-
Notifications
You must be signed in to change notification settings - Fork 28
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
6b8cbce
commit 5e773db
Showing
3 changed files
with
224 additions
and
2 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,209 @@ | ||
| import torch | ||
| from torch import nn, Tensor | ||
| from .layers import DropPath | ||
|
|
||
|
|
||
| class MLP(nn.Module): | ||
| def __init__(self, dim, hidden_dim, out_dim=None) -> None: | ||
| super().__init__() | ||
| out_dim = out_dim or dim | ||
| self.fc1 = nn.Linear(dim, hidden_dim) | ||
| self.act = nn.GELU() | ||
| self.fc2 = nn.Linear(hidden_dim, out_dim) | ||
|
|
||
| def forward(self, x: Tensor) -> Tensor: | ||
| return self.fc2(self.act(self.fc1(x))) | ||
|
|
||
|
|
||
| class LearnedAggreationLayer(nn.Module): | ||
| def __init__(self, dim, head=1): | ||
| super().__init__() | ||
| self.head = head | ||
| self.scale = (dim // head) ** -0.5 | ||
|
|
||
| self.q = nn.Linear(dim, dim) | ||
| self.k = nn.Linear(dim, dim) | ||
| self.v = nn.Linear(dim, dim) | ||
| self.proj = nn.Linear(dim, dim) | ||
|
|
||
| def forward(self, x: Tensor) -> Tensor: | ||
| B, N, C = x.shape | ||
| q = self.q(x[:, 0]).unsqueeze(1).reshape(B, 1, self.head, C//self.head).permute(0, 2, 1, 3) | ||
| k = self.k(x).reshape(B, N, self.head, C//self.head).permute(0, 2, 1, 3) | ||
| v = self.v(x).reshape(B, N, self.head, C//self.head).permute(0, 2, 1, 3) | ||
|
|
||
| q = q * self.scale | ||
|
|
||
| attn = (q @ k.transpose(-2, -1)) | ||
| attn = attn.softmax(dim=-1) | ||
|
|
||
| x_cls = (attn @ v).transpose(1, 2).reshape(B, 1, C) | ||
| x_cls = self.proj(x_cls) | ||
|
|
||
| return x_cls | ||
|
|
||
|
|
||
| class SqueezeExcite(nn.Module): | ||
| def __init__(self, in_chs, rd_ratio=0.25) -> None: | ||
| super().__init__() | ||
| rd_channels = round(in_chs * rd_ratio) | ||
| self.conv_reduce = nn.Conv2d(in_chs, rd_channels, 1) | ||
| self.act1 = nn.ReLU(True) | ||
| self.conv_expand = nn.Conv2d(rd_channels, in_chs, 1) | ||
| self.gate = nn.Sigmoid() | ||
|
|
||
| def forward(self, x): | ||
| x_se = x.mean((2, 3), keepdim=True) | ||
| x_se = self.act1(self.conv_reduce(x_se)) | ||
| x_se = self.conv_expand(x_se) | ||
| return x * self.gate(x_se) | ||
|
|
||
|
|
||
| class SEBlock(nn.Module): | ||
| def __init__(self, dim) -> None: | ||
| super().__init__() | ||
| self.qkv_pos = nn.Sequential( | ||
| nn.Conv2d(dim, dim, 1), | ||
| nn.GELU(), | ||
| nn.Conv2d(dim, dim, 3, 1, 1, groups=dim), | ||
| nn.GELU(), | ||
| SqueezeExcite(dim), | ||
| nn.Conv2d(dim, dim, 1) | ||
| ) | ||
|
|
||
| def forward(self, x): | ||
| B, N, C = x.shape | ||
| H = W = int(N ** 0.5) | ||
| x = x.transpose(-1, -2) | ||
| x = x.reshape(B, C, H, W) | ||
| x = self.qkv_pos(x) | ||
| x = x.reshape(B, C, N) | ||
| x = x.transpose(-1, -2) | ||
| return x | ||
|
|
||
|
|
||
| class Block(nn.Module): | ||
| def __init__(self, dim, dpr=0., init_values=1e-6) -> None: | ||
| super().__init__() | ||
| self.norm1 = nn.LayerNorm(dim) | ||
| self.attn = SEBlock(dim) | ||
| self.drop_path = DropPath(dpr) if dpr > 0. else nn.Identity() | ||
| self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)), requires_grad=True) | ||
|
|
||
| def forward(self, x): | ||
| return x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x))) | ||
|
|
||
|
|
||
| class BlockToken(nn.Module): | ||
| def __init__(self, dim, head, dpr=0., init_values=1e-6): | ||
| super().__init__() | ||
| self.norm1 = nn.LayerNorm(dim) | ||
| self.attn = LearnedAggreationLayer(dim, head) | ||
| self.drop_path = DropPath(dpr) if dpr > 0. else nn.Identity() | ||
|
|
||
| self.norm2 = nn.LayerNorm(dim) | ||
