Mmclassification training custom data

1 mmclassification install

         If the environment is installed mmclassification, Please skip this step .mmclassification Please refer to the blog for framework installation and debugging verification ：mmclassification Installation and commissioning _Coding Leaves blog -CSDN Blog _mmclassification install .

2 Data set preparation

        mmclassification The data set directory of is mainly composed of annotation files and image samples , The annotation file is stored in meta In the folder , Image sample exists train、val、test Under the folder , That is, they are used for training 、 Image samples for verification and testing . Image sample files are stored in train、val、test Under the folder , Pictures of the same category are stored in the same subfolder , The name of the subfolder is the name of the category to which the picture belongs .

        meta The folder mainly contains train.txt、val.txt and test.txt file .txt Each line in the file stores the image sample path and category id, As shown in the figure below .

         without meta Mark the file , Please refer to the blog ：mmclassification Annotation file generation _Coding Leaves blog -CSDN Blog , Generate meta Folder and under it txt file .

          The sample data of this article comes from minist Handwritten font visualization dataset , According to train、test Folder for storage , Download at ：minist Handwritten digital visualization data set - Deep learning document resources -CSDN download .

        Rename the downloaded dataset folder to Minist, also mmclassification New under the project directory data Folder , Put the data set in data Under folder . The storage path of the dataset is unlimited , Need to be below 3.3 Configure the corresponding path in section .

3 Custom datasets

3.1 newly build MyDataset

         stay mmclassification Under the project directory mmcls/datasets/ newly build mydataset.py file , Custom data loading class MyDataset, File name mydataset And class name MyDataset You can change .mydataset.py The contents of the document are as follows ： 

# -*- coding: utf-8 -*-
"""
 The official account of Lele perception school 
@author: https://blog.csdn.net/suiyingy
"""

import numpy as np

from .builder import DATASETS
from .base_dataset import BaseDataset



@DATASETS.register_module()
class MyDataset(BaseDataset):

    def load_annotations(self):
        assert isinstance(self.ann_file, str)

        data_infos = []
        with open(self.ann_file) as f:
            samples = [x.strip().split(' ') for x in f.readlines()]
            for filename, gt_label in samples:
                info = {'img_prefix': self.data_prefix}
                info['img_info'] = {'filename': filename}
                info['gt_label'] = np.array(gt_label, dtype=np.int64)
                data_infos.append(info)
            return data_infos

 3.2 take MyDataset Sign up to mmclassification frame

         stay mmcls/datasets/__init__.py Add the class defined above to the file MyDataset, As shown in the figure below ：

 3.3 Create a new dataset Profile

         stay mmclassification Project directory configs/_base_/datasets/ Under the folder , newly build mydataset.py file , It is mainly used to set the data set type 、 Data enhancement mode 、batch size (samples_per_gpu)、 Data set path and annotation file path 、 Model preservation cycle (interval). The contents of the document are as follows ：

# -*- coding: utf-8 -*-
"""
 The official account of Lele perception school 
@author: https://blog.csdn.net/suiyingy
"""
dataset_type = 'MyDataset'
classes = ['cat', 'bird', 'dog']  # The category names of your dataset

img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)

train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='RandomResizedCrop', size=224),
    dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='ImageToTensor', keys=['img']),
    dict(type='ToTensor', keys=['gt_label']),
    dict(type='Collect', keys=['img', 'gt_label'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='Resize', size=(256, -1)),
    dict(type='CenterCrop', crop_size=224),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='ImageToTensor', keys=['img']),
    dict(type='Collect', keys=['img'])
]

data = dict(
    train=dict(
        type=dataset_type,
        data_prefix='data/Minist/train',
        ann_file='data/Minist/meta/train.txt',
        classes=classes,
        pipeline=train_pipeline
    ),
    val=dict(
        type=dataset_type,
        data_prefix='data/Minist/test',
        ann_file='data/Minist/meta/test.txt',
        classes=classes,
        pipeline=test_pipeline
    ),
    test=dict(
        type=dataset_type,
        data_prefix='data/Minist/test',
        ann_file='data/Minist/meta/test.txt',
        classes=classes,
        pipeline=test_pipeline
    )
)
evaluation = dict(interval=1, metric='accuracy')

4 modify configs Model configuration file

         With configs/resnet/resnet18_8xb16_cifar10.py Configuration files, for example ,mmclassification The configuration file of usually contains the following 4 Parts of ：

_base_ = [
    '../_base_/models/resnet18_cifar.py', '../_base_/datasets/cifar10_bs16.py',
    '../_base_/schedules/cifar10_bs128.py', '../_base_/default_runtime.py'
]

        ../_base_/models/resnet18_cifar.py： Define model parameters , It mainly includes backbone network 、neck、head And number of categories .

        ../_base_/datasets/cifar10_bs16.py： Define data set enhancement methods and paths , That is to say 3.3 Section configuration file ,bs16 Express batch size by 16, namely samples_per_gpu=16.

        ../_base_/schedules/cifar10_bs128.py： Define training parameters , It mainly includes optimizer 、 Learning rate 、 Total training epoch Number .

        ../_base_/default_runtime.py： Define operating parameters , It mainly includes the model preservation cycle 、 Log output cycle, etc .

        configs The main modification is the data configuration file , Namely the '../_base_/datasets/cifar10_bs16.py' Replace it with 3.3 Section '../_base_/datasets/mydataset.py'. namely ：

5 Run the training program

        mmcls The basic training command is ：

python tools/train.py  Model configuration file 

         Example ：

python tools/train.py configs/resnet/resnet18_8xb16_cifar10.py

         Here we have put resnet18_8xb16_cifar10.py The document is in accordance with article 4 Section has been modified .

6 Running results

 7 【python Three dimensional deep learning 】python 3D point cloud from basic to deep learning _Coding Leaves blog -CSDN Blog _python Three dimensional point cloud

More 3D 、 Please pay attention to two-dimensional perception algorithm and financial quantitative analysis algorithm “ Lele perception school ” WeChat official account , And will continue to update .