Google Summer of Code 2021 proposed ideas

Students interested to participate should check which of the following projects fits their interests and skills.

Τo communicate with the mentors and ask questions about the projects, students should subscribe to this list and post relevant questions. Please follow the Proposal Template

For practical information, students visit this page.

---

GSOC Projects implemented in 2017 GSOC Projects implemented in 2018 GSOC Projects implemented in 2019

Sastix-CMS: A general purpose JAVA CMS in Spring Boot

Brief Explanation

Sastix CMS is a Spring Boot application having several content management features. It exposes specific REST APIs to store, cache and lock content (resources). A resource can be a simple html file, an image, a video or any other file. Under common/api package you can find the API interfaces used in the project. It is a good place to start looking. Since a client-server architecture is used, these APIs have two kind of implementations: A client service and server service.

The client can be used/imported as a separated library/dependency in any project and will provide all the rest calls needed towards the underlying server for managing and retrieving the content through spcific locking and caching mechanisms. The relative code is under “client” project .The server side service implements the same interfaces and can have additional implementations needed for the core CMS platform. The relative code is under “server” project.

In order to be able to run locally the Sastix CMS server you should have at least jdk 1.8 installed and a MySQL instance running. Sastix CMS is using ORM and you can find the relative configuration under application.properties. You should define a schema in your database in order to be able to run the platform successfully. The properties file is your guide to do that.

Expected Results

Implement a Frontend environment for Administrators and Content Creators/Consumers to easily use, manage and monitor this application through simple and secure flows. Angular or React or Thymeleaf can be used. An admin should be able to manage any kind of resources and monitor the usage of the application (eg net/disk io, memory/cpu usage, disk allocation). A creator should be able to create update and delete resources. A consumer should have access to the published content and be able to effectively use it (eg see a video with chunked transfer encoding support). It will also be essential to extend the already implemented code, integrate new features and support the new Frontend service. Scalability is important too; how it can be deployed in a clustered mode using docker or kubernetes so it can support more than 1000 concurrent users (eg. offering a video streaming service and we have 1000 users at the same time watching videos by consuming URIs from the CMS)

Related repositories

https://github.com/sastix/cms

Knowledge Prerequisites

java, maven, js, html, css, mysql, docker

Mentors:

Ioannis Skitsas - Panagiotis Bochalis

Creation of a multi user audio first annotation tool

Brief Explanation

The annotation tools exist to empower anyone, even those without a data science or engineering background, to build the next generation of powerful, useful and impactful Artificial Intelligence applications by providing tooling and standards for the creation, collaboration, labeling and formatting of datasets.

While there is a variety of open source tools available for labelling datasets for different types, including text, images and audio, most of them lack the multi user workflows that enable more than one annotators to work on the process at the same time, and also produce more accurate results by providing consensus strategies and options to handle conflicts amongst the different annotator's data.

This proposal aims to fix these problems, aiming at the creation of a modern and intuitive system that will enable multi user annotation for audio primarily, and other metadata types as well. The system should support User accounts, User Queues, Queues management and actions as assign data on the fly. Options for consensus and conflict management, and output to multiple formats.

Expected Results

An analysis on existing open source tools for labeling data, pros and cons for each.

Design and implementation of a system that will enable setting annotations on audio types primarily and output results in multiple formats. This should allow the drawing of regions over the audio waveform containing events (eg as BAT, on related repositories)

Distinction of users (managers, annotators) with different permissions (add data, set categories or tags, set policies)

More advanced functionality may include setting time schedules, setting consensus policies (majority, minimum ratio), individual annotator's scoring, filtering of results by annotator.

Related repositories

BAT: https://github.com/BlaiMelendezCatalan/BAT

Label Studio: https://github.com/heartexlabs/label-studio

Knowledge Prerequisites

A modern frontend/backend combination for building modern web applications, as Django/React

Mentors:

Pantelis Vikatos, Markos Gogoulos

Extend deepbots to support Evolutionary Algorithms

Brief Explanation

Over the recent years evolutionary algorithms have achieved impressive results on various tasks such as artificial locomotion and autonomous robot navigation. Additionally, neuroevolution is a promising alternative approach to gradient descent optimization, such as Deep Reinforcement Learning, being applied to optimize deep neural networks. Deepbots is a simple framework which is used as "middleware" between the free and open-source Cyberbotics' Webots (https://cyberbotics.com/) robot simulator and Reinforcement Learning algorithms. Deepbots is a framework which follows the OpenAI gym (https://gym.openai.com/) interface logic in order to achieve an established way of implementing DRL for robotics’ and other scenarios. In this way researchers and students can easily test and benchmark their algorithms on either predefined or novel scenarios. Deepbots can be used as an educational framework for DRL in high-end robotics scenarios while aiming to establish a testbed for researchers. Thus far, deepbots is accompanied by common research scenarios (such as cartpole, mountain car and an obstacle avoidance example) and well established DRL algorithms (such as DDPG (https://arxiv.org/abs/1509.02971) and PPO (https://arxiv.org/abs/1707.06347)). Therefore, deepbots can be easily extended to support and enable the use of neuroevolutionary algorithms, since Webots is a suitable platform to create delicate tasks and apply such algorithms to solve them, while providing a unified tool for benchmarking and educational purposes. The ultimate goal of deepbots is to provide testbed scenarios for robotics that are easily accessible for researchers in order to demonstrate and promote Machine Intelligence capabilities. - Deepbots was first presented (https://link.springer.com/chapter/10.1007/978-3-030-49186-4_6) at the international conference on Artificial Intelligence Applications and Innovations (AIAI) conference in June 2020. Additionally, it was presented at FOSSCOMM 2020. According to pepy.tech (https://pepy.tech/project/deepbots) it has more than 5k downloads last year while it has extensively been used as an education tool. Furthermore, deepbots has been used by OpenDR researchers (https://opendr.eu/) in the development of a modular, open and non-proprietary toolkit for core robotic functionalities by harnessing deep learning to provide advanced perception and cognition capabilities. Finally, mentors are constantly into an open discussion with webots developers to improve certain functionalities of the Webots open-source robotic simulator. Manos Kirtas has participated twice in previous GSoCs as a student.

Expected Results

The expected results are two-fold: 1) Extend deepbots to support evolutionary algorithms on Webots 2) Implement a hybrid neuroevolutionary deep reinforcement learning algorithm

Related Repositories

https://github.com/aidudezzz/deepbots, https://github.com/aidudezzz/deepworlds

Knowledge Prerequisites

Good knowledge of python, familiarity with python deep learning frameworks (such as pytroch and tensorflow). Any prior contribution to deepbots framework will be appreciated.

Mentors:

Manos Kirtas (https://github.com/ManosMagnus), Konstantinos Tsampazis (https://github.com/tsampazk), Nikolaos Passalis (https://github.com/passalis)

Extend deepbots to support stable-baselines and implement gym-style default Reinforcement Learning environments

Brief Explanation

Deep Reinforcement Learning (DRL) is increasingly used to train robots to perform complex and delicate tasks, while the development of simulators contributes to the acceleration of research on DRL for robotics. Deepbots is a simple framework which is used as "middleware" between the free and open-source Cyberbotics' Webots (https://cyberbotics.com/) robot simulator and Reinforcement Learning algorithms. Deepbots is a framework which follows the OpenAI gym (https://gym.openai.com/) interface logic in order to achieve an established way of implementing DRL for robotics’ and other scenarios. In this way researchers and students can easily test and benchmark their algorithms on either predefined or novel scenarios. Deepbots can be used as an educational framework for DRL in high-end robotics scenarios while aiming to establish a testbed for researchers. Thus far, deepbots is accompanied by common research scenarios (such as cartpole, mountain car and an obstacle avoidance example) and well established DRL algorithms (such as DDPG (https://arxiv.org/abs/1509.02971) and PPO (https://arxiv.org/abs/1707.06347)). The ultimate goal of deepbots is to provide testbed scenarios for robotics that are easily accessible for researchers in order to demonstrate and promote Machine Intelligence capabilities. Deepbots strives to be a plug-and-play solution that supports a wide range of DRL agents, similarly to gym, harnessing the power of Webots. Stable-baselines is an established repository including DRL algorithm implementations providing a massive potential for benchmarking using deepbots.

