 c") loc_ = rearrange(loc, 'nl c h w -> nl (h w) 1 c') outputs = [] for i, q in enumerate(loc_.unbind(dim=0)): # traverse all local patches v = pools[i] k = v outputs.append(self.attention[i](q, k, v)[0]) outputs = torch.cat(outputs, 1) src = loc.view(4, c, -1).permute(2, 0, 1) + self.dropout1(outputs) src = self.norm1(src) src = src + self.dropout2(self.linear4(self.dropout(self.activation(self.linear3(src)).clone()))) src = self.norm2(src) src = src.permute(1, 2, 0).reshape(4, c, h, w) # freshed loc glb = glb + F.interpolate(patches2image(src), size=glb.shape[-2:], mode='nearest') # freshed glb return torch.cat((src, glb), 0), token_attention_map class BEN_Base(nn.Module): def __init__(self): super().__init__() self.backbone = SwinTransformer(embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=12) emb_dim = 128 self.sideout5 = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) self.sideout4 = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) self.sideout3 = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) self.sideout2 = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) self.sideout1 = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) self.output5 = make_cbr(1024, emb_dim) self.output4 = make_cbr(512, emb_dim) self.output3 = make_cbr(256, emb_dim) self.output2 = make_cbr(128, emb_dim) self.output1 = make_cbr(128, emb_dim) self.multifieldcrossatt = MCLM(emb_dim, 1, [1, 4, 8]) self.conv1 = make_cbr(emb_dim, emb_dim) self.conv2 = make_cbr(emb_dim, emb_dim) self.conv3 = make_cbr(emb_dim, emb_dim) self.conv4 = make_cbr(emb_dim, emb_dim) self.dec_blk1 = MCRM(emb_dim, 1, [2, 4, 8]) self.dec_blk2 = MCRM(emb_dim, 1, [2, 4, 8]) self.dec_blk3 = MCRM(emb_dim, 1, [2, 4, 8]) self.dec_blk4 = MCRM(emb_dim, 1, [2, 4, 8]) self.insmask_head = nn.Sequential( nn.Conv2d(emb_dim, 384, kernel_size=3, padding=1), nn.InstanceNorm2d(384), nn.GELU(), nn.Conv2d(384, 384, kernel_size=3, padding=1), nn.InstanceNorm2d(384), nn.GELU(), nn.Conv2d(384, emb_dim, kernel_size=3, padding=1) ) self.shallow = nn.Sequential(nn.Conv2d(3, emb_dim, kernel_size=3, padding=1)) self.upsample1 = make_cbg(emb_dim, emb_dim) self.upsample2 = make_cbg(emb_dim, emb_dim) self.output = nn.Sequential(nn.Conv2d(emb_dim, 1, kernel_size=3, padding=1)) for m in self.modules(): if isinstance(m, nn.GELU) or isinstance(m, nn.Dropout): m.inplace = True @torch.inference_mode() @torch.autocast(device_type="cuda",dtype=torch.float16) def forward(self, x): real_batch = x.size(0) shallow_batch = self.shallow(x) glb_batch = rescale_to(x, scale_factor=0.5, interpolation='bilinear') final_input = None for i in range(real_batch): start = i * 4 end = (i + 1) * 4 loc_batch = image2patches(x[i,:,:,:].unsqueeze(dim=0)) input_ = torch.cat((loc_batch, glb_batch[i,:,:,:].unsqueeze(dim=0)), dim=0) if final_input == None: final_input= input_ else: final_input = torch.cat((final_input, input_), dim=0) features = self.backbone(final_input) outputs = [] for i in range(real_batch): start = i * 5 end = (i + 1) * 5 f4 = features[4][start:end, :, :, :] # shape: [5, C, H, W] f3 = features[3][start:end, :, :, :] f2 = features[2][start:end, :, :, :] f1 = features[1][start:end, :, :, :] f0 = features[0][start:end, :, :, :] e5 = self.output5(f4) e4 = self.output4(f3) e3 = self.output3(f2) e2 = self.output2(f1) e1 = self.output1(f0) loc_e5, glb_e5 = e5.split([4, 1], dim=0) e5 = self.multifieldcrossatt(loc_e5, glb_e5) # (4,128,16,16) e4, tokenattmap4 = self.dec_blk4(e4 + resize_as(e5, e4)) e4 = self.conv4(e4) e3, tokenattmap3 = self.dec_blk3(e3 + resize_as(e4, e3)) e3 = self.conv3(e3) e2, tokenattmap2 = self.dec_blk2(e2 + resize_as(e3, e2)) e2 = self.conv2(e2) e1, tokenattmap1 = self.dec_blk1(e1 + resize_as(e2, e1)) e1 = self.conv1(e1) loc_e1, glb_e1 = e1.split([4, 1], dim=0) output1_cat = patches2image(loc_e1) # (1,128,256,256) # add glb feat in output1_cat = output1_cat + resize_as(glb_e1, output1_cat) # merge final_output = self.insmask_head(output1_cat) # (1,128,256,256) # shallow feature merge shallow = shallow_batch[i,:,:,:].unsqueeze(dim=0) final_output = final_output + resize_as(shallow, final_output) final_output = self.upsample1(rescale_to(final_output)) final_output = rescale_to(final_output + resize_as(shallow, final_output)) final_output = self.upsample2(final_output) final_output = self.output(final_output) mask = final_output.sigmoid() outputs.append(mask) return torch.cat(outputs, dim=0) def loadcheckpoints(self,model_path): model_dict = torch.load(model_path, map_location="cpu", weights_only=True) self.load_state_dict(model_dict['model_state_dict'], strict=True) del model_path def inference(self,image,refine_foreground=False): set_random_seed(9) # image = ImageOps.exif_transpose(image) if isinstance(image, Image.Image): image, h, w,original_image = rgb_loader_refiner(image) if torch.cuda.is_available(): img_tensor = img_transform(image).unsqueeze(0).to(next(self.parameters()).device) else: img_tensor = img_transform32(image).unsqueeze(0).to(next(self.parameters()).device) with torch.no_grad(): res = self.forward(img_tensor) # Show Results if refine_foreground == True: pred_pil = transforms.ToPILImage()(res.squeeze()) image_masked = refine_foreground_process(original_image, pred_pil) image_masked.putalpha(pred_pil.resize(original_image.size)) return image_masked else: alpha = postprocess_image(res, im_size=[w,h]) pred_pil = transforms.ToPILImage()(alpha) mask = pred_pil.resize(original_image.size) original_image.putalpha(mask) # mask = Image.fromarray(alpha) return original_image else: foregrounds = [] for batch in image: image, h, w,original_image = rgb_loader_refiner(batch) if torch.cuda.is_available(): img_tensor = img_transform(image).unsqueeze(0).to(next(self.parameters()).device) else: img_tensor = img_transform32(image).unsqueeze(0).to(next(self.parameters()).device) with torch.no_grad(): res = self.forward(img_tensor) if refine_foreground == True: pred_pil = transforms.ToPILImage()(res.squeeze()) image_masked = refine_foreground_process(original_image, pred_pil) image_masked.putalpha(pred_pil.resize(original_image.size)) foregrounds.append(image_masked) else: alpha = postprocess_image(res, im_size=[w,h]) pred_pil = transforms.ToPILImage()(alpha) mask = pred_pil.resize(original_image.size) original_image.putalpha(mask) # mask = Image.fromarray(alpha) foregrounds.append(original_image) return foregrounds def segment_video(self, video_path, output_path="./", fps=0, refine_foreground=False, batch=1, print_frames_processed=True, webm = False, rgb_value= (0, 255, 0)): """ Segments the given video to extract the foreground (with alpha) from each frame and saves the result as either a WebM video (with alpha channel) or MP4 (with a color background). Args: video_path (str): Path to the input video file. output_path (str, optional): Directory (or full path) where the output video and/or files will be saved. Defaults to "./". fps (int, optional): The frames per second (FPS) to use for the output video. If 0 (default), the original FPS of the input video is used. Otherwise, overrides it. refine_foreground (bool, optional): Whether to run an additional “refine foreground” process on each frame. Defaults to False. batch (int, optional): Number of frames to process at once (inference batch size). Large batch sizes may require more GPU memory. Defaults to 1. print_frames_processed (bool, optional): If True (default), prints progress (how many frames have been processed) to the console. webm (bool, optional): If True (default), exports a WebM video with alpha channel (VP9 / yuva420p). If False, exports an MP4 video composited over a solid color background. rgb_value (tuple, optional): The RGB background color (e.g., green screen) used to composite frames when saving to MP4. Defaults to (0, 255, 0). Returns: None. Writes the output video(s) to disk in the specified format. """ cap = cv2.VideoCapture(video_path) if not cap.isOpened(): raise IOError(f"Cannot open video: {video_path}") original_fps = cap.get(cv2.CAP_PROP_FPS) original_fps = 30 if original_fps == 0 else original_fps fps = original_fps if fps == 0 else fps ret, first_frame = cap.read() if not ret: raise ValueError("No frames found in the video.") height, width = first_frame.shape[:2] cap.set(cv2.CAP_PROP_POS_FRAMES, 0) foregrounds = [] frame_idx = 0 processed_count = 0 batch_frames = [] total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) while True: ret, frame = cap.read() if not ret: if batch_frames: batch_results = self.inference(batch_frames, refine_foreground) if isinstance(batch_results, Image.Image): foregrounds.append(batch_results) else: foregrounds.extend(batch_results) if print_frames_processed: print(f"Processed frames {frame_idx-len(batch_frames)+1} to {frame_idx} of {total_frames}") break # Process every frame instead of using intervals frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) pil_frame = Image.fromarray(frame_rgb) batch_frames.append(pil_frame) if len(batch_frames) == batch: batch_results = self.inference(batch_frames, refine_foreground) if isinstance(batch_results, Image.Image): foregrounds.append(batch_results) else: foregrounds.extend(batch_results) if print_frames_processed: print(f"Processed frames {frame_idx-batch+1} to {frame_idx} of {total_frames}") batch_frames = [] processed_count += batch frame_idx += 1 if webm: alpha_webm_path = os.path.join(output_path, "foreground.webm") pil_images_to_webm_alpha(foregrounds, alpha_webm_path, fps=original_fps) else: cap.release() fg_output = os.path.join(output_path, 'foreground.mp4') pil_images_to_mp4(foregrounds, fg_output, fps=original_fps,rgb_value=rgb_value) cv2.destroyAllWindows() try: fg_audio_output = os.path.join(output_path, 'foreground_output_with_audio.mp4') add_audio_to_video(fg_output, video_path, fg_audio_output) except Exception as e: print("No audio found in the original video") print(e) def rgb_loader_refiner( original_image): h, w = original_image.size image = original_image # Convert to RGB if necessary if image.mode != 'RGB': image = image.convert('RGB') # Resize the image image = image.resize((1024, 1024), resample=Image.LANCZOS) return image.convert('RGB'), h, w,original_image # Define the image transformation img_transform = transforms.Compose([ transforms.ToTensor(), transforms.ConvertImageDtype(torch.float16), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]) img_transform32 = transforms.Compose([ transforms.ToTensor(), transforms.ConvertImageDtype(torch.float32), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ]) def pil_images_to_mp4(images, output_path, fps=24, rgb_value=(0, 255, 0)): """ Converts an array of PIL images to an MP4 video. Args: images: List of PIL images output_path: Path to save the MP4 file fps: Frames per second (default: 24) rgb_value: Background RGB color tuple (default: green (0, 255, 0)) """ if not images: raise ValueError("No images provided to convert to MP4.") width, height = images[0].size fourcc = cv2.VideoWriter_fourcc(*'mp4v') video_writer = cv2.VideoWriter(output_path, fourcc, fps, (width, height)) for image in images: # If image has alpha channel, composite onto the specified background color if image.mode == 'RGBA': # Create background image with specified RGB color background = Image.new('RGB', image.size, rgb_value) background = background.convert('RGBA') # Composite the image onto the background image = Image.alpha_composite(background, image) image = image.convert('RGB') else: # Ensure RGB format for non-alpha images image = image.convert('RGB') # Convert to OpenCV format and write open_cv_image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR) video_writer.write(open_cv_image) video_writer.release() def pil_images_to_webm_alpha(images, output_path, fps=30): """ Converts a list of PIL RGBA images to a VP9 .webm video with alpha channel. NOTE: