SAM——分割万物模型

介绍

SAM即为Segment Anything Model（分割万物模型），这是一个训练极其简单同时可以零样本迁移学习的小模型，同时这个模型的参数也十分的少，其功能也非常有利于促进CV的发展。

对于Transformer模型，相信一定很熟悉了吧，这里给出之前写过的博客，当然我会重说一次的。

Minloha的博客—Seq2Seq与Transformer

ViT

ViT即Vision Transformer，首先把图片分割为若干个小份，具体大小取决于Encoder的维度。

既然我们有了输入，自然就需要进行归一化，然后经过多头注意力机制后进入前馈神经网络中，得到编码向量，对于标准的Transformer模块，每一个输入都要求是一个向量，我们把图片的小部分看作一个折叠的向量即可。同时每次输入一个小块，最后就得到了输出的image embedding向量。

mask层通过卷积向量得到对应的编码，结果应为不同区域检测到的目标，最后就可以进行非极大值抑制，得到最后的选区

在得到图片的编码向量后，我们就需要训练掩码获取到最大的物体分割，具体的细节可以阅读下面的代码注释。

SAM

接下来就要使用SAM模型了，首先我们看看Transformer实现部分：

import torch
from torch import Tensor, nn

import math
from typing import Tuple, Type

from .common import MLPBlock



class TwoWayTransformer(nn.Module):
    def __init__(
        self,
        depth: int,
        embedding_dim: int,
        num_heads: int,
        mlp_dim: int,
        activation: Type[nn.Module] = nn.ReLU,
        attention_downsample_rate: int = 2,
    ) -> None:
        """
        编码转换器：
        	depth(int):转换器中的层数
        	embedding_dim(int): 输入嵌入的通道维度
            num_heads(int): 多头注意力的头数。必须分割嵌入_dim
			mlp_dim(int): mlp块内部的通道维度
			activation: 在MLP块中使用的激活函数
        """
        super().__init__()
        self.depth = depth
        self.embedding_dim = embedding_dim
        self.num_heads = num_heads
        self.mlp_dim = mlp_dim
        self.layers = nn.ModuleList()

        for i in range(depth):
            self.layers.append(
                TwoWayAttentionBlock(
                    embedding_dim=embedding_dim,
                    num_heads=num_heads,
                    mlp_dim=mlp_dim,
                    activation=activation,
                    attention_downsample_rate=attention_downsample_rate,
                    skip_first_layer_pe=(i == 0),
                )
            )

        self.final_attn_token_to_image = Attention(
            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
        )
        self.norm_final_attn = nn.LayerNorm(embedding_dim)

    def forward(
        self,
        image_embedding: Tensor,
        image_pe: Tensor,
        point_embedding: Tensor,
    ) -> Tuple[Tensor, Tensor]:
        """
        图片矩阵为： BCHW形式，即B个图片，C个通道，每个通道H×W大小的矩阵
        """
        # BxCxHxW -> BxHWxC == B x N_image_tokens x C
        bs, c, h, w = image_embedding.shape
        image_embedding = image_embedding.flatten(2).permute(0, 2, 1)
        image_pe = image_pe.flatten(2).permute(0, 2, 1)

        # Prepare queries
        queries = point_embedding
        keys = image_embedding

        # 应用归一层
        for layer in self.layers:
            queries, keys = layer(
                queries=queries,
                keys=keys,
                query_pe=point_embedding,
                key_pe=image_pe,
            )