| self.mlp = MLP(dim, int(dim*3)) | ||
|
|
||
| self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)), requires_grad=True) | ||
| self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)), requires_grad=True) | ||
|
|
||
| def forward(self, x: Tensor, x_cls: Tensor) -> Tensor: | ||
| u = torch.cat([x_cls, x], dim=1) | ||
| x_cls = x_cls + self.drop_path(self.gamma_1 * self.attn(self.norm1(u))) | ||
| x_cls = x_cls + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x_cls))) | ||
| return x_cls | ||
|
|
||
|
|
||
| class Stem(nn.Module): | ||
| def __init__(self, dim): | ||
| super().__init__() | ||
| self.proj = nn.Sequential( | ||
| nn.Sequential(nn.Conv2d(3, dim//8, 3, 2, 1, bias=False)), | ||
| nn.GELU(), | ||
| nn.Sequential(nn.Conv2d(dim//8, dim//4, 3, 2, 1, bias=False)), | ||
| nn.GELU(), | ||
| nn.Sequential(nn.Conv2d(dim//4, dim//2, 3, 2, 1, bias=False)), | ||
| nn.GELU(), | ||
| nn.Sequential(nn.Conv2d(dim//2, dim, 3, 2, 1, bias=False)) | ||
| ) | ||
|
|
||
| def forward(self, x: Tensor) -> Tensor: | ||
| return self.proj(x).flatten(2).transpose(1, 2) | ||
|
|
||
|
|
||
| patchconvnet_settings = { | ||
| 'S60': [384, 60, 0.], # [embed_dim, depth, drop_path_rate] | ||
| 'S120': [384, 120, 0.], | ||
| 'B60': [768, 60, 0.] | ||
| } | ||
|
|
||
|
|
||
| class PatchConvnet(nn.Module): | ||
| def __init__(self, model_name: str = 'S60', pretrained: str = None, num_classes: int = 1000, *args, **kwargs) -> None: | ||
| super().__init__() | ||
| assert model_name in patchconvnet_settings.keys(), f"PatchConvnet model name should be in {list(patchconvnet_settings.keys())}" | ||
| embed_dim, depth, drop_path_rate = patchconvnet_settings[model_name] | ||
|
|
||
| self.patch_embed = Stem(embed_dim) | ||
| self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) | ||
|
|
||
| dpr = [drop_path_rate for _ in range(depth)] # stochastic depth decay rule | ||
|
|
||
| self.blocks = nn.ModuleList([ | ||
| Block(embed_dim, dpr[i]) | ||
| for i in range(depth)]) | ||
|
|
||
| self.blocks_token_only = nn.ModuleList([ | ||
| BlockToken(embed_dim, 1, 0) | ||
| for i in range(1)]) | ||
|
|
||
| self.norm = nn.LayerNorm(embed_dim) | ||
| self.total_len = 1 + depth | ||
|
|
||
| self.head = nn.Linear(embed_dim, num_classes) | ||
| self._init_weights(pretrained) | ||
|
|
||
| def _init_weights(self, pretrained: str = None) -> None: | ||
| if pretrained: | ||
| self.load_state_dict(torch.load(pretrained, map_location='cpu')) | ||
| else: | ||
| for n, m in self.named_modules(): | ||
| if isinstance(m, nn.Linear): | ||
| if n.startswith('head'): | ||
| nn.init.zeros_(m.weight) | ||
| nn.init.zeros_(m.bias) | ||
| else: | ||
| nn.init.xavier_uniform_(m.weight) | ||
| if m.bias is not None: | ||
| nn.init.zeros_(m.bias) | ||
| elif isinstance(m, nn.LayerNorm): | ||
| nn.init.ones_(m.weight) | ||
| nn.init.zeros_(m.bias) | ||
| elif isinstance(m, nn.Conv2d): | ||
| nn.init.xavier_uniform_(m.weight) | ||
| if m.bias is not None: | ||
| nn.init.zeros_(m.bias) | ||
|
|
||
|
|
||
| def forward(self, x: Tensor) -> Tensor: | ||
| B = x.shape[0] | ||
| x = self.patch_embed(x) | ||
| cls_tokens = self.cls_token.expand(B, -1, -1) | ||
|
|
||
| for blk in self.blocks: | ||
| x = blk(x) | ||
|
|
||
| for blk in self.blocks_token_only: | ||
| cls_tokens = blk(x, cls_tokens) | ||
|
|
||
| x = torch.cat([cls_tokens, x], dim=1) | ||
| x = self.norm(x) | ||
| x = self.head(x[:, 0]) | ||
| return x | ||
|
|
||
|
|
||
| if __name__ == '__main__': | ||
| model = PatchConvnet('B60', 'C:\\Users\\sithu\\Documents\\weights\\backbones\\patchconvnet\\b60_224_1k.pth') | ||
| x = torch.zeros(1, 3, 224, 224) | ||
| y = model(x) | ||
| print(y.shape) |