Not all players will display alpha in WebM. Browsers like Chrome/Firefox typically do support VP9 alpha. """ if not images: raise ValueError("No images provided for WebM with alpha.") # Ensure output directory exists os.makedirs(os.path.dirname(output_path), exist_ok=True) with tempfile.TemporaryDirectory() as tmpdir: # Save frames as PNG (with alpha) for idx, img in enumerate(images): if img.mode != "RGBA": img = img.convert("RGBA") out_path = os.path.join(tmpdir, f"{idx:06d}.png") img.save(out_path, "PNG") # Construct ffmpeg command # -c:v libvpx-vp9 => VP9 encoder # -pix_fmt yuva420p => alpha-enabled pixel format # -auto-alt-ref 0 => helps preserve alpha frames (libvpx quirk) ffmpeg_cmd = [ "ffmpeg", "-y", "-framerate", str(fps), "-i", os.path.join(tmpdir, "%06d.png"), "-c:v", "libvpx-vp9", "-pix_fmt", "yuva420p", "-auto-alt-ref", "0", output_path ] subprocess.run(ffmpeg_cmd, check=True) print(f"WebM with alpha saved to {output_path}") def add_audio_to_video(video_without_audio_path, original_video_path, output_path): """ Check if the original video has an audio stream. If yes, add it. If not, skip. """ # 1) Probe original video for audio streams probe_command = [ 'ffprobe', '-v', 'error', '-select_streams', 'a:0', '-show_entries', 'stream=index', '-of', 'csv=p=0', original_video_path ] result = subprocess.run(probe_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) # result.stdout is empty if no audio stream found if not result.stdout.strip(): print("No audio track found in original video, skipping audio addition.") return print("Audio track detected; proceeding to mux audio.") # 2) If audio found, run ffmpeg to add it command = [ 'ffmpeg', '-y', '-i', video_without_audio_path, '-i', original_video_path, '-c', 'copy', '-map', '0:v:0', '-map', '1:a:0', # we know there's an audio track now output_path ] subprocess.run(command, check=True) print(f"Audio added successfully => {output_path}") ### Thanks to the source: https://huggingface.co/ZhengPeng7/BiRefNet/blob/main/handler.py def refine_foreground_process(image, mask, r=90): if mask.size != image.size: mask = mask.resize(image.size) image = np.array(image) / 255.0 mask = np.array(mask) / 255.0 estimated_foreground = FB_blur_fusion_foreground_estimator_2(image, mask, r=r) image_masked = Image.fromarray((estimated_foreground * 255.0).astype(np.uint8)) return image_masked def FB_blur_fusion_foreground_estimator_2(image, alpha, r=90): # Thanks to the source: https://github.com/Photoroom/fast-foreground-estimation alpha = alpha[:, :, None] F, blur_B = FB_blur_fusion_foreground_estimator(image, image, image, alpha, r) return FB_blur_fusion_foreground_estimator(image, F, blur_B, alpha, r=6)[0] def FB_blur_fusion_foreground_estimator(image, F, B, alpha, r=90): if isinstance(image, Image.Image): image = np.array(image) / 255.0 blurred_alpha = cv2.blur(alpha, (r, r))[:, :, None] blurred_FA = cv2.blur(F * alpha, (r, r)) blurred_F = blurred_FA / (blurred_alpha + 1e-5) blurred_B1A = cv2.blur(B * (1 - alpha), (r, r)) blurred_B = blurred_B1A / ((1 - blurred_alpha) + 1e-5) F = blurred_F + alpha * \ (image - alpha * blurred_F - (1 - alpha) * blurred_B) F = np.clip(F, 0, 1) return F, blurred_B def postprocess_image(result: torch.Tensor, im_size: list) -> np.ndarray: result = torch.squeeze(F.interpolate(result, size=im_size, mode='bilinear'), 0) ma = torch.max(result) mi = torch.min(result) result = (result - mi) / (ma - mi) im_array = (result * 255).permute(1, 2, 0).cpu().data.numpy().astype(np.uint8) im_array = np.squeeze(im_array) return im_array def rgb_loader_refiner( original_image): h, w = original_image.size # # Apply EXIF orientation image = ImageOps.exif_transpose(original_image) if original_image.mode != 'RGB': original_image = original_image.convert('RGB') image = original_image # Convert to RGB if necessary # Resize the image image = image.resize((1024, 1024), resample=Image.LANCZOS) return image, h, w,original_image