MMPretrain代码课笔记（配置文件）

需要预先安装pytorch与cuda

clone mmpretrain源码，安装openmim

git clone https://github.com/open-mmlab/mmpretrain

cd mmpretrain

pip install openmim

安装mmpretrain算法库,注意要在mmpretrain目录下

mim install -e “.[multimodal]”

后进入python环境

In[1]:import mmpretrain

In[2]:mmpretrain.__version__

In[3]:from mmpretrain import get_model, list_models, inference_model

In[4]:list_models(task='Image Classification', pattern='resnet18') #获取分类任务相关的名字带有resnet18的所有模型

In[5]:list_models(task='Image Caption', pattern='blip') #获取图像描述相关的名字带有blip的所有模型

In[6]: model = get_model('resnet18_8xb16_cifar10') #获取resnet18在cifar10上的模型

In[7]:type(model) #获取类型--是ImageClassifier

In[8]:type(model.backbone)#应该是ResNet类型

In[9]: inference.model(model, 'demo/bird.jpeg', show=True)#此时模型未训练结果混乱
#直接get_model没有额外参数获得的模型是没有预训练权重的

In[10]：inference.model('resnet18_8xb16_cifar10'', 'demo/bird.jpeg',show=True)#正确用法

接下来进行在一个数据集用ResNet的微调训练

准备数据集

#从clone的mmpretrain目录出发
mkdir data
cd data
tar -xf ~/Downloads/cats_dogs_dataset.tar#提前下载的数据集https://download.openmmlab.com/mmclassification/dataset/cats_dogs_dataset.tar
cd cats_dogs_dataset 

cd ../../

#编写配置文件
#config/resnet目录有有可参考的ResNet18的配置文件

#使用最简单resnet18_8xb32_in1k.py
vi configs/resnet/resnet18_8xb32_in1k.py

 这里将四个文件的配置信息组合在了一起：

 模型配置文件：一些初始化参数，backbone主干网络，一个resnet网络用于特征提取，neck是pooling层把backbone提取每张图的特征转为1维向量，head分类头。

可以用如下方法修改配置信息

from mmengine impor Config

cfg = Config.fromfile("./configs/resnet/resnet18_8xb32_in1k.py")

cfg.model

cfg.model.head.num_classes = 2 #修改配置

cfg。model

数据集配置：

# dataset settings
dataset_type = 'ImageNet'
data_preprocessor = dict(
    num_classes=1000,
    # RGB format normalization parameters
    mean=[123.675, 116.28, 103.53],
    std=[58.395, 57.12, 57.375],
    # convert image from BGR to RGB
    to_rgb=True,
)

train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='RandomResizedCrop', scale=224),
    dict(type='RandomFlip', prob=0.5, direction='horizontal'),
    dict(type='PackInputs'),
]

test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='ResizeEdge', scale=256, edge='short'),
    dict(type='CenterCrop', crop_size=224),
    dict(type='PackInputs'),
]

train_dataloader = dict(
    batch_size=32,
    num_workers=5,
    dataset=dict(
        type=dataset_type,
        data_root='data/imagenet',
        ann_file='meta/train.txt',
        data_prefix='train',
        pipeline=train_pipeline),
    sampler=dict(type='DefaultSampler', shuffle=True),
)

val_dataloader = dict(
    batch_size=32,
    num_workers=5,
    dataset=dict(
        type=dataset_type,
        data_root='data/imagenet',
        ann_file='meta/val.txt',
        data_prefix='val',
        pipeline=test_pipeline),
    sampler=dict(type='DefaultSampler', shuffle=False),
)
val_evaluator = dict(type='Accuracy', topk=(1, 5))

# If you want standard test, please manually configure the test dataset
test_dataloader = val_dataloader
test_evaluator = val_evaluator

 注意：配置文件在加载完成后，中间变量的连接关系就没有了，即train_dataloader的type与变量dataset_type没有连接了。

规划配置：

 记录了训练和测试的流程，optim_wrapper配置了优化器参数；param_scheduler参数归化器，用于训练期间根据训练结果修改学习率等参数

运行参数配置：

# defaults to use registries in mmpretrain
default_scope = 'mmpretrain'

# configure default hooks
default_hooks = dict(
    # record the time of every iteration.
    timer=dict(type='IterTimerHook'),

    # print log every 100 iterations.
    logger=dict(type='LoggerHook', interval=100),

    # enable the parameter scheduler.
    param_scheduler=dict(type='ParamSchedulerHook'),

    # save checkpoint per epoch.
    checkpoint=dict(type='CheckpointHook', interval=1),

    # set sampler seed in distributed evrionment.
    sampler_seed=dict(type='DistSamplerSeedHook'),

    # validation results visualization, set True to enable it.
    visualization=dict(type='VisualizationHook', enable=False),
)

