Code the Light Fantastic
Share:
twitter social icon linkedin social icon facebook social icon
Finished
competition-bg

Code the Light Fantastic

Simulate an adaptive matrix headlight

KTM AG
Machine Learning/AI
Enterprise
Total Prize 24,000
Join Datathon
Scroll To Top

Datathon Type

challenge_cup

Challenge with monetary prizes

No. of Users

156

No. of Submission

147

No. of Teams

12

Community chat

Join our Discord

Description

Description of the competition

KTM AG, a global frontrunner in two-wheeler innovation, is pushing forward into the domain of artificial intelligence and deep learning, and we're inviting you to be a part of this groundbreaking journey. Introducing the KTM AG inaugural Code Challenge, an exciting 3-month online journey designed to harness the collective intelligence, ignite passion, and translate visionary ideas into transformative two-wheeler technology. 

Tasks Description: At the heart of this challenge, participants are tasked with developing an algorithm for a high-beam lighting system that uses a pixel matrix. The primary goal is to:
  • Detect objects in front of the vehicle from a captured image.
  • Identify and map the exact region these objects occupy within the image onto the pixel matrix.
  • Dim or turn off the corresponding pixels in the lighting system.

This creates an adaptive high-beam system that targets and dims only specific areas aligned with detected objects, ensuring the rest of the road remains well-lit. Further guidelines are available in the Dataset subsection.

The Competion Structure:

The datathon unfolds in a 3-tiered cascade model:

Level 1
Participants are provided with a video filmed in "easy" conditions, and with a number of "Key Frames" which show the ideal output of the algorithm at these snapshots in time. The illumination state of the pixels at that point in time are encoded into a CSV file. The algorithm of the participant should aim to output the pixels as close as possible to what is requested in the key frame. This algorithm is then assessed by an automatic assessment, and then a manual, on board assessment, to verify the code will run on the target hardware. A score is provided on how accurate the participants' key frames match those of KTM AG and if the minimum score is reached the candidate may pass to Level 2.

Level 2
Participants are provided with a new video in more difficult conditions along with another set of Key Frames.The assessment method is similar to Level 1, except the tolerance on the Key Frames will be tighter and the minimum score to pass to Level 3 will be higher.

Level 3
Participants are provided with another new video, with further difficult conditions and the corresponding Key Frames. The first two assessment phases are similar to that of Level 1 and Level 2, and the quality level will be similar to that of Level 2.  Unlike Level 2, the efficiency and processing footprint of the algorithm will also be analysed and the submission with the highest quality and smallest footprint is eligible to win to the final prize.

Prize Pool:

  • Level 1: 25 prizes at €200 each.
  • Level 2: 5 prizes at €600 each.
  • Level 3: 1st Place wins €10,000, with the 2nd Place receiving €6,000.

Prizes for Level 1 and 2 are allocated on a first come basis, so the quicker you are, the better your chances of winning a prize, however even if there are no prizes left, new entrants can still compete for the Level 3 prize. Prizes are also cumulative, so the final winner could take home a maximum of €10,800!

With hefty prizes, intriguing challenges, and the chance to shape the future of two-wheeler technology, KTM AG's datathon promises an unmatched experience. Join us in this Code the light fantastic challenge! 

Description of the competition

KTM AG, a global frontrunner in two-wheeler innovation, is pushing forward into the domain of artificial intelligence and deep learning, and we're inviting you to be a part of this groundbreaking journey. Introducing the KTM AG inaugural Code Challenge, an exciting 3-month online journey designed to harness the collective intelligence, ignite passion, and translate visionary ideas into transformative two-wheeler technology. 

Tasks Description: At the heart of this challenge, participants are tasked with developing an algorithm for a high-beam lighting system that uses a pixel matrix. The primary goal is to:
  • Detect objects in front of the vehicle from a captured image.
  • Identify and map the exact region these objects occupy within the image onto the pixel matrix.
  • Dim or turn off the corresponding pixels in the lighting system.

This creates an adaptive high-beam system that targets and dims only specific areas aligned with detected objects, ensuring the rest of the road remains well-lit. Further guidelines are available in the Dataset subsection.

The Competion Structure:

The datathon unfolds in a 3-tiered cascade model:

Level 1
Participants are provided with a video filmed in "easy" conditions, and with a number of "Key Frames" which show the ideal output of the algorithm at these snapshots in time. The illumination state of the pixels at that point in time are encoded into a CSV file. The algorithm of the participant should aim to output the pixels as close as possible to what is requested in the key frame. This algorithm is then assessed by an automatic assessment, and then a manual, on board assessment, to verify the code will run on the target hardware. A score is provided on how accurate the participants' key frames match those of KTM AG and if the minimum score is reached the candidate may pass to Level 2.

Level 2
Participants are provided with a new video in more difficult conditions along with another set of Key Frames.The assessment method is similar to Level 1, except the tolerance on the Key Frames will be tighter and the minimum score to pass to Level 3 will be higher.

Level 3
Participants are provided with another new video, with further difficult conditions and the corresponding Key Frames. The first two assessment phases are similar to that of Level 1 and Level 2, and the quality level will be similar to that of Level 2.  Unlike Level 2, the efficiency and processing footprint of the algorithm will also be analysed and the submission with the highest quality and smallest footprint is eligible to win to the final prize.

Prize Pool:

  • Level 1: 25 prizes at €200 each.
  • Level 2: 5 prizes at €600 each.
  • Level 3: 1st Place wins €10,000, with the 2nd Place receiving €6,000.

