data

Here you have the possibility to define the datasets.

1. Dataset directory structure

Note: As an example, consider the structure in ./Syntetich_complete dataset.
Each directory is named with the following pattern: [type][underscore][note]. Each directory be structured as follow:

├───[type][underscore][note]
	├───annotations
	├───pz[id unique]
	├───     .
	├───     .
	├───pz[id unique]
	└───tfrecord
		├───configuration_1
		├───configuration_2
		├───     .
		├───     .
		└───configuration_n
			├───[type]_train.tfrecord
			├───[type]_valid.tfrecord
			├───[type]_test.tfrecord
			└───set_configs.pkl
				

The content of each folder is described below:

• pz[id unique]: these directories contain the images about a specific infant identified by the id unique
• annotations: this directory will contain the related annotations files for each of the pz[id unique] directory
• tfrecord: this directory contain the dataset's configurations. The dataset configuration is the source data for the framework. Each configuration is rapresented by a sub-folder.
  In each of them we have:
  • train/valid/test sets in .tfrecord format. These are the set to use during the training and evaluation phase
  • sets_configs.pkl that describe the carateristics about the configuration (radius_key=2, flip=True, etc..)
  • dic_history.pkl dictonary in which we have for each sets the pair formed and the related positin in tfrecord file
  These files are created by the ./Dataset_configuration_generator.py script described in section 2.

2. Create dataset configuration with Dataset_configuration_generator.py

This script creates the dataset configuration that can be used by the framework. To execute it, it is necessary to set the information on the configuration part. After that you can run the script:

python Dataset_configuration_generator.py