mmdetection之dataset概述

dataset类在tools/train文件中实例化，然后传入到train_detector()函数中用于构建dataloader

datasets = [build_dataset(cfg.data.train)]  # 实例化dataset
if len(cfg.workflow) == 2:  # 有验证流程的话则实例化验证数据集类
    val_dataset = copy.deepcopy(cfg.data.val)
    val_dataset.pipeline = cfg.data.train.pipeline
    datasets.append(build_dataset(val_dataset))

实例化dataset时用到的参数在_base_/dataset/文件夹中进行配置

data = dict(
    # samples_per_gpu=2,
    # workers_per_gpu=2,
    samples_per_gpu=1, #batch_sizes
    workers_per_gpu=0, #number_works
    train=dict( # 实例化train_dataset时用到的参数
        type=dataset_type,
        ann_file='E:\新建文件夹\datesets\SIXray\ImageSet/ann.txt',
        img_prefix=data_root,
        pipeline=train_pipeline
    ),
    val=dict( # 实例化val_dataset时用到的参数
        type=dataset_type,
        ann_file='E:\新建文件夹\datesets\SIXray\ImageSet/ann.txt',
        img_prefix=data_root,
        pipeline=test_pipeline),
    test=dict( # 实例化test_dataset时用到的参数
        type=dataset_type,
        ann_file='E:\新建文件夹\datesets\SIXray\ImageSet/ann.txt',
        img_prefix=data_root,
        pipeline=test_pipeline))

base/default_runtime.py文件中的workflow用于控制实例化的dataset的类型

workflow = [('train', 1),('val', 1)]