VMambe应用目标检测backbone之VSSM结构打印

[Releases · MzeroMiko/VMamba · GitHub](https://github.com/MzeroMiko/VMamba/releases)

安装torchvision等：

尝试版本1：wins上python=3.10, torch=1.13.1  成功

    pip install torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117

尝试版本2 ：wins上python=3.10, torch=2.1.0

  pip install torchvision==0.16.0+cu118 torchaudio==2.1.0 --extra-index-url https://download.pytorch.org/whl/cu118

模型pipline:

---

熟悉模型结构：

import os
from functools import partial
from typing import Callable

import torch
from torch import nn
from torch.utils import checkpoint

from mmengine.model import BaseModule
from mmdet.registry import MODELS as MODELS_MMDET
from mmseg.registry import MODELS as MODELS_MMSEG

def import_abspy(name="models", path="classification/"):
    import sys
    import importlib
    path = os.path.abspath(path)
    assert os.path.isdir(path)
    sys.path.insert(0, path)
    module = importlib.import_module(name)
    sys.path.pop(0)
    #打印
    print("detection imported module: {}".format(module.__name__))
    return module

def main():
    build = import_abspy(
        "models",
        os.path.join(os.path.dirname(os.path.abspath(__file__)), "../classification/"),
    )

    Backbone_VSSM = build.vmamba.Backbone_VSSM

    class MM_VSSM(BaseModule, Backbone_VSSM):
        def __init__(self, *args, **kwargs):
            BaseModule.__init__(self)
            Backbone_VSSM.__init__(self, *args, **kwargs)

    # 测试初始化 MM_VSSM 类
    model = MM_VSSM()

    # 打印模型的类名，确保导入和初始化正确
    print("Model class:", model.__class__.__name__)
    print("Model:", model)
    print(model.__dict__)


if __name__ == "__main__":
    main()

print("Model:", model)

Model: MM_VSSM( 
  (patch_embed): Sequential(                                                                                          VMamba1427
    (0): Conv2d(3, 96, kernel_size=(4, 4), stride=(4, 4))
    (1): Permute()                   
    (2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
  )
  (layers): ModuleList(
    (0): Sequential(
      (blocks): Sequential(
        (0): VSSBlock(
          (norm): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (op): SS2D(
            (out_norm): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
            (in_proj): Linear(in_features=96, out_features=384, bias=False)
            (act): SiLU()
            (conv2d): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
            (out_act): Identity()
            (out_proj): Linear(in_features=192, out_features=96, bias=False)
            (dropout): Identity()
          )
          (drop_path): timm.DropPath(0.0)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
        )
        (1): VSSBlock(
          (norm): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (op): SS2D(
            (out_norm): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
            (in_proj): Linear(in_features=96, out_features=384, bias=False)
            (act): SiLU()
            (conv2d): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
            (out_act): Identity()
            (out_proj): Linear(in_features=192, out_features=96, bias=False)
            (dropout): Identity()
          )
          (drop_path): timm.DropPath(0.0071428571827709675)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
        )
      )
      (downsample): Sequential(
        (0): Permute()
        (1): Conv2d(96, 192, kernel_size=(2, 2), stride=(2, 2))
        (2): Permute()
        (3): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
      )
    )
    (1): Sequential(
      (blocks): Sequential(
        (0): VSSBlock(
          (norm): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (op): SS2D(
            (out_norm): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
            (in_proj): Linear(in_features=192, out_features=768, bias=False)
            (act): SiLU()
            (conv2d): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
            (out_act): Identity()
            (out_proj): Linear(in_features=384, out_features=192, bias=False)
            (dropout): Identity()
          )
          (drop_path): timm.DropPath(0.014285714365541935)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
        )
        (1): VSSBlock(
          (norm): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (op): SS2D(
            (out_norm): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
            (in_proj): Linear(in_features=192, out_features=768, bias=False)
            (act): SiLU()
            (conv2d): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
            (out_act): Identity()
            (out_proj): Linear(in_features=384, out_features=192, bias=False)
            (dropout): Identity()
          )
          (drop_path): timm.DropPath(0.02142857201397419)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
        )
      )
      (downsample): Sequential(
        (0): Permute()
        (1): Conv2d(192, 384, kernel_size=(2, 2), stride=(2, 2))
        (2): Permute()
        (3): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
      )
    )
    (2): Sequential(
      (blocks): Sequential(
        (0): VSSBlock(
          (norm): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (op): SS2D(
            (out_norm): LayerNorm((768,), eps=1e-05, elementwi