深度学习主干网络构建利器——timm

---

一、Timm创建模型

加载预训练模型

代码如下：

import timm
m = timm.create_model('vit_large_r50_s32_224.augreg_in21k', pretrained=True)
m.eval()

二、查看模型列表

1.引入库

代码如下：

import timm
from pprint import pprint
model_names = timm.list_models(pretrained=True)
pprint(model_names)

输出结果
[‘bat_resnext26ts.ch_in1k’,
‘beit_base_patch16_224.in22k_ft_in22k’,
‘beit_base_patch16_224.in22k_ft_in22k_in1k’,
‘beit_base_patch16_384.in22k_ft_in22k_in1k’,
‘beit_large_patch16_224.in22k_ft_in22k’,
‘beit_large_patch16_224.in22k_ft_in22k_in1k’,
‘beit_large_patch16_384.in22k_ft_in22k_in1k’,
‘beit_large_patch16_512.in22k_ft_in22k_in1k’,
‘beitv2_base_patch16_224.in1k_ft_in1k’,
‘beitv2_base_patch16_224.in1k_ft_in22k’,
‘beitv2_base_patch16_224.in1k_ft_in22k_in1k’,
‘beitv2_large_patch16_224.in1k_ft_in1k’,
…]

## 根据模型名称进行过滤
import timm
from pprint import pprint
model_names = timm.list_models('*resne*t*')
pprint(model_names)

2.模型微调

代码如下（示例）：

model = timm.create_model('mobilenetv3_large_100', pretrained=True, num_classes=2)
pprint(model)

模型输入通道的修改

---

3.特征提取

import timm
import torch
x = torch.randn(1, 3, 224, 224)
model = timm.create_model('mobilenetv3_large_100', pretrained=True)
pprint(model)
features = model.forward_features(x)
print(features.shape)

4.指定模型特征输出

# 对输出特征进行限制
import torch
import timm
# # 指定输出索引，并修改stride
m = timm.create_model('resnet34', features_only=True,output_stride = 8, pretrained=True)
# print("模型修改前")
# pprint(m)
# print("模型修改后")
# output_stride降采样的倍数
# m = timm.create_model('resnet34', output_stride = 8,out_indices = [4,],features_only=True, pretrained=True)
# pprint(m)
print(f'Feature channels: {m.feature_info.channels()}')
print(f'Feature reduction: {m.feature_info.reduction()}')
o = m(torch.randn(2, 3, 256, 256))
for x in o:
    print(x.shape)
print(o[0].shape)

总结

timm还支持图像变换，详细还是去官网看文档吧：Timm