Deepbots was first presented at the international conference on Artificial Intelligence Applications and Innovations (AIAI) conference in June 2020 (https://link.springer.com/chapter/10.1007/978-3-030-49186-4_6). Additionally, it was presented at FOSSCOMM 2020. According to pepy.tech (https://pepy.tech/project/deepbots) it has more than 5k downloads last year while it has extensively been used as an education tool. Furthermore, deepbots has been used by OpenDR researchers (https://opendr.eu/) in the development of a modular, open and non-proprietary toolkit for core robotic functionalities by harnessing deep learning to provide advanced perception and cognition capabilities. Finally, mentors are constantly into an open discussion with webots developers to improve certain functionalities of the Webots open-source robotic simulator. Manos Kirtas has participated twice in previous GSoCs as a student.

Expected Results

The expected results are three-fold: 1) Integrate stable-baselines into deepbots 2) Create default environments without RL agents, similar to gym 3) Create easy-to-use setup tools for robotic environments

Related Repositories

https://github.com/aidudezzz/deepbots, https://github.com/aidudezzz/deepworlds

Knowledge Prerequisites

Good knowledge of python, familiarity with python deep learning frameworks (such as pytroch and tensorflow). Any prior contribution to deepbots framework will be appreciated.

Mentors:

Manos Kirtas (https://github.com/ManosMagnus), Konstantinos Tsampazis (https://github.com/tsampazk), Nikolaos Passalis (https://github.com/passalis)

Development of a NodeRED compatible domain-specific language for Home Assistant

Brief Explanation

Home automation has become a standard for most houses and is a research and market trend with expected high growth. Among other smart home management solutions, Home Assistant (https://www.home-assistant.io/) is an open-source framework that clearly stands out. Home Assistant enables the integration of several commercial as well as custom-developed smart devices. Although Home Automation and other similar frameworks include tools for creating simple automation rules, they lack flexibility, i.e. more complex automation rules and flows are hard to create. Thus, advanced programming skills are needed if more sophisticated scenarios are envisaged. In this context we propose the development of a domain specific language (DSL) for home-assistant automation (HA-auto); HA-auto will enable citizen developers to easily create flows that manage complex everyday scenarios. To do so, HA-auto will provide integration to the NodeRed low-code platform (https://nodered.org/), this way making home automation flow programming even easier.

Related Repositories

https://github.com/home-assistant, https://github.com/node-red/node-red

Knowledge Prerequisites

Required: JavaScript, Python. Desired: NodeRed, Broker-based communication protocols, such as MQTT/AMQP, System Modeling / DSL knowledge

Mentors:

Konstantinos Panayiotou (klpanagi@issel.ee.auth.gr), Emmanouil Tsardoulias (etsardou@ece.auth.gr), Andreas Symeonidis (symeonid@ece.auth.gr)

Font Quality Improvement Project for Greek glyphs in Open Source Fonts

Brief Explanation

Many of the Open Source fonts (e.g., available at https://fonts.google.com), include glyphs for Greek language support. Some of these fonts are of poor quality and would benefit from a quality check and design improvements. This QA could vary from a simple character set report and design quality to the font's meta-data, metrics, OT features etc.

The aim of this project is to improve the quality of these fonts and automate the process of QA specifically to the Greek character set. This automation could be incorporated into the FontBakery tool. The exact set of fonts to be completed will be determined in discussions between the student and the mentor(s).

Expected Results

Reports on the quality of a number of Open Source fonts containing Greek characters.

Knowledge Prerequisites

Python (for FontBakery test framework) or other environment, font technologies, type design. Please note that this is a special project, where coding skills, in the traditional sense, might not be enough.

Mentors:

Alexios Zavras, Irene Vlachou Εmilios Τheofanous

Development of a DIY robot kit for educators

Brief Explanation

The aim of the project will be to develop all the designs, guidelines and sample code for a starter DIY robot kit that can be 3d-printed, assembled and operated using basic electronics and sensors. This is expected to create a low-cost alternative to commercial robot kits (e.g. Lego Mindstorms) that does not require expert staff in robotics, electronics or IoT programming (e.g. using Arduino/Raspberry kits). The ability to 3D-print everything and combine it with low-cost basic electronics and sensors will allow regional open technologies initiatives to provide schools with starter kits and a full 'Robotic 101' introductory course.

The kit that will be developed and opened must comprise 3D-designs for all the necessary parts of a modular robot that can be printed and assembled following the assemble guidelines. The target audience of the project can be educators (e.g. high school ICT teachers), with minimum expertise in robotics, electronics, and programming. So the print and assembly guidelines must be detailed and simple. In addition, the project must have a modular structure that allows educators to guide their students to the step-by-step development of the robot and to the implementation of simple navigation or sensing scenarios, that require basic programming skills.

