mmdetection3d-train_pipeline

train_pipeline = [
    dict(type='LoadImageFromFileMono3D', backend_args=backend_args),
    dict(
        type='LoadAnnotations3D',
        with_bbox=True,
        with_label=True,
        with_attr_label=False,
        with_bbox_3d=True,
        with_label_3d=True,
        with_bbox_depth=True),
    dict(type='Resize', scale=(1242, 375), keep_ratio=True),
    dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
    dict(
        type='Pack3DDetInputs',
        keys=[
            'img', 'gt_bboxes', 'gt_bboxes_labels', 'gt_bboxes_3d',
            'gt_labels_3d', 'centers_2d', 'depths'
        ]),
]
1、LoadImageFromFileMono3D

加载当前图像

2、LoadAnnotations3D

with_bboxes:数据集标注中给出的目标物体对应的边界框

with_labels:数据集标注中给出的目标物体对应的标签

with_bbox_3d:数据集标注中给出的目标物体对应的3d边框

with_label_3d:数据集标注中给出的目标物体对应的3d标签

with_bbox_depth:数据集标注中给出的目标物体对应的深度

3、Resize

scale=(1242, 375), keep_ratio=True

scale缩放之后图像的最终尺度， keep_ratio保持宽高比

4、RandomFlip

flip_ratio_bev_horizontal:翻转概率

horizontal:翻转方向

5、Pack3DDetInputs

输出指标:keys=['img', 'gt_bboxes', 'gt_bboxes_labels', 'gt_bboxes_3d',

'gt_labels_3d', 'centers_2d', 'depths']