Full project source

Since the main project is made up of multiple submodules, containing even more submodules (such as LVGL, micropython, etc.), it is not always possible to get the full source and state of the project from GitHubs download functionality alone.

This release contains the full source of the project, with all submodules initialized, excluding the .git folder.

It is the base of reference for everything mentioned and described in the paper.

Be aware, due to the amount of submodules in the repositories such as LVGL and micropython, the archived contents are actually quite big for only being source code material.

An additional archive with suffix _minimal is added, which excludes all of those external dependencies and only contains code written by the Author. Be aware that the lvgl_ui_generator and lvgl_ui_generator_v2 projects actually require those dependencies to compile their binaries.

Time: 16.05.2024 20:18

Final paper results

In this release, the data of the final thesis results is uploaded.

It contains three datasets of the two generator modes:

• 04da75baa7084aee83f3b31602c408c2_random
• 5c8e248bdcc24ee2bb57760f0961c690_design_original
• 2fe5e8b06a6249fcb9360c9645e0f7a6_design_corrected

The design corrected version was manually adjusted to remove faulty/distorted images due to an issue in the generator during encoding. To fill up removed images, additionally generated designs from other generations were added in.

The release also contains the two training runs on these datasets:

• train371 was trained on the random dataset
• train373 was trained on the design_corrected dataset

ClearML exported task results

This release contains the fully exported tasks of various projects created in ClearML during paper development.

• lvgl_ui_detector is a folder containing the main tasks of the project, where specific experiments took place to either improve the pipelines or to test out the generator in various forms.
• YOLOv8 is a folder containing tasks related to tuning hyperparameters for either the random or design dataset
• HPO is a folder containing tasks related to custom hyperparameter optimization using ClearMLs optimizer tasks, alongside the integration of Optuna.

These tasks are generally provided as an export of all gathered experiments which took place during paper development, in case any are referenced.
Be aware, that these archives also contain aborted, failed or debugging tasks from the paper development.

Each folder contains tasks in the form of subfolders, where the Task ID is the name of the folder.
A full list of experiments with additional information for each ID can be viewed in the experiments.json file inside each project folder mentioned above.

ClearML exported datasets

This release contains the fully exported datasets and their corresponding tasks of various projects created in ClearML during paper development.

These tasks are generally provided as an export of all created datasets which took place during paper development, in case any are referenced.

Be aware, that this archive also contains aborted, failed or debugging datasets/tasks from the dataset development.

The archive contains datasets in the form of subfolders, where the Dataset ID is the name of the folder.
A full list of datasets with additional information for each ID can be viewed in the datasets.json file inside the archive.

Local model training & tuning runs

This release contains the training runs that were stored locally on my machine.

It's the full contents of the 'runs' directory.

Time: 16.05.2024 19:40