Swin_transformer网络

已于 2022-03-06 16:36:56 修改
阅读量350 收藏 1
点赞数
分类专栏： 代码训练 文章标签： transformer 深度学习
于 2021-09-09 11:18:52 首次发布
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_42820467/article/details/120195458
版权
本文深入探讨了Swin Transformer网络结构，它是一种用于计算机视觉任务的先进模型。该模型由多个基本层组成，每个层包含窗口自注意力机制、多层感知机和残差连接。通过逐步下采样，模型能够处理不同分辨率的特征，适用于图像分类、目标检测等任务。
摘要由CSDN通过智能技术生成

在这里插入图片描述

# 打印出来的网络结构
ft_net_swin(
  (model): SwinTransformer(
    (patch_embed): PatchEmbed(
      (proj): Conv2d(3, 128, kernel_size=(4, 4), stride=(4, 4))
      (norm): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
    )
    (pos_drop): Dropout(p=0.0, inplace=False)
    (layers): Sequential(
      (0): BasicLayer(
        dim=128, input_resolution=(56, 56), depth=2
        (blocks): ModuleList(
          (0): SwinTransformerBlock(
            (norm1): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=128, out_features=384, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=128, out_features=128, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): Identity()
            (norm2): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=128, out_features=512, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=512, out_features=128, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (1): SwinTransformerBlock(
            (norm1): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=128, out_features=384, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=128, out_features=128, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=128, out_features=512, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=512, out_features=128, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
        )
        (downsample): PatchMerging(
          input_resolution=(56, 56), dim=128
          (reduction): Linear(in_features=512, out_features=256, bias=False)
          (norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
        )
      )
      (1): BasicLayer(
        dim=256, input_resolution=(28, 28), depth=2
        (blocks): ModuleList(
          (0): SwinTransformerBlock(
            (norm1): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=256, out_features=768, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=256, out_features=256, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=256, out_features=1024, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=1024, out_features=256, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (1): SwinTransformerBlock(
            (norm1): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=256, out_features=768, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=256, out_features=256, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=256, out_features=1024, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=1024, out_features=256, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
        )
        (downsample): PatchMerging(
          input_resolution=(28, 28), dim=256
          (reduction): Linear(in_features=1024, out_features=512, bias=False)
          (norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        )
      )
      (2): BasicLayer(
        dim=512, input_resolution=(14, 14), depth=18
        (blocks): ModuleList(
          (0): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (1): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (2): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (3): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (4): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (5): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (6): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (7): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (8): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (9): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (10): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (11): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (12): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (13): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (14): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (15): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (16): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (17): SwinTransformerBlock(
            (norm1): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=512, out_features=1536, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=512, out_features=512, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=512, out_features=2048, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=2048, out_features=512, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
        )
        (downsample): PatchMerging(
          input_resolution=(14, 14), dim=512
          (reduction): Linear(in_features=2048, out_features=1024, bias=False)
          (norm): LayerNorm((2048,), eps=1e-05, elementwise_affine=True)
        )
      )
      (3): BasicLayer(
        dim=1024, input_resolution=(7, 7), depth=2
        (blocks): ModuleList(
          (0): SwinTransformerBlock(
            (norm1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=1024, out_features=3072, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=1024, out_features=1024, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=1024, out_features=4096, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=4096, out_features=1024, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
