Syntetich_complete

This dataset refers to MINI-RGBD (7 Gb) available online. We perform on MINI-RGBD some modifications to obtain the Syntetcih_complete. The modifications are describerd below.

1. pz[id unique]

We have named each directory with pz[id unique]. [id unique] is an integer with which we uniquely distinguish each infant.
Each of the directory refers to a specific directory in the MINI-RGBD dataset. We described it in the README file of each pz[id unique] directory.

Each of pz[id unique] directory contains:

• 1000 16-bit images, already in the MINI-RGBD and viewable by downloading the Fiji program
• 1000 8-bit images, not present in the MINI-RGBD. We obtained them performing a conversion in uint8 from 16-bit images

The images have not been pushed to GitHub, but you can download it and apply the trasformation describe in section 4.1.

2. annotations

In the annotations directory, for each pz[id unique] there is result_pz[id unique].csv file containing the images keypoints annotations. In particular, each keypoint annotation are written as follows: [coordinata_x],[coordinata_y].

The annotations files have been pushed to GitHub. If you want to know more details on annotations transformations respect to MINI-RGBD please refer to section 4.2.

3. tfrecord

In the tfrecord directory there are the sub-folders related to dataset's configurations.
In each of configuration directory there are:

• the train, valid e test sets files in .tfrecord format
• sets_config file in pikle (.pkl) format that containg the infos on configuration.
• dic_history.pkl dictonary in which we have for each sets the pairs formed and the related position in the tfrecord file

For the Syntetich complete, we created only one configuration called: negative_no_flip_camp_5_keypoints_2_mask_1. The configuration and related files are created by the ../Dataset_configuration_generator.py and are used by the framework.

4. Trasformations

4.1 Images

Each of the images in the folders pz[id unique] has the following changes compared with the MINI-RGBD:

• 90 degree clockwise rotation of images

4.2 Annotations

In order to cope with images transformations, we also rotated of 90 degree clockwise the keypoints coordinates.
In addition, we have reduced the original MINI-RGBD annotations:

• Simplification of keypoints (from 24 keypoints to 14 keypoints)
• New positional IDs given to keypoints

4.3 Example

Below we post an example of image and reletive annotations of Syntetch_complete dataset: