train_dataloader

train_dataloader = dict(
    batch_size=2,
    num_workers=2,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file='kitti_infos_train.pkl',
        data_prefix=dict(img='training/image_2'),
        pipeline=train_pipeline,
        modality=input_modality,
        load_type='fov_image_based',
        test_mode=False,
        metainfo=metainfo,
        # we use box_type_3d='Camera' in monocular 3d
        # detection task
        box_type_3d='Camera',
        backend_args=backend_args))
val_dataloader = dict(
    batch_size=1,
    num_workers=2,
    persistent_workers=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        data_prefix=dict(img='training/image_2'),
        ann_file='kitti_infos_val.pkl',
        pipeline=test_pipeline,
        modality=input_modality,
        load_type='fov_image_based',
        metainfo=metainfo,
        test_mode=True,
        box_type_3d='Camera',
        backend_args=backend_args))
test_dataloader = val_dataloader
 

1、batch_size=2 批量大小

2、num_workers=2多线程

3、persistent_workers=True

4、sampler=dict(type='DefaultSampler', shuffle=True)随机打乱

5、type=dataset_type数据类型

6、data_root=data_root数据目录

7、ann_file='kitti_infos_train.pkl'注释文件

8、data_prefix=dict(img='training/image_2')

9、pipeline=train_pipeline训练流程

10、load_type='fov_image_based'

11、test_mode=False测试模式

12、metainfo=metainfo数据类别

13、box_type_3d='Camera'相机

14、backend_args=backend_args