          (1): SwinTransformerBlock(
            (norm1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
            (attn): WindowAttention(
              (qkv): Linear(in_features=1024, out_features=3072, bias=True)
              (attn_drop): Dropout(p=0.0, inplace=False)
              (proj): Linear(in_features=1024, out_features=1024, bias=True)
              (proj_drop): Dropout(p=0.0, inplace=False)
              (softmax): Softmax(dim=-1)
            )
            (drop_path): DropPath()
            (norm2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
            (mlp): Mlp(
              (fc1): Linear(in_features=1024, out_features=4096, bias=True)
              (act): GELU()
              (fc2): Linear(in_features=4096, out_features=1024, bias=True)
              (drop): Dropout(p=0.0, inplace=False)
            )
          )
        )
      )
    )
    (norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
    (avgpool): AdaptiveAvgPool1d(output_size=1)
    (head): Sequential()
  )
  (classifier): ClassBlock(
    (add_block): Sequential(
      (0): Linear(in_features=1024, out_features=512, bias=True)
      (1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (2): Dropout(p=0.5, inplace=False)
    )
    (classifier): Sequential(
      (0): Linear(in_features=512, out_features=751, bias=True)
    )
  )
)
多笑一笑啊
关注
专栏目录
Swin-Transformer网络结构详解
@bangbang的博客
09-10 459
论文这篇论文是由微软亚洲研究院于2021年在ICCV上发表的，作者基本上都是华人，该论文也获得2021年的best paper。关于它的性能超越了之前的所有网络，在coco数据集上在目标检测任务中超过之前网络2.7个AP,在实例分割任务中提升有2.6个点, 在视觉的多个领域中都带来了很大的进度。通过给的这些参数可以知道的性能是非常好的，一般我们自己通过网络创新精度提升一个点，就可以发文章了。Pytorch实现代码： pytorch_classification/swin_transformer。
使用swin_transformer做自己数据集的图像分类
04-30
使用的数据集共有5种类别，使用Swin-T预训练模型进行训练。
理解Swin Transformer（结合代码）
Orange_sparkle的博客
09-09 2089
1）输入一张RGB彩色图片2）经过PatchEmbed层得到shape=[H/4, W/4, embed_dim]的特征矩阵3）通过循环设置的BasicLayer（Swin Transformer Block + Patch Merging）结构得到shape=[H/32, W/32, embed_dim*8]的提取出来的特征图4）通过最后的head进行分类，head包括全局平均池化，输出shape=[B, C, 1]，再通过全连接层，输出维度为类别数，得到分类结果。
狗都能看懂的Swin Transformer的讲解和代码实现
个人学习笔记
08-21 1763
狗都能看懂的Swin Transformer和代码复现
【计算机视觉】Swin-Transformer 网络结构解析
Danger的博客
04-02 1776
【计算机视觉】Swin-Transformer 网络结构解析
30_Swin-Transformer网络结构详解
https://github.com/foxpup11?tab=repositories
07-19 1448
https://www.bilibili.com/video/BV1pL4y1v7jC/?spm_id_from=333.999.0.0&vd_source=7dace3632125a1ef7fd32c285eb2fbac
查看swintransformer的网络结构
qq_45893319的博客
01-04 1017
import os import json import torch from PIL import Image from torchvision import transforms import matplotlib.pyplot as plt #导入对应的模型 from model import swin_tiny_patch4_window7_224 as create_model device = torch.device("cuda:0" if torch.cuda.is_available()
神经网络学习笔记5——Swin-Transformer网络
qq_45848817的博客
10-21 1万+
Swim Transformer是特为视觉领域设计的一种分层Transformer结构。Swin 的两大特性是滑动窗口和分层表示。滑动窗口在局部不重叠的窗口中计算自注意力，并允许跨窗口连接。分层结构允许模型适配不同尺度的图片，并且计算复杂度与图像大小呈线性关系，也因此被人成为披着transformer皮的CNN。
本科课设-基于swin_transformer实现了道路车道的实例分割python源码.zip
最新发布
09-18
本科课设-基于swin_transformer实现了道路车道的实例分割python源码.zip本科课设-基于swin_transformer实现了道路车道的实例分割python源码.zip本科课设-基于swin_transformer实现了道路车道的实例分割python源码....
基于swin_transformer算法实现了道路车道线实例分割python源码.zip
08-22
基于Swin_Transformer算法的道路车道线实例分割项目 是一个利用深度学习技术，特别是Swin Transformer模型，来实现道路图像中车道线的精确实例分割的项目。Swin Transformer是一种基于Transformer结构的模型，它在...
Swin-Transformer:这是“变形金刚”的官方实现
05-10
变形金刚 作者：*，*，*，*，，，和。 此回购协议是的正式实现。 当前，它包含用于以下任务的代码和模型： 图像分类：包含在此仓库中。 有关快速，请参见 。 对象检测和实例分割：有关对象检测，请参见 。 语义分割：有关语义分割的信息，请参见 。 更新 2021年4月12日 初始提交： 提供了ImageNet-1K（ ， ， ）和ImageNet-22K（ ， ）上。 提供了用于ImageNet-1K图像分类，COCO对象检测和ADE20K语义分割的受支持代码和模型。 在分支提供了用于的cuda内核实现。 介绍 欧亚变压器（名称Swin代表对于s hifted赢得DOW）最初描述 ，其干练作为一个通用的骨干，为计算机视觉。 它基本上是一个分层的Transformer，其表示是通过移动窗口来计算的。 通过将自注意计算限制为不重叠的局部窗口，同时允许跨窗口连接，移位的窗
swin_transformer pytorch代码
04-19
首先，`model.py`是核心的模型定义文件，其中包含了Swin Transformer的网络结构。Swin Transformer的核心是其窗口内的自注意力机制，它将输入的图像分成多个非重叠的窗口，并在每个窗口内进行自注意力计算，减少了...
swin_tiny_patch4_window7_224.pth
05-07
Swin Transformer是近年来提出的一种基于Transformer架构的新型卷积神经网络（CNN），由卢卡斯·吴（Liang-Chieh Chen）等人在2021年提出。这个模型主要应用于图像识别、分割等计算机视觉任务，并因其高效性和准确性...
SwinTransformer网络架构
QT-Smile
06-11 844
SwinTransformer网络架构
【论文阅读】Swin-transformer网络结构详解
aixiaomi123的博客
01-19 2282
Swin-transformer 网络架构详解
Swin-Transformer详解
热门推荐
hellozhxy的博客
10-19 1万+
Swin-Transformer是2021年微软研究院发表在ICCV上的一篇文章，并且已经获得的荣誉称号。虽然在图像分类方面的结果令人鼓舞，但是由于其低分辨率特性映射和复杂度随图像大小的二次增长，其结构不适合作为密集视觉任务或高分辨率输入图像的通过骨干网路。为了最佳的精度和速度的权衡，提出了Swin-Transformer结构。论文名称原论文地址官方开源代码地址Pytorch实现代码Tensorflow2实现代码如下图所示为：Swin-Transformer与ViT的对比结构。采样方式。
经典网络模型系列——Swin-Transformer详细讲解与代码实现
hellozhxy的博客
10-19 4923
【代码】经典网络模型系列——Swin-Transformer详细讲解与代码实现。
swin-transformer
qq_54809548的博客
07-14 5175
通过将图像或文本分割成Token，可以更好地理解和处理图像或文本数据，从而提高计算机视觉任务的性能和效果。swin transformer用CNN的方式去构建网络，用卷积局部化的思想，基于self-attention的方式去搭建网络，实现图像编码。其主要原因在于：transformer的模型结构就导致学习数据非常的难，因此需要更多的hidden layer来学习数据的本质分布。例如，在图像分类或目标检测任务中，可以将图像划分为不同的图像块，并将每个图像块视为一个Token来进行处理。
残差网络 mobilenet swin_transformer
04-14
Swin Transformer是一种基于Transformer的图像分类模型，它在传统Transformer的基础上进行了改进，以适应图像数据的处理。Swin Transformer引入了分层的窗口机制，将图像划分为多个窗口，并在每个窗口内进行...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值