Showmax is working closely with the Faculty of Information Technology at the Czech Technical University. We collaborate under the umbrella of ShowmaxLab. It involves Showmax engineers visiting the faculty regularly and working directly with students. Lab has a dedicated workspace on the 13th floor.
ShowmaxLab is a program that allows students to get closer to the real world, get paid for doing interesting stuff, and (usually) gain academic credit while they do it. Students can select from a range of available projects (check below) across the different Showmax development teams to work on. These include everything from AI and machine learning (ML), highly available and distributed Internet systems, to monitoring, testing and API documentation.
Machine Learning
Machine learning is becoming an important tool for solving some of the day-to-day engineering challenges. ML has an indispensable place in the efforts to better understand our users, content, and how our service can be better. This also includes content recommendations - one of the most-important parts of the user experience, and the perfect testing ground for advanced ML applications.
| Project title | Researcher(s) | Project Summary |
|---|---|---|
| Action detection in videos | Ondřej Bíža | The aim of this project is to recognize what people are doing in videos. As recognizing human actions is an incredibly complex, we use very deep convolutional networks trained on large datasets. |
| Detecting scene locations | Lukáš Lopatovský | In this project, we tested different techniques for detecting where movie scenes take place. The final system used convolutional neural networks trained on the Imagenet dataset and fine-tuned on our custom dataset. |
Open Projects
These projects are typically eligible for being submitted as bachelor’s or even master’s theses. We will provide you with technical supervisor (odborný garant práce) and/or reviewer (oponent práce). In order to see the details and get involved, you need to be a CTU (ČVUT) student. If you don’t have an account, will also need to register yourself at SSP to get enrolled to the project or please, drop us an email at students@showmax.engineering.
| Project title | Estimated time complexity | Reward level | Enroll | |
|---|---|---|---|---|
| Adaptive Video Buffer Management | 120 hours | 💰💰💰💰 |
|
|
| Client-side video players employ adaptive bitrate (ABR) algorithms to optimize user quality of experience (QoE). Despite the abundance of recently proposed schemes, state-of-the-art ABR algorithms suffer from a key limitation: they use fixed control rules based on simplified or inaccurate models of the deployment environment. As a result, existing schemes inevitably fail to achieve optimal performance across a broad set of network conditions and QoE objectives. | ||||
| Annotation of movie scripts with video positions | 100 hours | 💰💰 |
|
|
| Movie subtitles carry a lot of semantic information that can be used in machine learning models. However, they reflect only on what is spoken aloud and occasionally on significant sounds / noises. On the other hand, there are full movie scripts available from internet databases that describe everything with a very high level of detail. The subtitles are annotated with movie positions while the script lines are not. However, by matching text from the former data set with the latter one, the positions of script line can be found to a varying degree of precision. | ||||
| Automatic subtitles synchronisation | 60 hours | 💰 |
|
|
| We have a lot of subtitles for movies and series. Some of the subtitles are not aligned with the video properly. We would like to detect if the subtitles are matching the movie and align them automatically in case they are incorrectly synchronised. We would like to fix the cases when the whole subtitles file is shifted compared to the movie and we would also like to fine tune timing of each subtitle line. | ||||
| Better targeting of win-back campaigns | 60 hours | 💰💰 |
|
|
| We have sent out 1.2 million vouchers to some of our customers. A portion of the vouchers have not been used. We would like to investigate what are the similarities between the customers who used the offer and those who didn’t. | ||||
| Compare "Brotli" and "gzip" compression algorithm on nginx HTTP server | 40 hours | 💰 |
|
|
| Currently we use “gzip” compression algorithm in Showmax.com for making data amount of HTTP results smaller. For that purpose we use simply native “gzip” module in nginx HTTP server. | ||||
| Deep learning features and similarity of movies based on their subtitles | 30 hours | 💰 |
|
|
| Subtitles of movies are usually very informative. Deep learning methods can capture higher level meaning … | ||||
| Highly available and scalable Prometheus | 40 hours | 💰 |
|
|
| Prometheus is a powerful and very useful monitoring system and a time series database and we just love it. Help let it grow together with our platform by finding a way to make it highly available and scalable. | ||||
| Highly available Docker registry | 40 hours | 💰 |
|
|
| Modern services built upon micro-service architecture are usually dependent on containers (Docker). When the Docker registry is not highly available, it poses a serious threat to the whole platform as it is a SPOF (Single Point Of Failure). Find a solution to the problem and make Docker registry highly available. | ||||
| Neural Adaptive Content-aware Internet Video Delivery | 150 hours | 💰💰💰 |
|
|
| Neural Adaptive Content-aware Internet Video Delivery seems to be very promising technique to improve quality of experience for users with slow internet connection and powerful hardware. This method us using pre-trained deep neural network to improve the quality of the video from lower resolution into higher resolution using computing power of the client hardware. We want to investigate if similar principles could be used in Showmax environment. | ||||
| Prediction of time when to send message to user | 60 hours | 💰💰 |
|
|
| We are sending various messages to the users. The most important message is reminding users to prolong the subscription. In this kind of messages it is crucial to get the timing of the message right. We would like to put machine learning in place to predict the ideal time when to send the message to the user. | ||||
| Research of full-disk encryption | 60 hours | 💰 |
|
|
| With the goal of ever improving the security of our platform and securing the personal data of our customers, full disk encryption is a necessity. Help us improve our platform by investigating the ways to do full disk encryption on Debian powered machines. | ||||
| Viewing funnel analysis and clustering | 60 hours | 💰💰 |
|
|
| Our customers have different viewing habits. They have also came for different content. Goal of this work is to understand what is the viewing funnel and what are the similar groups of users. For example we envision that there could be group of people who came to watch just one/couple blockbusters and then left. Others who came because Game of Thrones but then have got hooked on our original production. | ||||
Learn More
Drop us an email at lab@showmax.com to schedule a meeting. We go to the lab every second Monday mornings.