        # 使用自注意力层进行计算
        q = queries + point_embedding
        k = keys + image_pe
        attn_out = self.final_attn_token_to_image(q=q, k=k, v=keys)
        queries = queries + attn_out
        queries = self.norm_final_attn(queries)

        return queries, keys


class TwoWayAttentionBlock(nn.Module):
    def __init__(
        self,
        embedding_dim: int,
        num_heads: int,
        mlp_dim: int = 2048,
        activation: Type[nn.Module] = nn.ReLU,
        attention_downsample_rate: int = 2,
        skip_first_layer_pe: bool = False,
    ) -> None:
        """
        Transformer共有四层，分别为：自注意力层(Norm)，交叉验证(Norm)，FNN层(Norm)
        """
        super().__init__()
        self.self_attn = Attention(embedding_dim, num_heads)
        self.norm1 = nn.LayerNorm(embedding_dim)

        self.cross_attn_token_to_image = Attention(
            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
        )
        self.norm2 = nn.LayerNorm(embedding_dim)

        self.mlp = MLPBlock(embedding_dim, mlp_dim, activation)
        self.norm3 = nn.LayerNorm(embedding_dim)

        self.norm4 = nn.LayerNorm(embedding_dim)
        self.cross_attn_image_to_token = Attention(
            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
        )

        self.skip_first_layer_pe = skip_first_layer_pe

    def forward(
        self, queries: Tensor, keys: Tensor, query_pe: Tensor, key_pe: Tensor
    ) -> Tuple[Tensor, Tensor]:
        # Self attention block
        if self.skip_first_layer_pe:
            queries = self.self_attn(q=queries, k=queries, v=queries)
        else:
            q = queries + query_pe
            attn_out = self.self_attn(q=q, k=q, v=queries)
            queries = queries + attn_out
        queries = self.norm1(queries)

        # Cross attention block, tokens attending to image embedding
        q = queries + query_pe
        k = keys + key_pe
        attn_out = self.cross_attn_token_to_image(q=q, k=k, v=keys)
        queries = queries + attn_out
        queries = self.norm2(queries)

        # MLP block
        mlp_out = self.mlp(queries)
        queries = queries + mlp_out
        queries = self.norm3(queries)

        # Cross attention block, image embedding attending to tokens
        q = queries + query_pe
        k = keys + key_pe
        attn_out = self.cross_attn_image_to_token(q=k, k=q, v=queries)
        keys = keys + attn_out
        keys = self.norm4(keys)

        return queries, keys


class Attention(nn.Module):
    """
    自注意力需要三个参数，K，Q，V，都是超参数，可以用偏差进行优化学习
    """

    def __init__(
        self,
        embedding_dim: int,
        num_heads: int,
        downsample_rate: int = 1,
    ) -> None:
        super().__init__()
        self.embedding_dim = embedding_dim
        self.internal_dim = embedding_dim // downsample_rate
        self.num_heads = num_heads
        assert self.internal_dim % num_heads == 0, "num_heads must divide embedding_dim."

        self.q_proj = nn.Linear(embedding_dim, self.internal_dim)
        self.k_proj = nn.Linear(embedding_dim, self.internal_dim)
        self.v_proj = nn.Linear(embedding_dim, self.internal_dim)
        self.out_proj = nn.Linear(self.internal_dim, embedding_dim)

    def _separate_heads(self, x: Tensor, num_heads: int) -> Tensor:
        b, n, c = x.shape
        x = x.reshape(b, n, num_heads, c // num_heads)
        return x.transpose(1, 2)  # B x N_heads x N_tokens x C_per_head

    def _recombine_heads(self, x: Tensor) -> Tensor:
        b, n_heads, n_tokens, c_per_head = x.shape
        x = x.transpose(1, 2)
        return x.reshape(b, n_tokens, n_heads * c_per_head)  # B x N_tokens x C

    def forward(self, q: Tensor, k: Tensor, v: Tensor) -> Tensor:
        # Input projections
        q = self.q_proj(q)
        k = self.k_proj(k)
        v = self.v_proj(v)

        # Separate into heads
        q = self._separate_heads(q, self.num_heads)
        k = self._separate_heads(k, self.num_heads)
        v = self._separate_heads(v, self.num_heads)

        # Attention
        _, _, _, c_per_head = q.shape
        attn = q @ k.permute(0, 1, 3, 2)  # B x N_heads x N_tokens x N_tokens
        attn = attn / math.sqrt(c_per_head)
        attn = torch.softmax(attn, dim=-1)

