Expert Data Annotation for Driverless Experience at Its Best

1787 HOURS SPENT ON LABELING TASKS
2244 IMAGES ANNOTATED
25951 BOXES ADDED
171572 KEYPOINTS ADDED

table of contents:

Let Go of Control and Get Professional Help

HOW MUCH WOULD YOUR ANNOTATTION COST?

High Quality + Certified Security

Get Your Quote

The rapid transfer of knowledge from one person to another has allowed humanity to develop fast and create technologies that no single generation could possibly create on its own. The specialization of labor, in this case, became a necessity.

Today, it’s best to let the experts do their job – a crucial principle in IT. It is extremely resource- and cost-intensive to gather and train a team of data specialists. To lighten the load, it’s essential to focus on the core details while delegating separate tasks of your project to a team of dedicated professionals.

The culture of finding expert partners has become so vital for modern entrepreneurship. This is why our partner from Germany, Elefant Racing, was able to achieve astounding, unpredicted success in their project. The combination of the Elefant team’s strong vision and focus and data annotation expertise from Label Your Data has made it possible for the first autonomous race car of Elefant Racing to exist!

The nearly two-year successful partnership with Elefant became a telling case worth learning from, so LET’S GET STARTED! cone

Background: Elefant Racing, 2004-2022

Elefant Racing started back in March 2004. It was a student organization consisting of passionate young engineers and business leaders. Their goal was to build the best racing cars that could win the famed Formula Student events taking place all over the world.

What’s Formula Student?

Formula Student (FS) is the established event meant to foster young talents in business and engineering. The idea is simple: build a race car from scratch and compete with other student teams.

What do you need to participate in FS?

team

partners & sponsors

race car

Only the combination of the best components, planning, and construction efforts results in a car that can not only compete, but win a Formula student event!

car-desc-img

Initially, the founder and CTO of Elefant Racing, Tobias Friedrich was fascinated by the possibilities that driverless vehicles offered. However, building a driverless race car at the time was still an expensive and somewhat extravagant project. It was less known than other types of racing cars (a traditional one with a combustion engine fire-img or a comparative novelty, an electric car bolt-img ).

For this reason, as well as due to the absence of engineers and developers who were able to take on the task, it was decided to stick to the more traditional vehicle types.

Group 48 4th/5th place FS Czech Republic 16th/39th place FSG 11th/21st place FSN FR19 Loki 2019 13th place FSG FR5 Katana 2006 17th place FSG 2007 FR7 Aquila 2007 41st place FSG FR8 Chromo 2008 60th place FSG FR9 Fantom 2009 61st place FSG FR10 Phoenix 2010 16th place FSG FR11 Cocoon 2011 13th place FSG FR12 Vilja 2012 12th place FSG 13th place FSA FR13 Cyrano 2013 25th place FSG 10th place FSA FR14 Norsu 2014 1st place FS Czech Republic 7th place FSG/FSA FR15 Valkyrie 2015 27th place FSG 18th place FSA FR16 Parsifal 2016 23th place FSG/FSH 17th place FSN FR17 Wotan 2017 did not participate because of damage FR18 Siegfried 2018 21st place FSG 5th place FSE FR20 Ragnarök 2020 FSG – Formula Student Germany FSA – Formula Student Austria FSH – Formula Student Hungary FSN – Formula Student Netherlands FSE – Formula Student East, former FSH *

In 2020, the decision was made to start building a driverless race car. cone

Problem: Lack of Essential Expertise

There are many things a project like a driverless vehicle requires. From components and materials to manpower with sufficient level of expertise, FR20 Ragnarök was undoubtedly the most ambitious race car ever built by the Elefant team.

And it was tricky: the Elefant team, despite their rich experience in building racing cars, have never before built a driverless model. They didn't know where to start or what they'll need to get the job done. As they started designing a model to support the driverless mode, it became apparent that training the algorithm would not be easy. It required massive amounts of data to train properly.

Autonomous driving is among the most hazardous dangerous and thus rigorously regulated machine learning areas. In Formula Student, there is a whole set of disciplines designed to test the security of a driverless vehicle.

Soon the team understood that not only the amount of the data but its quality mattered, too. Building the best race car and equipping it with a sophisticated driverless algorithm would be for naught if the data wasn’t properly annotated for the training.

However, the team didn’t have enough experience and even less human resources to dedicate to this time-consuming task. For this reason, their search for an annotation partner started.

The Elefant Racing team had a few requirements for their potential partners and sponsors:

  • Path Expertise was obviously the top priority as the quality of annotation had to be top-notch
  • Oval A knack for innovation the team was looking for a company that shared their burning passion for new technologies
  • Path Copy 3 Given the competitive nature of Formula Student, the Elefant team was interested in the security of all processes
  • Path Copy 5 The possibility to support the project during the whole year, from September 2020 to August 2021

And thus the quest began. The team searched through the Internet in hopes to find the perfect partner. Their effort paid off when they stumbled upon Label Your Data, a comparatively small but ambitious Ukrainian start-up with focus on high-quality and secure data annotation.

car-experience

Why Do You Need Data Annotation?

Data annotation is a crucial step of building and training a machine learning algorithm. When you have a vast amount of data on your hands, you cannot simply feed it into your model and get excellent results.

