8.1.1.1.1.1. blueoil.cmd.build_tfds

8.1.1.1.1.1.1. Module Contents

8.1.1.1.1.1.1.1. Functions

_get_tfds_settings(config_file)

_get_tfds_builder_class(dataset_class)

_copy_directory_recursively(src, dst)

run(config_file, overwrite)

Build custom TFDS datasets from config file

main(config_file, overwrite)

A script to build custom TFDS datasets

blueoil.cmd.build_tfds._get_tfds_settings(config_file)
blueoil.cmd.build_tfds._get_tfds_builder_class(dataset_class)
blueoil.cmd.build_tfds._copy_directory_recursively(src, dst)
blueoil.cmd.build_tfds.run(config_file, overwrite)

Build custom TFDS datasets from config file

blueoil.cmd.build_tfds.main(config_file, overwrite)

A script to build custom TFDS datasets

 This script can build custom TFDS datasets from existing dataset classes. You can use training config files to specify which dataset class to be used as data source. The following settings are required in the config file.

DATASET_CLASS = <dataset class> DATASET.TFDS_KWARGS = {

“name”: “<a dataset name to be generated>”, “data_dir”: “<a directory path to output generated dataset>”, “image_size”: <image size array like [128, 128]>,

 Note: Images will be resized into the specified size when TFRecords are loaded.

Images stored in TFRecords still have original size.

 If you have a training config file with the settings above, you can execute this script and the training script with the same config file. Then the generated TFDS dataset will be used for training.

python blueoil/cmd/build_tfds.py -c common_config_file.py python blueoil/cmd/train.py -c common_config_file.py ```