Spaces:
Running
on
Zero
Running
on
Zero
| import torch | |
| import torch.nn as nn | |
| import math | |
| def reshape_tensor(x, heads): | |
| bs, length, width = x.shape | |
| #(bs, length, width) --> (bs, length, n_heads, dim_per_head) | |
| x = x.view(bs, length, heads, -1) | |
| # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head) | |
| x = x.transpose(1, 2) | |
| # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head) | |
| x = x.reshape(bs, heads, length, -1) | |
| return x | |
| def FeedForward(dim, mult=4): | |
| inner_dim = int(dim * mult) | |
| return nn.Sequential( | |
| nn.LayerNorm(dim), | |
| nn.Linear(dim, inner_dim, bias=False), | |
| nn.GELU(), | |
| nn.Linear(inner_dim, dim, bias=False), | |
| ) | |
| class PerceiverAttention(nn.Module): | |
| def __init__(self, *, dim, dim_head=64, heads=8): | |
| super().__init__() | |
| self.scale = dim_head**-0.5 | |
| self.dim_head = dim_head | |
| self.heads = heads | |
| inner_dim = dim_head * heads | |
| self.norm1 = nn.LayerNorm(dim) | |
| self.norm2 = nn.LayerNorm(dim) | |
| self.to_q = nn.Linear(dim, inner_dim, bias=False) | |
| self.to_kv = nn.Linear(dim, inner_dim * 2, bias=False) | |
| self.to_out = nn.Linear(inner_dim, dim, bias=False) | |
| def forward(self, x, latents): | |
| """ | |
| Args: | |
| x (torch.Tensor): image features | |
| shape (b, n1, D) | |
| latent (torch.Tensor): latent features | |
| shape (b, n2, D) | |
| """ | |
| x = self.norm1(x) | |
| latents = self.norm2(latents) | |
| b, l, _ = latents.shape | |
| q = self.to_q(latents) | |
| kv_input = torch.cat((x, latents), dim=-2) | |
| k, v = self.to_kv(kv_input).chunk(2, dim=-1) | |
| q = reshape_tensor(q, self.heads) | |
| k = reshape_tensor(k, self.heads) | |
| v = reshape_tensor(v, self.heads) | |
| # attention | |
| scale = 1 / math.sqrt(math.sqrt(self.dim_head)) | |
| weight = (q * scale) @ (k * scale).transpose(-2, -1) # More stable with f16 than dividing afterwards | |
| weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype) | |
| out = weight @ v | |
| out = out.permute(0, 2, 1, 3).reshape(b, l, -1) | |
| return self.to_out(out) | |
| class FacePerceiverResampler(torch.nn.Module): | |
| def __init__( | |
| self, | |
| *, | |
| dim=768, | |
| depth=4, | |
| dim_head=64, | |
| heads=16, | |
| embedding_dim=1280, | |
| output_dim=768, | |
| ff_mult=4, | |
| ): | |
| super().__init__() | |
| self.proj_in = torch.nn.Linear(embedding_dim, dim) | |
| self.proj_out = torch.nn.Linear(dim, output_dim) | |
| self.norm_out = torch.nn.LayerNorm(output_dim) | |
| self.layers = torch.nn.ModuleList([]) | |
| for _ in range(depth): | |
| self.layers.append( | |
| torch.nn.ModuleList( | |
| [ | |
| PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads), | |
| FeedForward(dim=dim, mult=ff_mult), | |
| ] | |
| ) | |
| ) | |
| def forward(self, latents, x): | |
| x = self.proj_in(x) | |
| for attn, ff in self.layers: | |
| latents = attn(x, latents) + latents | |
| latents = ff(latents) + latents | |
| latents = self.proj_out(latents) | |
| return self.norm_out(latents) | |
| class ProjPlusModel(torch.nn.Module): | |
| def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, clip_embeddings_dim=1280, num_tokens=4): | |
| super().__init__() | |
| self.cross_attention_dim = cross_attention_dim | |
| self.num_tokens = num_tokens | |
| self.proj = torch.nn.Sequential( | |
| torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2), | |
| torch.nn.GELU(), | |
| torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens), | |
| ) | |
| self.norm = torch.nn.LayerNorm(cross_attention_dim) | |
| self.perceiver_resampler = FacePerceiverResampler( | |
| dim=cross_attention_dim, | |
| depth=4, | |
| dim_head=64, | |
| heads=cross_attention_dim // 64, | |
| embedding_dim=clip_embeddings_dim, | |
| output_dim=cross_attention_dim, | |
| ff_mult=4, | |
| ) | |
| def forward(self, id_embeds, clip_embeds, shortcut = True, scale = 1.0): | |
| x = self.proj(id_embeds) | |
| x = x.reshape(-1, self.num_tokens, self.cross_attention_dim) | |
| x = self.norm(x) | |
| out = self.perceiver_resampler(x, clip_embeds) | |
| if shortcut: | |
| out = x + scale * out | |
| return out |