        # Get output
        out = attn @ v
        out = self._recombine_heads(out)
        out = self.out_proj(out)

        return out

我们再看看图像编码是怎么完成的：

import torch
import torch.nn as nn
import torch.nn.functional as F

from typing import Optional, Tuple, Type

from .common import LayerNorm2d, MLPBlock

# Vision Transformer结构
class ImageEncoderViT(nn.Module):
    def __init__(
        self,
        img_size: int = 1024,
        patch_size: int = 16,
        in_chans: int = 3,
        embed_dim: int = 768,
        depth: int = 12,
        num_heads: int = 12,
        mlp_ratio: float = 4.0,
        out_chans: int = 256,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Module] = nn.LayerNorm,
        act_layer: Type[nn.Module] = nn.GELU,
        use_abs_pos: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        global_attn_indexes: Tuple[int, ...] = (),
    ) -> None:
        """
        	img_size（int）：输入图像大小。
            patch_size（int）：步长大小。
            in_chans（int）：输入图像通道的数量。
            embed_dim（int）：补丁嵌入维度。
            depth（int）：ViT的深度。
            num_heads（int）：每个ViT块中的注意力头的数量。
            mlp_ratio（float）：mlp隐藏dim与嵌入dim的比率。
            qkv_bias（bool）：如果为True，则为查询、键、值添加可学习的偏差。
            norm_layer（nn.Module）：规范化层。
            act_layer（nn.Module）：激活层。
            use_abs_pos（bool）：如果为True，则使用绝对位置嵌入。
            use_rel_pos（bool）：如果为True，则将相对位置嵌入添加到注意力映射中。
            rel_pos_zero_init（bool）：如果为True，zero初始化相对位置参数。
            window_size（int）：窗口注意力块的窗口大小。
            global_attn_indexes（list）：使用全局注意力的块的索引。
        """
        super().__init__()
        self.img_size = img_size

        self.patch_embed = PatchEmbed(
            kernel_size=(patch_size, patch_size),
            stride=(patch_size, patch_size),
            in_chans=in_chans,
            embed_dim=embed_dim,
        )

        self.pos_embed: Optional[nn.Parameter] = None
        if use_abs_pos:
            # Initialize absolute positional embedding with pretrain image size.
            self.pos_embed = nn.Parameter(
                torch.zeros(1, img_size // patch_size, img_size // patch_size, embed_dim)
            )

        self.blocks = nn.ModuleList()
        for i in range(depth):
            block = Block(
                dim=embed_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                norm_layer=norm_layer,
                act_layer=act_layer,
                use_rel_pos=use_rel_pos,
                rel_pos_zero_init=rel_pos_zero_init,
                window_size=window_size if i not in global_attn_indexes else 0,
                input_size=(img_size // patch_size, img_size // patch_size),
            )
            self.blocks.append(block)

        self.neck = nn.Sequential(
            nn.Conv2d(
                embed_dim,
                out_chans,
                kernel_size=1,
                bias=False,
            ),
            LayerNorm2d(out_chans),
            nn.Conv2d(
                out_chans,
                out_chans,
                kernel_size=3,
                padding=1,
                bias=False,
            ),
            LayerNorm2d(out_chans),
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.patch_embed(x)
        if self.pos_embed is not None:
            x = x + self.pos_embed

        for blk in self.blocks:
            x = blk(x)

        x = self.neck(x.permute(0, 3, 1, 2))

        return x


class Block(nn.Module):
    """支持窗口注意力和残差传播块的Transformer"""
    """
    	窗口注意力就是一种掩码
    """