Deliverables of the project, apart from the robot parts' designs, include a detailed list of the necessary electronics and sensors and the specifications for a Raspberry pi or similar single board computer (SBG).

Detailed assembly instructions, images, and videos from the assembly process are desirable.

The open source code that will be installed and run on the SBG and will allow controlling the robot through a simple programming interface, along with installation guidelines must be developed.

The robot will be operated either manually using a browser that wirelessly connects with the robot, or automatically by uploading robot control scripts through the same environment.

Some sample control scripts and robot programming scenarios will also be developed.

Expected Results

In the three months of the project it is expected to have the basic robot designs, the libraries for controling basic sensors (ultrasonic sensor, IR sensor, micro switches, optical odometer, servo/dc motor), the core operating software for controlling the robot and some simple robot programming assignments.

The three months plan of the project must define: a) The selection of electronics parts, SBC, and motors. b) The 3D designs of the printed parts of the robot. c) The libraries and software for controlling the robot. d) The development of assembly guidelines and the creation of demo scenarios for the class.

Related Repositories

https://hackaday.io/project/26007-versatile-educational-2wd-robot

Knowledge Prerequisites

Electronics, Robotics, Programming.

Mentors:

Hariton Polatoglou, Iraklis Varlamis, Theodoros Karounos

Plot ThanCad vector graphics to large plotters in Windows and Linux

Brief Explanation

ThanCad is a free 2 dimensional CAD which is command compatible with AutoCad. It is interoperable with AutoCad through standard dxf files. ThanCad runs in Windows and Linux, and has been experimentally ported to macOS-X. ThanCad has been used as teaching tool in National Technical University of Athens. Although ThanCad can plot to desktop printers, it lacks the capability to plot to large size (A1, A0) plotter devices, for example through the HPGL/2 gtraphics language.

Expected Results

ThanCad will be able to plot to large plotters and let the users select a plotter through the Windows printer interface.

Related repositories

https://sourceforge.net/projects/thancad/

Knowledge Prerequisites

Python

Mentors:

Thanasis Stamos, Dimitra Vassilaki

PackageInfo WebApp

Brief Explanation

A web application (site) to provide read-only access to a database of information about software packages.

For packages provided in popular Linux distributions like Debian, Ubuntu, etc., a number of characteristics is being collected (e.g., name, version, size) and stored into a database. The project would be to create a web interface to query the information in the database, in a read-only way. Users should be able to specify/search based on distribution, package name, version, etc. to drill down to a specific package and get information for it.

Expected Results

implementation of new web site/app

Related repositories

new project

Knowledge Prerequisites

front end: JavaScript (any framework) back end: Python (any framework) DB content: shell, basic SQL, optionally docker

Mentors:

Alexios Zavras

hashesDB

Brief Explanation

Tools to create, maintain, and search a database of hashes of files.

For each file, a number of distinct hashes is computed and stored into a database. Upon encountering a new file, the database can be searched to answer whether this file has been already seen. Given a huge archive of files (e.g. 6.5B files on Software Heritage), this would produce a simple method of quickly checking the provenance of files that were copied from elsewhere. Fuzzy hashes like ssdeep can extend the functionality to discover similar but not identical files.

Pointers for further info: https://www.softwareheritage.org/ https://multiformats.io/multihash/ https://ssdeep-project.github.io/ssdeep/index.html

Expected Results

implementation of new tools

Related repositories

new project

Knowledge Prerequisites

Python (optional C), basic SQL

Mentors:

Alexios Zavras

sch-webapps: aggregator for web-based educational material

Brief Explanation

Photodentro is the official repository of K12 educational material for Greece. It contains more than 15,000 learning objects or applications which are implemented in HTML, Adobe Flash, Java or other technologies, and it's used by thousands of students and teachers. Unfortunately, it has the following limitations:

• Flash based applications (2,760) are no longer viewable since Adobe deprecated Flash at the end of 2020.
• Applications can't be operated offline, for example when Internet connectivity is an issue.
• There's no way for users to create their own collections, which would ease navigation.
• Finally, teachers should be able to expose their offline collections on the school LAN so that students would use them via any browser.

A new sch-webapps application should be developed that would address these four limitations, hugely benefiting K12 education and possibly many others.

