论文 - MMDetection: Open MMLab Detection Toolbox and Benchmark - 2019

Github - open-mmlab/mmdetection

mmdetection - Data preparation pipeline

更新日期：2019.08.23

mmdetection 的数据准备管道(data preparaion pipeline)和数据集的处理过程是进行了分解的. 一般情况下，数据集定义了如何处理标注数据；数据准备管道定义了数据字典(data dict)准备的全部步骤. 数据管道包含一系列序列化的操作，每个操作都采用 dict 作为输入，并同样输出一个 dict.

如图，蓝色块表示管道操作. 随着管道的进行，每个操作子会添加新的 keys(标记为绿色) 到输出 dict 中，或者更新已有的 keys(标记为橙色).

这些数据操作被归类为：数据加载，预处理，格式化，测试数据增强.

例如，Faster R-CNN 的数据管道示例：

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='LoadAnnotations', with_bbox=True),
    dict(type='Resize', img_scale=(1333, 800), keep_ratio=True),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1333, 800),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='Pad', size_divisor=32),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img']),
        ])
]

[1] - 数据加载(Data loading)

LoadImageFromFile
  • add: img, img_shape, ori_shape
LoadAnnotations
  • add: gt_bboxes, gt_bboxes_ignore, gt_labels, gt_masks, gt_semantic_seg, bbox_fields, mask_fields
LoadProposals
  • add: proposals

[2] - 预处理(Pre-processing)

Resize
  • add: scale, scale_idx, pad_shape, scale_factor, keep_ratio
  • update: img, img_shape, bbox_fields, mask_fields
RandomFlip
  • add: flip
  • update: img, bbox_fields, mask_fields
Pad
  • add: pad_fixed_size, pad_size_divisor
  • update: img, pad_shape, *mask_fields
RandomCrop
  • update: img, pad_shape, gt_bboxes, gt_labels, gt_masks, *bbox_fields
Normalize
  • add: img_norm_cfg
  • update: img
SegResizeFlipPadRescale
  • update: gt_semantic_seg
PhotoMetricDistortion
  • update: img
Expand
  • update: img, gt_bboxes
MinIoURandomCrop
  • update: img, gt_bboxes, gt_labels
Corrupt
  • update: img

[3] - 格式化(Formatting)

ToTensor
  • update: specified by keys.
ImageToTensor
  • update: specified by keys.
Transpose
  • update: specified by keys.
ToDataContainer
  • update: specified by fields.
DefaultFormatBundle
  • update: img, proposals, gt_bboxes, gt_bboxes_ignore, gt_labels, gt_masks, gt_semantic_seg
Collect
  • add: img_meta (the keys of img_meta is specified by meta_keys)
  • remove: all other keys except for those specified by keys

[4] - 测试数据增强(Test time augmentation)

MultiScaleFlipAug
Last modification：September 3, 2019
© Allow specification reprint
If you think my article is useful to you, please feel free to appreciate