    def __init__(
        self,
        dim: int,
        num_heads: int,
        mlp_ratio: float = 4.0,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Module] = nn.LayerNorm,
        act_layer: Type[nn.Module] = nn.GELU,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        """
            dim（int）：输入通道的数量。
            num_heads（int）：每个ViT块中的注意力头的数量。
            mlp_ratio（float）：mlp隐藏dim与嵌入dim的比率。
            qkv_bias（bool）：如果为True，则为K,Q,V添加可学习的偏差。
            norm_layer（nn.Module）：规范化层。
            act_layer（nn.Module）：激活层。
            use_rel_pos（bool）：如果为True，则将相对位置嵌入添加到注意力映射中。
            rel_pos_zero_init（bool）：如果为True，zero初始化相对位置参数。
            window_size（int）：窗口注意力块的窗口大小。如果它等于0，那么利用全球关注。
            input_size（int或None）：用于计算相对位置的输入分辨率参数大小
        """
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads,
            qkv_bias=qkv_bias,
            use_rel_pos=use_rel_pos,
            rel_pos_zero_init=rel_pos_zero_init,
            input_size=input_size if window_size == 0 else (window_size, window_size),
        )

        self.norm2 = norm_layer(dim)
        self.mlp = MLPBlock(embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer)

        self.window_size = window_size

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        shortcut = x
        x = self.norm1(x)
        # Window partition
        if self.window_size > 0:
            H, W = x.shape[1], x.shape[2]
            x, pad_hw = window_partition(x, self.window_size)

        x = self.attn(x)
        # Reverse window partition
        if self.window_size > 0:
            x = window_unpartition(x, self.window_size, pad_hw, (H, W))

        x = shortcut + x
        x = x + self.mlp(self.norm2(x))

        return x


class Attention(nn.Module):
    """这是多头注意力"""

    def __init__(
        self,
        dim: int,
        num_heads: int = 8,
        qkv_bias: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.proj = nn.Linear(dim, dim)

        self.use_rel_pos = use_rel_pos
        if self.use_rel_pos:
            assert (
                input_size is not None
            ),
            # initialize relative positional embeddings
            self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
            self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        B, H, W, _ = x.shape
        # qkv with shape (3, B, nHead, H * W, C)
        qkv = self.qkv(x).reshape(B, H * W, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        # q, k, v with shape (B * nHead, H * W, C)
        q, k, v = qkv.reshape(3, B * self.num_heads, H * W, -1).unbind(0)

        attn = (q * self.scale) @ k.transpose(-2, -1)

        if self.use_rel_pos:
            attn = add_decomposed_rel_pos(attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W))

        attn = attn.softmax(dim=-1)
        x = (attn @ v).view(B, self.num_heads, H, W, -1).permute(0, 2, 3, 1, 4).reshape(B, H, W, -1)
        x = self.proj(x)

        return x


def window_partition(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, Tuple[int, int]]:
    """
    	堆叠不重复的窗口，一种不同于padding的方式。
    """
    B, H, W, C = x.shape

    pad_h = (window_size - H % window_size) % window_size
    pad_w = (window_size - W % window_size) % window_size
    if pad_h > 0 or pad_w > 0:
        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h))
    Hp, Wp = H + pad_h, W + pad_w

    x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows, (Hp, Wp)


def window_unpartition(
    windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
) -> torch.Tensor:
    """
    Window unpartition into original sequences and removing padding.
    Args:
        x (tensor): input tokens with [B * num_windows, window_size, window_size, C].
        window_size (int): window size.
        pad_hw (Tuple): padded height and width (Hp, Wp).
        hw (Tuple): original height and width (H, W) before padding.

    Returns:
        x: unpartitioned sequences with [B, H, W, C].
    """
    Hp, Wp = pad_hw
    H, W = hw
    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
    x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1)
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1)

    if Hp > H or Wp > W:
        x = x[:, :H, :W, :].contiguous()
    return x


def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor:
    """
    Get relative positional embeddings according to the relative positions of
        query and key sizes.
    Args:
        q_size (int): size of query q.
        k_size (int): size of key k.
        rel_pos (Tensor): relative position embeddings (L, C).

