当前位置：首页 > news >正文

timm教程翻译：（一）Overview

news 2025/10/19 6:58:04

https://timm.fast.ai/

1.How to use(如何使用)

1.1 Create a model（创造一个model）

import timm 
import torchmodel = timm.create_model('resnet34')
x     = torch.randn(1, 3, 224, 224)
model(x).shape

torch.Size([1, 1000])

使用 timm 创建模型就是这么简单。create_model 函数是一个工厂方法，可用于创建 timm 库中超过 300 个模型。
要创建预训练模型，只需传入 pretrained=True。

pretrained_resnet_34 = timm.create_model('resnet34', pretrained=True)

Downloading: "https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet34-43635321.pth" to /home/tmabraham/.cache/torch/hub/checkpoints/resnet34-43635321.pth

要创建具有自定义类别数的模型，只需传入 num_classes=<number_of_classes>。

import timm 
import torchmodel = timm.create_model('resnet34', num_classes=10)
x     = torch.randn(1, 3, 224, 224)
model(x).shape

torch.Size([1, 10])

1.2 List Models with Pretrained Weights(列出所有预训练过的模型)

timm.list_models() 返回 timm 中可用模型的完整列表。要查看预训练模型的完整列表，请在 list_models 中传入 pretrained=True。

avail_pretrained_models = timm.list_models(pretrained=True)
len(avail_pretrained_models), avail_pretrained_models[:5]

(592,['adv_inception_v3','bat_resnext26ts','beit_base_patch16_224','beit_base_patch16_224_in22k','beit_base_patch16_384'])

There are a total of 271 models with pretrained weights currently available in timm!

1.3 Search for model architectures by Wildcard

It is also possible to search for model architectures using Wildcard as below:
还可以使用 Wildcard 搜索模型架构，如下所示：

all_densenet_models = timm.list_models('*densenet*')
all_densenet_models

['densenet121','densenet121d','densenet161','densenet169','densenet201','densenet264','densenet264d_iabn','densenetblur121d','tv_densenet121']

1.4 Fine-tune timm model in fastai

The fastai library has support for fine-tuning models from timm:
fastai 库支持来自 timm 的微调模型：

from fastai.vision.all import *path = untar_data(URLs.PETS)/'images'
dls = ImageDataLoaders.from_name_func(path, get_image_files(path), valid_pct=0.2,label_func=lambda x: x[0].isupper(), item_tfms=Resize(224))# if a string is passed into the model argument, it will now use timm (if it is installed)
learn = vision_learner(dls, 'vit_tiny_patch16_224', metrics=error_rate)learn.fine_tune(1)