augmentations.md

Augmentations and Preprocesses

EOD supports several data augmentations and preprocesses, including various augmentations such as Flip, Resize, StitchExpand, ImageCrop, etc. and preprocesses such as normalization, to_tenser, pad, etc. Details are as follows.

EOD imports augmentations directly from config settings:

Flip:

flip: &flip   
 type: flip
 kwargs:
   flip_p: 0.5

Resize:

resize: &train_resize
 type: keep_ar_resize
 kwargs:
   scales: [640, 672, 704, 736, 768, 800]
   max_size: 1333
   separate_wh: True

StitchExpand:

expand: &stitch_expand
  type: stitch_expand
  kwargs:
    expand_ratios: 2.0
    expand_prob: 0.5

ImageCrop:

crop: &crop
  type: crop
  kwargs:
    means: [123.675, 116.280, 103.530]
    scale: 1024
    crop_prob: 0.5

Normalization:

normalize: &normalize
 type: normalize
 kwargs:
   mean: [0.485, 0.456, 0.406] # ImageNet pretrained statics
   std: [0.229, 0.224, 0.225]

ToTensor:

to_tensor: &to_tensor
  type: to_tensor

BatchPad: be usually added into dataloader config

dataloader:
    type: base
    kwargs:
      num_workers: 4
      alignment: 32
      pad_value: 0
      pad_type: batch_pad

• All augmentations need to be added into dataset.kwargs.transformer in order as follows:

dataset:
  type: coco
  kwargs:
    meta_file: coco/annotations/instances_train2017.json
    image_reader:
      type: fs_opencv
      kwargs:
        image_dir: coco/train2017
        color_mode: RGB
    transformer: [*flip, *train_resize, *to_tensor, *normalize]   # add here in order