1. 自定义一个pipeline

新建一个mmdet/datasets/pipelines/middleway.py

pipeline的运作逻辑从call(self, results)的results入，从return的results。

  1. # Copyright (c) OpenMMLab. All rights reserved.
  2. from ..builder import PIPELINES
  4. try:
  5.     from panopticapi.utils import rgb2id
  6. except ImportError:
  7.     rgb2id = None
  10. @PIPELINES.register_module()
  11. class MiddleDebug:
  12.     """Load an image from file.
  14.     Required keys are "img_prefix" and "img_info" (a dict that must contain the
  15.     key "filename"). Added or updated keys are "filename", "img", "img_shape",
  16.     "ori_shape" (same as `img_shape`), "pad_shape" (same as `img_shape`),
  17.     "scale_factor" (1.0) and "img_norm_cfg" (means=0 and stds=1).
  19.     Args:
  20.         to_float32 (bool): Whether to convert the loaded image to a float32
  21.             numpy array. If set to False, the loaded image is an uint8 array.
  22.             Defaults to False.
  23.         color_type (str): The flag argument for :func:`mmcv.imfrombytes`.
  24.             Defaults to 'color'.
  25.         file_client_args (dict): Arguments to instantiate a FileClient.
  26.             See :class:`mmcv.fileio.FileClient` for details.
  27.             Defaults to ``dict(backend='disk')``.
  28.     """
  30.     def __init__(self, mode=1):
  31.         self.mode = 1
  33.     def __call__(self, results):
  34.         """Call functions to load image and get image meta information.
  36.         Args:
  37.             results (dict): Result dict from :obj:`mmdet.CustomDataset`.
  39.         Returns:
  40.             dict: The dict contains loaded image and meta information.
  41.         """
  42.         print(results)
  43.         return results

2. 使用自定义的pipeline

  1.         pipeline=[
  2.             dict(type='LoadImageFromFile'),
  3.             dict(type='MiddleDebug'),
  4.             dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
  5.             dict(type='MiddleDebug'),
  6.             dict(type='Resize', img_scale=(1333, 800), keep_ratio=True),
  7.             dict(type='RandomFlip', flip_ratio=0.5),
  8.             dict(type='MiddleDebug'),
  9.             dict(
  10.                 type='Normalize',
  11.                 mean=[123.675, 116.28, 103.53],
  12.                 std=[58.395, 57.12, 57.375],
  13.                 to_rgb=True),
  14.             dict(type='Pad', size_divisor=32),
  15.             dict(type='DefaultFormatBundle'),
  16.             dict(
  17.                 type='Collect',
  18.                 keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks'])
  19.         ]),