MMLAB-AI训练营笔记——第一次作业

这里记录一下作业的流程以及心得。

1. 安装依赖库

主要是根据mmclassification的文档，安装了pytorch，mmcv以及mmclassification，具体可以参考这篇文档。

2. 下载数据集

根据作业中给出的数据集官网，下载好后，在mmclassification的文件夹中新建一个data文件夹，把数据放入后解压。

3. 对数据集进行分割，并生成.txt文件

话不多说，直接上代码

import os
from shutil import copy, rmtree
import random

def make_file(file_path: str):
    if os.path.exists(file_path):
        # 如果文件夹存在，则先删除原文件夹在重新创建
        rmtree(file_path)
    os.makedirs(file_path)

# 保证随机可复现
random.seed(0)#保证每次随机抽取的都可以复现

# 将数据集中20%的数据划分到验证集中
split_rate = 0.2 #这里填多少 就是验证集的比例是多少，比如填0.1就是验证集的数量占总数据集的10%

data_path = './data/flower_dataset'#数据集存放的地方，建议在程序所在的文件夹下新建一个data文件夹，将需要划分的数据集存放进去
data_root = './data/flower_dataset'  #这里是生成的训练集和验证集所处的位置，这里设置的是在当前文件夹下。
data_class = [cla for cla in os.listdir(data_path)]
print("数据的种类分别为：")
print(data_class)# 输出数据种类，数据种类默认为读取的文件夹的名称
# 建立保存训练集的文件夹
train_data_root = os.path.join(data_root, "train") #训练集的文件夹名称为 train
make_file(train_data_root)
for num_class in data_class:
    # 建立每个类别对应的文件夹
    make_file(os.path.join(train_data_root, num_class))
# 建立保存验证集的文件夹
val_data_root = os.path.join(data_root, "val")#验证集的文件夹名称为 val
make_file(val_data_root)
for num_class in data_class:
    # 建立每个类别对应的文件夹
    make_file(os.path.join(val_data_root, num_class))

for num_class in data_class:
    num_class_path = os.path.join(data_path, num_class)
    images = os.listdir(num_class_path)
    num = len(images)
    val_index = random.sample(images, k=int(num*split_rate))   #随机抽取图片
    for index, image in enumerate(images):
        if image in val_index:
            # 将划分到验证集中的文件复制到相应目录
            data_image_path = os.path.join(num_class_path, image)
            val_new_path = os.path.join(val_data_root, num_class)
            copy(data_image_path, val_new_path)
            try:
                eval_f = open("./data/flower_dataset/val.txt", "a+")
                eval_f.write(str(num_class)+"/"+str(image))
                eval_f.write(" ")
                if num_class == "daisy":
                    eval_f.write("0")
                elif num_class == "dandelion":
                    eval_f.write("1")
                elif num_class == "rose":
                    eval_f.write("2")
                elif num_class == "sunflower":
                    eval_f.write("3")
                else:
                    eval_f.write("4")
                eval_f.write("\n")
            except FileExistsError as e:
                print(e)
                exit(1)
        else:
            # 将划分到训练集中的文件复制到相应目录
            data_image_path = os.path.join(num_class_path, image)
            train_new_path = os.path.join(train_data_root, num_class)
            copy(data_image_path, train_new_path)
            try:
                train_f = open("./data/flower_dataset/train.txt", "a+")
                train_f.write(str(num_class)+"/"+str(image))
                train_f.write(" ")
                if num_class == "daisy":
                    train_f.write("0")
                elif num_class == "dandelion":
                    train_f.write("1")
                elif num_class == "rose":
                    train_f.write("2")
                elif num_class == "sunflower":
                    train_f.write("3")
                else:
                    train_f.write("4")
                train_f.write("\n")
            except FileExistsError as e:
                print(e)
                exit(1)
    print("\r[{}] split_rating [{}/{}]".format(num_class, index+1, num), end="")  # processing bar
    print()
print("       ")
print("       ")
print("划分完成") 

4. 生成classes.txt文件

因为就只有5个类别，我直接手写了

daisy 0
dandelion 1
rose 2
sunflower 3
tulip 4

5. 修改配置文件

将训练和验证集的数据路径进行修改，同样还有数据集标注列表，以及类别名文件路径。此外，还有评估方法修改为仅使用 top-1 分类错误率。

model = dict(
    type='ImageClassifier',
    backbone=dict(type='MobileNetV2', widen_factor=1.0),
    neck=dict(type='GlobalAveragePooling'),
    head=dict(
        type='LinearClsHead',
        num_classes=5,
        in_channels=1280,
        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
        topk=(1, 5)))
load_from = '/home/xilm/python_file/mmclassification-master/resources/mobilenet_v2_batch256_imagenet_20200708-3b2dc3af.pth'
dataset_type = 'CustomDataset'
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, backend='pillow'),
    dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
    dict(
        type='Normalize',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        to_rgb=True),
    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), backend='pillow'),
    dict(type='CenterCrop', crop_size=224),
    dict(
        type='Normalize',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        to_rgb=True),
    dict(type='ImageToTensor', keys=['img']),
    dict(type='Collect', keys=['img'])
]
data = dict(
    samples_per_gpu=32,
    workers_per_gpu=2,
    train=dict(
        type='CustomDataset',
        data_prefix=
        '/home/xilm/python_file/mmclassification-master/data/flower_dataset/train',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='RandomResizedCrop', size=224, backend='pillow'),
            dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='ToTensor', keys=['gt_label']),
            dict(type='Collect', keys=['img', 'gt_label'])
        ]),
    val=dict(
        type='CustomDataset',
        data_prefix=
        '/home/xilm/python_file/mmclassification-master/data/flower_dataset/val',
        ann_file=
        '/home/xilm/python_file/mmclassification-master/data/flower_dataset/val.txt',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='Resize', size=(256, -1), backend='pillow'),
            dict(type='CenterCrop', crop_size=224),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img'])
        ]),
    test=dict(
        type='CustomDataset',
        data_prefix=
        '/home/xilm/python_file/mmclassification-master/data/flower_dataset/val',
        ann_file=
        '/home/xilm/python_file/mmclassification-master/data/flower_dataset/val.txt',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='Resize', size=(256, -1), backend='pillow'),
            dict(type='CenterCrop', crop_size=224),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img'])
        ]))
evaluation = dict(
    interval=1, metric='accuracy', metric_options=dict(topk=(1, )))#此处将指标修改为top1
optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=4e-05)
optimizer_config = dict(grad_clip=None)
lr_config = dict(policy='step', gamma=0.5, step=1)
runner = dict(type='EpochBasedRunner', max_epochs=10)
checkpoint_config = dict(interval=5)
log_config = dict(interval=10, hooks=[dict(type='TextLoggerHook')])
dist_params = dict(backend='nccl')
log_level = 'INFO'
resume_from = None
workflow = [('train', 1)]
work_dir = './work_dirs/mobilenet-v2_flower'
gpu_ids = range(0, 1)

6. 训练

在终端执行mim train mmcls [配置文件名]

---

以上就是完成的整个流程了！