Expected Results

• Documentation for installing sch-webapps in Windows and Debian or Ubuntu Linux along with its dependencies: the nodejs environment and the Palemoon browser for viewing Flash applications.
• sch-webapps should start a web server at http://localhost:7100 on demand.
• LAN users would connect there and be able to browse the local collection of web applications, similar to https://ts.sch.gr/software.
• The local user (e.g. the teachers) would in addition see an administration page, from where they'd be able to add or remove applications.
• Creating a local collection should be as easy as pasting a list of URLs like "http://photodentro.edu.gr/lor/r/8521/10760". sch-webapps should then automatically locate and download application.zip and metadata.xml, decompress and parse them, and generate a main index page with all the locally installed applications.

- The main index.html should be viewable even via the file:/// protocol, without a web service.

Related repositories

sch-webapps will completely replace sch-webapp-launcher and all the 40+ Debian packages of web based applications under https://gitlab.com/ts.sch.gr.

Knowledge Prerequisites

Javascript, HTML, CSS, nodejs, git, experience with educational applications.

Application tasks

• Download a few web applications from http://photodentro.edu.gr/lor and decompress them in the same folder to manually create a "collection".
• Then design an example index web page for that collection, similar to https://ts.sch.gr/software.
• Finally, upload the result to a github/gitlab pages repository and share it with the mentors.

Mentors

Foteini Tsiami, Siahos Yiannis

Addition of chant glyphs in Open Source chant Font

Brief Explanation

Nowadays we have fonts to write Byzantine music (e.g. available at https://github.com/t-bullock/KA-Arvanitis), but we don’t have yet a font with all glyphs for all the chant notes which belong to the various periods of the evolution of Byzantine music. The aim of this project is to improve this situation and add the missing glyphs which we take mainly from manuscripts of the 9th-19th centuries.* Up to date, in Unicode we have only 246 signs. The exact number of new glyphs to be completed will be determined in discussions between the student and the mentors. This is not a typical programming project, but a complex task, requiring skills both in font technologies, paleography of Byzantine music and design. It’ s important that today we have about 200 years since the first printed chant book and 30 years since the development of the first chant font. For the project we will take as a point of departure the font of Gregorios Stathis.

• A second step would be a proposal in Unicode to add codepoints for the new signs.

This is not a typical programming project. If you have never designed fonts before, it is probably not for you.

Expected Results

Full support for all the neumes of Byzantine music in an Open Source font.

Knowledge Prerequisites

Font technologies, Paleography of Byzantine music, Type design. Please note that this is a special project, where coding, in the traditional sense, will not be enough.

Mentors:

Nikolaos Siklafidis: Phd Candidate in Theology at the Aristotle University of Thessaloniki, nikosikl@past.auth.gr
Maria Alexandru: Associate professor in Byzantine music at the Aristotle University of Thessaloniki, malexand@mus.auth.gr

MediaCMS - modern, fully featured video and media CMS

Brief Explanation

MediaCMS is a content management system developed to meet the needs of modern web platforms for viewing and sharing media. It is built with Django (backend) and React (frontend) and features a REST API. It includes a responsive design with two themes, advanced categorization options (tags, categories, playlists), full text search, an enhanced video player, subtitles, plus streaming through HLS.

The suggested work for GSoC 2021 includes working on any of these two directions: extend the software and provide Machine Learning powered functionality.

Extend the software ideas:

• software internalization
• improvements to system notifications
• admin dashboards
• provide a command line client that utilizes the REST API
• block storage support
• implement remote encoding services

Machine Learning powered functionality:

• improve the related media algorithm, using collaborative filtering or other approaches. This can make the related media much more interesting
• provide auto-tagging for media, based on existing but also new tags. New tags could be linked with user provided images. This will allow for efficient auto-tagging for large numbers of images and videos
• automatic titles and descriptions based on the metadata extracted
• video thumbnail extraction, that is representative of the video.

Expected Results

All functionality created as part of this work will be integrated to the software through multiple Pull Requests. This work can have a massive impact over the software adoption.

Related repositories

https://github.com/mediacms-io/mediacms
https://mediacms.io
https://demo.mediacms.io

Knowledge Prerequisites

Python/Django for backend related work
Javascript/React for frontend related work
Machine Learning or Deep Learning frameworks knowledge, for the ML related suggestions

Mentors:

Markos Gogoulos