    Returns:
        Extracted positional embeddings according to relative positions.
    """
    max_rel_dist = int(2 * max(q_size, k_size) - 1)
    # Interpolate rel pos if needed.
    if rel_pos.shape[0] != max_rel_dist:
        # Interpolate rel pos.
        rel_pos_resized = F.interpolate(
            rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
            size=max_rel_dist,
            mode="linear",
        )
        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
    else:
        rel_pos_resized = rel_pos

    # Scale the coords with short length if shapes for q and k are different.
    q_coords = torch.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
    k_coords = torch.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)

    return rel_pos_resized[relative_coords.long()]


def add_decomposed_rel_pos(
    attn: torch.Tensor,
    q: torch.Tensor,
    rel_pos_h: torch.Tensor,
    rel_pos_w: torch.Tensor,
    q_size: Tuple[int, int],
    k_size: Tuple[int, int],
) -> torch.Tensor:
    """
    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
    Args:
        attn (Tensor): attention map.
        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).

    Returns:
        attn (Tensor): attention map with added relative positional embeddings.
    """
    q_h, q_w = q_size
    k_h, k_w = k_size
    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
    Rw = get_rel_pos(q_w, k_w, rel_pos_w)

    B, _, dim = q.shape
    r_q = q.reshape(B, q_h, q_w, dim)
    rel_h = torch.einsum("bhwc,hkc->bhwk", r_q, Rh)
    rel_w = torch.einsum("bhwc,wkc->bhwk", r_q, Rw)

    attn = (
        attn.view(B, q_h, q_w, k_h, k_w) + rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]
    ).view(B, q_h * q_w, k_h * k_w)

    return attn


class PatchEmbed(nn.Module):
    """
    Image to Patch Embedding.
    """

    def __init__(
        self,
        kernel_size: Tuple[int, int] = (16, 16),
        stride: Tuple[int, int] = (16, 16),
        padding: Tuple[int, int] = (0, 0),
        in_chans: int = 3,
        embed_dim: int = 768,
    ) -> None:
        """
        Args:
            kernel_size (Tuple): kernel size of the projection layer.
            stride (Tuple): stride of the projection layer.
            padding (Tuple): padding size of the projection layer.
            in_chans (int): Number of input image channels.
            embed_dim (int):  embed_dim (int): Patch embedding dimension.
        """
        super().__init__()

        self.proj = nn.Conv2d(
            in_chans, embed_dim, kernel_size=kernel_size, stride=stride, padding=padding
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.proj(x)
        # B C H W -> B H W C
        x = x.permute(0, 2, 3, 1)
        return x

我们再看看掩码层：

import torch
from torch import nn
from torch.nn import functional as F

from typing import List, Tuple, Type

from .common import LayerNorm2d


# 掩码的解码层，负责处理编码图片向量
class MaskDecoder(nn.Module):
    def __init__(
        self,
        *,
        transformer_dim: int,
        transformer: nn.Module,
        num_multimask_outputs: int = 3,
        activation: Type[nn.Module] = nn.GELU,
        iou_head_depth: int = 3,
        iou_head_hidden_dim: int = 256,
    ) -> None:
        super().__init__()
        self.transformer_dim = transformer_dim
        self.transformer = transformer

        self.num_multimask_outputs = num_multimask_outputs

        self.iou_token = nn.Embedding(1, transformer_dim)
        self.num_mask_tokens = num_multimask_outputs + 1
        self.mask_tokens = nn.Embedding(self.num_mask_tokens, transformer_dim)

        # 实现方法是卷积后打平，然后计算注意力
        self.output_upscaling = nn.Sequential(
            nn.ConvTranspose2d(transformer_dim, transformer_dim // 4, kernel_size=2, stride=2),
            LayerNorm2d(transformer_dim // 4),
            activation(),
            nn.ConvTranspose2d(transformer_dim // 4, transformer_dim // 8, kernel_size=2, stride=2),
            activation(),
        )
        self.output_hypernetworks_mlps = nn.ModuleList(
            [
                MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3)
                for i in range(self.num_mask_tokens)
            ]
        )

        self.iou_prediction_head = MLP(
            transformer_dim, iou_head_hidden_dim, self.num_mask_tokens, iou_head_depth
        )

    def forward(
        self,
        image_embeddings: torch.Tensor,
        image_pe: torch.Tensor,
        sparse_prompt_embeddings: torch.Tensor,
        dense_prompt_embeddings: torch.Tensor,
        multimask_output: bool,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        
        """
        	为了避免多次划分物体，这里使用了IOU方法进行抑制
        """
        masks, iou_pred = self.predict_masks(
            image_embeddings=image_embeddings,
            image_pe=image_pe,
            sparse_prompt_embeddings=sparse_prompt_embeddings,
            dense_prompt_embeddings=dense_prompt_embeddings,
        )