Prizes for Level 1 and 2 are allocated on a first come basis, so the quicker you are, the better your chances of winning a prize, however even if there are no prizes left, new entrants can still compete for the Level 3 prize. Prizes are also cumulative, so the final winner could take home a maximum of €10,800!

With hefty prizes, intriguing challenges, and the chance to shape the future of two-wheeler technology, KTM AG's datathon promises an unmatched experience. Join us in this Code the light fantastic challenge! 

Submission Guidelines

The participant must submit the output of their model for selected key-frames in a specific format for automated evaluation, multiple submissions can be made by the participant as they build and improve their models. These results will be considered for the leaderboard. 

Submissions must follow the following guidelines.
  1. Please submit the output for these frames: [1000, 1136,  125, 1520, 2060, 2140, 2310,  275, 2800, 3165, 3246, 3305,3543, 3740, 3794, 4752, 5050,5800,593]
  2. CSV Unicode UTF 8 format
  3. “CompositeKey” column, this is a unique identifier generated as follows: “Key frame number”-”Pixel Row”-”Pixel Column”. Eg:”1000-2-120”
  4. “Value” column, with the predicted value for each pixel. 
  5. High-beam On = 1
  6. High-beam off = 0

A sample file has been generated using the provided sample pixel matrix CSVs: 
For full evaluation, participant must submit to the platform their proposed solution, note that participants will not be considered for the prizes in the absence of the following:
  1. A script to be run within the code template
  2. CSV files with the results for the key frames, please match the sample key frames CSV format(256 columns x 64 rows)
  3. Any additional material necessary for the functioning of their solution
  4. Documentation with the resources used and a description of the code

Resources

  • Code template
  • Sample key frames
  • Solved CSV files correspond to each of the above key frames(256x64 pixels) with 1 denoting high-beam on and 0 denoting high-beam off. Note that the CSV files only covers the specific area applicable to the high-beam(identified on the code template)
  • Sample CSV file with submission format for automated evaluation.
  • Riding Information CSV:  As support for the project, an optional csv file is attached with information about speed, lean and pitch angles and the moment when the rider turns the High-Beam on or off for each specific frame (optional material for the participant in case he/she finds it useful). 
The need for cloud processing
For participants concerned about resource limitations for training their models, we recommend participants to leverage the power and convenience of Google Colab's free offering for their model training needs. Google Colab provides an exceptional platform that combines the prowess of Python programming with the accessibility of cloud computing. With its seamless integration with Google Drive and pre-installed libraries, participants can efficiently develop, prototype, and train their machine learning models without the hassle of setting up complex environments. The generous allocation of GPU resources ensures faster training times, enabling participants to iterate on their models quickly.
 
Performance
System-on-module: AGX Xavier [64GB]: 512 Core GPU, 64 Tensor Cores, 32 TOPs
Operating System: Linux Ubuntu 18.04
JetPack Version: JetPack 4.6

Graphics
AI Accelerator: NVIDIA Jetson SOM
Number of GPUs: Single GPU
GPU Architecture: NVIDIA Volta 

Timeline

22

Challenge starts

August at 06:00 UTC

Add to Calendar

12

Challenge ends (Public leaderboard)

November at 21:59 UTC

Add to Calendar

12

Challenge ends (Private leaderboard)

November at 21:58 UTC

Add to Calendar

December at 20:59 UTC

The deadline to submit for the Level 2 Challenge. Ensure you turn in your submissions by 5th November. Success in this stage will secure your spot in the coveted Level 3. If you've successfully navigated Level 1, don't stop now – your journey continues. Level 3 awaits those who rise to the challenge!
Add to Calendar

January at 22:59 UTC

Mark it on your calendar, because this is the ultimate deadline for Level 3 submissions. Ensure your work is submitted in time to be part of the challenge.
Add to Calendar

Description of Timeline


Registration Details:

Though registrations will remain open, we strongly recommend registering at the earliest to ensure your participation, and to begin receiving crucial event updates. Early birds will also have an advantage in getting familiarized with the datathon's requirements and nuances.
Leaderboards and Competition Structure  (LEVEL1)

  1. Private Leaderboard:
    • This leaderboard is reserved for those who have adhered strictly to our submission guidelines.
    • Only those ranked on the private leaderboard are eligible for the final prize.
    • Though you might notice fewer competitors here compared to the public leaderboard, rest assured this is intentional.
    • End Date for Private Leaderboard submissions: 12 November, 2023 CET
      Remember, only entries submitted within this timeline qualify for the prize consideration.

Please Note: Any alterations in the timeline lie solely at the discretion of the organizers. Staying updated is crucial. Hence, we once again urge all participants to register promptly to ensure they receive all pertinent updates related to the datathon.

Join us for an event filled with innovation. Code the light, and may the best data enthusiast win!

Evaluation

Performance & Evaluation

The detailed evaluation methodology for our datathon will be published alongside the official launch.

Our aim is to ensure that all participants are on a level playing field and understand how their solutions will be assessed. We believe this transparency will not only help in guiding your efforts during the event but also ensure that the results are judged in a fair and consistent manner.

Stay tuned for more updates as we approach the official launch date! In the meantime, we encourage you to start forming teams, brainstorming, and getting ready for the challenge.

We look forward to seeing the innovative solutions you'll bring to the table!

Join our private community in Discord

Keep up to date by participating in our global community of data scientists and AI enthusiasts. We discuss the latest developments in data science competitions, new techniques for solving complex challenges, AI and machine learning models, and much more!

Join our Discord