add augmentor support

[fumihwh]

No description provided.

[leapmindadmin]

Can one of the admins verify this patch?

[hadusam]

run test

[hadusam]

run test

[hadusam]

run dlk test

[hadusam]

run test

[hadusam]

run test

[lm-jira]

Library for FPGA can be generated from this code without any problem.
The library can also be implemented without problem

[hadusam]

run test

[hadusam]

ok to test

[hadusam]

@lm-jira Thank you for your checking!
@ruimashita Please review this PR. 👍

[ruimashita]

May not need this line. I will fix it.

[ruimashita]

Didn't use this var CHOOSE_TOP_TWO_AU. need to delete this line.

[ruimashita]

The HTML format of augmentation page on document site is not readbility especially params section.

Should be like below:

[ruimashita]

@fumihwh oh❗️ I thought of doing to edit, but you did it. Thanks👍👍👍

[ruimashita]

data_augmentor page is empty content
https://deploy-preview-2--blueoil.netlify.com/reference/data_augmentor.html

[fumihwh]

No idea about that. It works good locally.

[ruimashita]

@fumihwh OK. I will do the document things 👍

[ruimashita]

@lm-jira
When I select RandomErasingForDetection and ./blueoil.sh train the error comes.

yaml.scanner.ScannerError: mapping values are not allowed here
  in "/home/blueoil/config/aa.yml", line 34, column 29
ERROR: Training exited with a non-zero status
(.venv) #( 11/02/18@ 5:12PM )( wakisaka@leapmind-se

That's way the generated yaml has broken format.

# supported task types are 'classification', 'object_detection' and 'semantic_segmentation'.
task_type: classification

network_name: LmnetV1Quantize

dataset:
  format: Caltech101
  train_path:
  test_path:

trainer:
  batch_size: 10
  epochs: 100

common:
  image_size:
    - 128  # height
    - 128  # width

  # set pretrain model name. currently, this feature is not supported, always ignored.
  pretrain_model: false

  # enable dataset prefetch, set false if weired problem happens
  dataset_prefetch: true

  data_augmentation:
    - RandomErasingForDetection:
      - probability: 0.5- sl: 0.02- sh: 0.2- r1: 0.3- content_type: mean- mean: [125, 122, 114]- i_a: True- o_a: True

What kind of manually test, did you do ??

[lm-jira]

I trained the model without editing anything in the config file.