        # Select the correct mask or masks for outptu
        if multimask_output:
            mask_slice = slice(1, None)
        else:
            mask_slice = slice(0, 1)
        masks = masks[:, mask_slice, :, :]
        iou_pred = iou_pred[:, mask_slice]

        # Prepare output
        return masks, iou_pred

    def predict_masks(
        self,
        image_embeddings: torch.Tensor,
        image_pe: torch.Tensor,
        sparse_prompt_embeddings: torch.Tensor,
        dense_prompt_embeddings: torch.Tensor,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Predicts masks. See 'forward' for more details."""
        # Concatenate output tokens
        output_tokens = torch.cat([self.iou_token.weight, self.mask_tokens.weight], dim=0)
        output_tokens = output_tokens.unsqueeze(0).expand(sparse_prompt_embeddings.size(0), -1, -1)
        tokens = torch.cat((output_tokens, sparse_prompt_embeddings), dim=1)

        # Expand per-image data in batch direction to be per-mask
        src = torch.repeat_interleave(image_embeddings, tokens.shape[0], dim=0)
        src = src + dense_prompt_embeddings
        pos_src = torch.repeat_interleave(image_pe, tokens.shape[0], dim=0)
        b, c, h, w = src.shape

        # Run the transformer
        hs, src = self.transformer(src, pos_src, tokens)
        iou_token_out = hs[:, 0, :]
        mask_tokens_out = hs[:, 1 : (1 + self.num_mask_tokens), :]

        # Upscale mask embeddings and predict masks using the mask tokens
        src = src.transpose(1, 2).view(b, c, h, w)
        upscaled_embedding = self.output_upscaling(src)
        hyper_in_list: List[torch.Tensor] = []
        for i in range(self.num_mask_tokens):
            hyper_in_list.append(self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]))
        hyper_in = torch.stack(hyper_in_list, dim=1)
        b, c, h, w = upscaled_embedding.shape
        masks = (hyper_in @ upscaled_embedding.view(b, c, h * w)).view(b, -1, h, w)

        # Generate mask quality predictions
        iou_pred = self.iou_prediction_head(iou_token_out)

        return masks, iou_pred


# MLP神经元，不使用Torch自带的前馈神经网络
class MLP(nn.Module):
    def __init__(
        self,
        input_dim: int,
        hidden_dim: int,
        output_dim: int,
        num_layers: int,
        sigmoid_output: bool = False,
    ) -> None:
        super().__init__()
        self.num_layers = num_layers
        h = [hidden_dim] * (num_layers - 1)
        self.layers = nn.ModuleList(
            nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim])
        )
        self.sigmoid_output = sigmoid_output

    def forward(self, x):
        for i, layer in enumerate(self.layers):
            x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x)
        if self.sigmoid_output:
            x = F.sigmoid(x)
        return x

最后我们组合这些层，得到SAM模型：

import torch
from torch import nn
from torch.nn import functional as F

from typing import Any, Dict, List, Tuple

from .image_encoder import ImageEncoderViT
from .mask_decoder import MaskDecoder
from .prompt_encoder import PromptEncoder


class Sam(nn.Module):
    mask_threshold: float = 0.0
    image_format: str = "RGB"

