| import torch | |
| from torch import nn | |
| import torch.nn.functional as F | |
| from einops import rearrange | |
| class Residual(nn.Module): | |
| def __init__(self, fn): | |
| super().__init__() | |
| self.fn = fn | |
| def forward(self, x, **kwargs): | |
| return self.fn(x, **kwargs) + x | |
| class PreNorm(nn.Module): | |
| def __init__(self, dim, fn): | |
| super().__init__() | |
| self.norm = nn.LayerNorm(dim) | |
| self.fn = fn | |
| def forward(self, x, **kwargs): | |
| return self.fn(self.norm(x), **kwargs) | |
| class FeedForward(nn.Module): | |
| def __init__(self, dim, hidden_dim): | |
| super().__init__() | |
| self.net = nn.Sequential( | |
| nn.Linear(dim, hidden_dim), | |
| nn.GELU(), | |
| nn.Linear(hidden_dim, dim) | |
| ) | |
| def forward(self, x): | |
| return self.net(x) | |
| class Attention(nn.Module): | |
| def __init__(self, dim, heads=8): | |
| super().__init__() | |
| self.heads = heads | |
| self.scale = dim ** -0.5 | |
| self.to_qkv = nn.Linear(dim, dim * 3, bias=False) | |
| self.to_out = nn.Linear(dim, dim) | |
| def forward(self, x, mask = None): | |
| b, n, _, h = *x.shape, self.heads | |
| qkv = self.to_qkv(x) | |
| q, k, v = rearrange(qkv, 'b n (qkv h d) -> qkv b h n d', qkv=3, h=h) | |
| dots = torch.einsum('bhid,bhjd->bhij', q, k) * self.scale | |
| if mask is not None: | |
| mask = F.pad(mask.flatten(1), (1, 0), value = True) | |
| assert mask.shape[-1] == dots.shape[-1], 'mask has incorrect dimensions' | |
| mask = mask[:, None, :] * mask[:, :, None] | |
| dots.masked_fill_(~mask, float('-inf')) | |
| del mask | |
| attn = dots.softmax(dim=-1) | |
| out = torch.einsum('bhij,bhjd->bhid', attn, v) | |
| out = rearrange(out, 'b h n d -> b n (h d)') | |
| out = self.to_out(out) | |
| return out | |
| class Transformer(nn.Module): | |
| def __init__(self, dim, depth, heads, mlp_dim): | |
| super().__init__() | |
| self.layers = nn.ModuleList([]) | |
| for _ in range(depth): | |
| self.layers.append(nn.ModuleList([ | |
| Residual(PreNorm(dim, Attention(dim, heads = heads))), | |
| Residual(PreNorm(dim, FeedForward(dim, mlp_dim))) | |
| ])) | |
| def forward(self, x, mask=None): | |
| for attn, ff in self.layers: | |
| x = attn(x, mask=mask) | |
| x = ff(x) | |
| return x | |
| class CViT(nn.Module): | |
| def __init__(self, image_size=224, patch_size=7, num_classes=2, channels=512, | |
| dim=1024, depth=6, heads=8, mlp_dim=2048): | |
| super().__init__() | |
| assert image_size % patch_size == 0, 'image dimensions must be divisible by the patch size' | |
| self.features = nn.Sequential( | |
| nn.Conv2d(3, 32, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=32), | |
| nn.ReLU(), | |
| nn.Conv2d(32, 32, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=32), | |
| nn.ReLU(), | |
| nn.Conv2d(32, 32, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=32), | |
| nn.ReLU(), | |
| nn.MaxPool2d(kernel_size=2, stride=2), | |
| nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=64), | |
| nn.ReLU(), | |
| nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=64), | |
| nn.ReLU(), | |
| nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=64), | |
| nn.ReLU(), | |
| nn.MaxPool2d(kernel_size=2, stride=2), | |
| nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=128), | |
| nn.ReLU(), | |
| nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=128), | |
| nn.ReLU(), | |
| nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=128), | |
| nn.ReLU(), | |
| nn.MaxPool2d(kernel_size=2, stride=2), | |
| nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=256), | |
| nn.ReLU(), | |
| nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=256), | |
| nn.ReLU(), | |
| nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=256), | |
| nn.ReLU(), | |
| nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=256), | |
| nn.ReLU(), | |
| nn.MaxPool2d(kernel_size=2, stride=2), | |
| nn.Conv2d(256, 512, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=512), | |
| nn.ReLU(), | |
| nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=512), | |
| nn.ReLU(), | |
| nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=512), | |
| nn.ReLU(), | |
| nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1), | |
| nn.BatchNorm2d(num_features=512), | |
| nn.ReLU(), | |
| nn.MaxPool2d(kernel_size=2, stride=2) | |
| ) | |
| num_patches = (image_size // patch_size) ** 2 | |
| self.max_sequence_length = num_patches+1 | |
| patch_dim = channels * patch_size ** 2 | |
| self.patch_size = patch_size | |
| self.pos_embedding = nn.Parameter(torch.randn(1, self.max_sequence_length, dim)) | |
| self.patch_to_embedding = nn.Linear(patch_dim, dim) | |
| self.cls_token = nn.Parameter(torch.randn(1, 1, dim)) | |
| self.transformer = Transformer(dim, depth, heads, mlp_dim) | |
| self.to_cls_token = nn.Identity() | |
| self.mlp_head = nn.Sequential( | |
| nn.Linear(dim, mlp_dim), | |
| nn.ReLU(), | |
| nn.Linear(mlp_dim, num_classes) | |
| ) | |
| def forward(self, img, mask=None): | |
| p = self.patch_size | |
| x = self.features(img) | |
| y = rearrange(x, 'b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = p, p2 = p) | |
| y = self.patch_to_embedding(y) | |
| cls_tokens = self.cls_token.expand(y.shape[0], -1, -1) | |
| x = torch.cat((cls_tokens, y), dim=1) | |
| x += self.pos_embedding[:, :x.size(1)] | |
| x = self.transformer(x, mask) | |
| x = self.to_cls_token(x[:, 0]) | |
| return self.mlp_head(x) |