# configure environment
env_cfg = dict(
    # whether to enable cudnn benchmark
    cudnn_benchmark=False,

    # set multi process parameters
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),

    # set distributed parameters
    dist_cfg=dict(backend='nccl'),
)

# set visualizer
vis_backends = [dict(type='LocalVisBackend')]
visualizer = dict(type='UniversalVisualizer', vis_backends=vis_backends)

# set log level
log_level = 'INFO'

# load from which checkpoint
load_from = None

# whether to resume training from the loaded checkpoint
resume = False

# Defaults to use random seed and disable `deterministic`
randomness = dict(seed=None, deterministic=False)

指和实验相关的附加功能，环境配置等参数，大部分情况不需要修改。介绍几个常用的

logger日志配置，interval=100表示间隔100次迭代打印日常

checkpoint权重保存配置，可以设置max_keep_ckpys=5参数保存最后几个权重，save_best='auto'就会保存目前为止精度最高的模型

randomness可以设置随机数种子

在projects目录下新建一个目录存放自定义配置文件

mkdir projects/cat_dog
cd projects/cat_dog
vi resnent18_finetune.py

将上述四个配置文件内容全部复制到这个新建文件中。

然后加载预训练权重，

在模型配置处加入init_cfg

# model settings
model = dict(
    type='ImageClassifier',
    backbone=dict(
        type='ResNet',
        depth=18,
        num_stages=4,
        out_indices=(3, ),
        style='pytorch'),
    neck=dict(type='GlobalAveragePooling'),
    head=dict(
        type='LinearClsHead',
        num_classes=1000,
        in_channels=512,
        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
        topk=(1, 5),
    )
    init_cfg=dict(type='Rewtrained', checkpoint='xxx')#xxx为模型链接
)

在数据集配置中修改dataset_type = 'CustomDataset'；修改train_dataloder等中的data_root为新数据路径，然后删除无关数据。将评测指标相关配置的的val_evaluator中的topk改成1

# dataset settings
dataset_type = 'CustomDataset'
data_preprocessor = dict(
    num_classes=1000,
    # RGB format normalization parameters
    mean=[123.675, 116.28, 103.53],
    std=[58.395, 57.12, 57.375],
    # convert image from BGR to RGB
    to_rgb=True,
)

train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='RandomResizedCrop', scale=224),
    dict(type='RandomFlip', prob=0.5, direction='horizontal'),
    dict(type='PackInputs'),
]

test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='ResizeEdge', scale=256, edge='short'),
    dict(type='CenterCrop', crop_size=224),
    dict(type='PackInputs'),
]

train_dataloader = dict(
    batch_size=32,
    num_workers=5,
    dataset=dict(
        type=dataset_type,
        data_root='../../data/cats_dogs_dataset/training_set',
        pipeline=train_pipeline),
    sampler=dict(type='DefaultSampler', shuffle=True),
)

val_dataloader = dict(
    batch_size=32,
    num_workers=5,
    dataset=dict(
        type=dataset_type,
        data_root='../../data/cats_dogs_dataset/val_set/',
        pipeline=test_pipeline),
    sampler=dict(type='DefaultSampler', shuffle=False),
)
val_evaluator = dict(type='Accuracy', topk=1)
# If you want standard test, please manually configure the test dataset
val_dataloader = dict(
    batch_size=32,
    num_workers=5,
    dataset=dict(
        type=dataset_type,
        data_root='../../data/cats_dogs_dataset/test_set/',
        pipeline=test_pipeline),
    sampler=dict(type='DefaultSampler', shuffle=False),
)


test_evaluator = val_evaluator

最后修改训练配置将epoch改成5，因为有加载预训练权重，模型收敛很快。

配置文件准备好就可以开始训练了

使用mim命令可以在任意位置很容易开始训练,传入训练脚本与保存位置

mim train mmpretrain resnet18_finetune.py --work-dir=./exp

同样，mim test mmpretrain传入模型与训练出的权重文件就可以进行测试

mim test mmpretrain resnet18_finetune.py --checkpoint exp/epoch_5.pth

mim test mmpretrain resnet18_finetune.py --checkpoint exp/epoch_5.pth --put result.pkl

加上--out result.pkl就可以将每个样本的测试结果都保存在result.pkl文件里

然后就可以使用mmpretrain中的各种分析工具进行分析,例如

mim run mmpretrain analyze_result resnet18_finetune.py result.pkl --out-dir analyze

cd analuze

ls

tree ./

就可以看到测试正确和错误结果以及每张图测试得分。

使用ImageClassificationInferencer API这样inferen就可以进行单张图片的预测

from mmpretrain import ImageClassificationInferencer

inferencer = ImageClassificationInferencer('./resnet18_finetune.py', pretrained='exp/epoch_5.pth')

inferen("xxx.jpg", show=Tyre)