    def __init__(
        self,
        image_encoder: ImageEncoderViT,
        prompt_encoder: PromptEncoder,
        mask_decoder: MaskDecoder,
        pixel_mean: List[float] = [123.675, 116.28, 103.53],
        pixel_std: List[float] = [58.395, 57.12, 57.375],
    ) -> None:
        """
            对象掩码：
            	image_encoder（ImageEncoderViT）：用于对图像到图像嵌入，这允许有效的掩模预测。
                prompt_encoder（PromptEncoder）：对各种类型的输入提示进行编码。
                mask_edecoder（MaskDecoder）：从图像嵌入中预测掩码以及编码提示。
                pixel_mean（list（float））：对输入图像中的像素进行归一化的平均值。
                pixel_sd（list（float））：用于对输入图像中的像素进行标准化的标准值。
        """
        super().__init__()
        self.image_encoder = image_encoder
        self.prompt_encoder = prompt_encoder
        self.mask_decoder = mask_decoder
        self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
        self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)

    @property
    def device(self) -> Any:
        return self.pixel_mean.device

    @torch.no_grad()
    def forward(
        self,
        batched_input: List[Dict[str, Any]],
        multimask_output: bool,
    ) -> List[Dict[str, torch.Tensor]]:
        """
        	根据预训练的图像获得相应的掩码（最后的过程）
        		首先需要确定输入图片的大小，结合Mask层的输出，就可以获得带有标签的掩码列表，在计算出最后的边框之后返回边框的掩码
        """
        input_images = torch.stack([self.preprocess(x["image"]) for x in batched_input], dim=0)
        image_embeddings = self.image_encoder(input_images)

        outputs = []
        for image_record, curr_embedding in zip(batched_input, image_embeddings):
            if "point_coords" in image_record:
                points = (image_record["point_coords"], image_record["point_labels"])
            else:
                points = None
            sparse_embeddings, dense_embeddings = self.prompt_encoder(
                points=points,
                boxes=image_record.get("boxes", None),
                masks=image_record.get("mask_inputs", None),
            )
            low_res_masks, iou_predictions = self.mask_decoder(
                image_embeddings=curr_embedding.unsqueeze(0),
                image_pe=self.prompt_encoder.get_dense_pe(),
                sparse_prompt_embeddings=sparse_embeddings,
                dense_prompt_embeddings=dense_embeddings,
                multimask_output=multimask_output,
            )
            masks = self.postprocess_masks(
                low_res_masks,
                input_size=image_record["image"].shape[-2:],
                original_size=image_record["original_size"],
            )
            masks = masks > self.mask_threshold
            outputs.append(
                {
                    "masks": masks,
                    "iou_predictions": iou_predictions,
                    "low_res_logits": low_res_masks,
                }
            )
        return outputs

    def postprocess_masks(
        self,
        masks: torch.Tensor,
        input_size: Tuple[int, ...],
        original_size: Tuple[int, ...],
    ) -> torch.Tensor:
        """
        	移除填充对原图像的影响（逆padding）
        """
        masks = F.interpolate(
            masks,
            (self.image_encoder.img_size, self.image_encoder.img_size),
            mode="bilinear",
            align_corners=False,
        )
        masks = masks[..., : input_size[0], : input_size[1]]
        masks = F.interpolate(masks, original_size, mode="bilinear", align_corners=False)
        return masks

    def preprocess(self, x: torch.Tensor) -> torch.Tensor:
        """ 规范化最后的输出 """
        # Normalize colors
        x = (x - self.pixel_mean) / self.pixel_std

        # Pad
        h, w = x.shape[-2:]
        padh = self.image_encoder.img_size - h
        padw = self.image_encoder.img_size - w
        x = F.pad(x, (0, padw, 0, padh))
        return x

成果

我们可以随便找一个图片实验一下：

这是一枚STM32F407VET6，我们放进算法里面看看掩码图：

我们发现，目标已经被老老实实的过滤出来了，甚至可以看到上面的按键

总结

本期博文主要介绍了ViT模型的实现SAM，我们从中看得出来Transformer的跨领域能力有多强，不仅可以横扫NLP，也可以跨界到CV行业进行物体分割，当然，我们可以为其增加一些新成分，比如过滤识别