Our expert data annotation ensures that:

  • Path The tasks that your ML model performs have a broad range of application
  • Path Copy 5 The training is much more efficient with labeled data (clean and processed)

Processing the data means removing all the low-quality data (think low-resolution images, blurry or underexposed photos, etc.).

Annotating the data means adding meaningful tags to the elements of the data (such as objects on an image) that will explain to a machine how to interpret this object.

TODAY, it’s best to let the experts do their job!

Solution: Quest for the Perfect Partner

The team contacted the CEO of Label Your Data, Karyna Naminas, with an interesting proposition: help to build a driverless race car.

At Label Your Data, we always offer a pilot so that a client could check if the annotation experts understand the task correctly and if the quality, security, and speed meet the expectations. For Elefant Racing, the trust between our teams was established almost immediately. Label Your Data annotator started working on the project with frequent feedback exchanges. Within a few weeks, the Elefant Racing team was 100% sure that we understood what kind of annotation they required.

quote-icon

I was immediately interested and intrigued by the offer that the Elefant Racing team made.

Label Your Data was a young start-up looking for curious and innovative projects to strengthen domain expertise. Together with our co-founder, Daria Leshchenko, we were driven by a keen interest in the automotive industry and Sebastian’s dedication to the project. Our experience in supporting annotation projects across multiple industries made it even more exciting for Label Your Data to help the Elefant team with their glorious undertaking! quote-icon

block-bg-photo Karyna Naminas
CEO, Label Your Data
block-bg-photo Sebastian Sartor
Operations Lead at Elefant Racing
quote-icon

I remember we were really surprised.

Fabian and Maksym had the first meeting and mentioned to me that they didn't think it’d be successful. But then, surprisingly, [the Label Your Data team] was interested. We were very happy and relieved. Maybe not everyone knew, but for me, this meant soooo much improvement of our team efficiency. quote-icon

December 2020: Label Your Data provides a labeling expert at full disposal of the Elefant Racing team.
As the project progressed, so did the demand for annotation. Step-by-step, the process unfolded.
From December until the end of July, Label Your Data annotators were working on the labeling tasks.
The Elefant team proceeded with training, final tests, and the start of the Formula Racing events.

Results: Annotation Process and Competition Achievements

The tests started in the Spring of 2021 and kept going until the roll-out in late June. This was the moment of truth when FR20 Ragnarök was first demonstrated to the wide public. By that time, the annotators of Label Your Data spent almost 2 000 hours on the labeling tasks, which consisted of annotating traffic cones using two major types of annotation.

Major Annotation Types Used for Data Labeling for Elefant Racing

  • Path Bounding boxes: commonly used for autonomous driving tasks, this type of annotation (also known as 2D boxes) looks like a frame around an object, which was necessary to train the Elefant model to recognize the sizes and dimensions of other race cars, people, etc.
  • Path Copy 5 Keypoint annotation: aka landmark annotation, the keypoint type is crucial for defining the main (key) points for a natural object, which helped train an ML algorithm of the Elefant race car to predict the shapes and movement of said objects. Keypoint annotation was used to track the position of traffic cones.

Performance of FR20 Ragnarök During Formula Student Events

As a result of this annotation process, the Elefant Racing team was able to train the model in the nick of time, ready for the presentation of FR20 Ragnarök during the FS in Zalaegerszeg, Hungary.

Taking place on August 7-11, Hungarian Formula Student presented a set of challenges for the Elefant team during both static and dynamic events. Static events required the team to demonstrate their knowledge of both the business and engineering parts of building a race car. At this time, a lot of competing teams with driverless vehicles failed the check-ups, either due to malfunctioning hardware or low-quality model design and training.

For the Elefant team, however, the presentations and security check-up went smoothly and the team got their permission to take part in the dynamic events, including acceleration event, driving eights, and unknown tracks racing. Despite competing against larger and more well-known teams that had experience in building driverless vehicles, the Elefant Racing team was able to get the 1st place in Design Judging at Formula Student East! The second FS competition that the Elefant Racing team participated in took place in Hockenheimring, Germany. During the events in FS Germany, the team set a few new records, including:

  • 🏅 The lowest weight under 205 kg
  • 🏅 Fastest time during the skidpad test at 5.19 seconds

These achievements are surprising and very significant for the team without previous experience in building autonomously driven race cars. This would not be possible without the passion and dedication of the team, as well as the successful partnership of Elefant Racing and Label Your Data.

performance-img
performance-img
Conclusion: The Stellar Collaboration of Elefant Racing and Label Your Data
conclusion-img

Building a project as complex and multifaceted as a driverless race car is a challenging and ambitious endeavor. The Elefant Racing team had to plan and propose a working business model, to find all the necessary components and materials to build a physical car, and then to design and train an ML algorithm that would enable it to drive autonomously.
It is said that a friend in need is a friend indeed. We at Label Your Data are happy that we were able to help the Elefant Racing team when they were on the lookout for the annotation partner. For us, it was an intriguing experience that taught us a lot about passion and determination. For Elefant Racing, it was a learning process that demonstrated how experts in different fields can come together and build something as sophisticated as FR20 Ragnarök. And despite the lack of faith in the victory of the first ever Elefant driverless race car, the dedication and hard work paid off and resulted in a high place during FS events, as well as a few new Elefant team records!

TASK US WITH YOUR SAMPLE DATA & SLA SUBMIT SAMPLE DATA & GET YOUR QUOTE