Competition Details

LPCVC addresses three foundational and critical directions in computer vision. Utilizing the Qualcomm AI-Hub website and toolbox, this challenge focuses on optimizing and deploying models on edge devices, including mobile phones and AI PCs. Participants will submit their models online, with various acceptable formats like Pytorch, Tensorflow, Tflite and ONNX, making the process convenient and accessible to a global community of developers.

This challenge will be helpful for advancing efficient computer vision, encouraging developers to create CV applications that run directly and easily on real devices. Unlike other existing competitions, LPCVC’s topics are derived from practical, real-world applications on edge devices. Participants do not necessarily use large or heavy cloud GPU servers to train those models targeted at edge devices, significantly lowering the hardware requirements for developers, researchers, and students.

Open-source sample solutions will be provided as references and qualification. If a submitted solution is inferior to the sample solution, the submission is disqualified. The winning solutions will be publicly released along with the test dataset, fostering further development within the efficient computer vision community.

All teams need to register on lpcv.ai website (registration will be open later) with the signed Participant Agreement and Acknowledgement. Visit each track’s introduction page for more details.

Participation Steps

1. Subscribe to the newsletter or join the Qualcomm AI Hub Slack workspace. Make sure to join channel #lpcvc for competition related announcements and for your questions.
2. Register accounts on Qualcomm AI Hub for testing the sample solutions and for your own evaluation. All submitted models for all tracks need to be compiled and tested successfully on the AI Hub with corresponding hardwares, before submitting to our referee system. If our referee system could not process the model ID, the submitted solution will be considered invalid.
3. Develop your models for the tracks you would like to participate in. Try out the sample solution and referee system GitHub repos for each track on your own to see the results with sample data. You can also replace the sample solution model with your model to test the evaluation results with sample data.
4. Team registration portal will be open on February 3, 2025 on our website.
5. After team registration, the submission portal will be open on each track’s information page after the submission window open date (March 1, 2025). Teams are expected to submit the model ID and the compile job ID for the referee system to retrieve the model from the AI Hub database for evaluation.
6. During the submission window, only submissions from registered teams will be evaluated. Additionally, we will only evaluate the most recent submission from each team made on the previous day.
7. The submitted model will be run with both public sample data and hidden testing data. The results will be shown on the public leaderboard on our website.
8. After submission window closes, winners will be announced on our website, through emails, and on #lpcvc Slack channel.

Organizers  / Advisors

FAQ

1. Can we participate in the LPCVC2025 if my team is not located in the United States?

Answer: Yes. Anyone is welcome to submit to the 2025 LPCVC any track.

2. How can I receive the prize, if my submission wins the track?

Answer: The winners will be invited to our workshop at 2025 CVPR (pending acceptance), and we will hand over the winner’s prizes. We only support shipping for United States addresses upon request.

3. How do I submit the model to participate in one of the tracks?

Answer: Team registration will be enabled before the submission window opens. Stay tuned for any updates/announcements through the Newsletter or join Qualcomm AI Hub Slack Workspace! Make sure to join channel #lpcvc for competition related notifications.

4. What is the timeline for this challenge?

Answer: The 2025 LPCVC submission window will open from 11:59 PM Pacific Time on March 1, 2025, to 11:59 PM Pacific Time on April 1, 2025.

5. Do we need to provide or open-source our winning solutions?

Answer: Yes. For a team to win in each track, the winning solutions are required to be open-sourced. More details can be found in the Participant Agreement.

6. What happens if the https://app.aihub.qualcomm.com/ is down?

Answer: The Qualcomm AI Hub website is updated every Monday at 3:00 PM PST. Please wait for around two hours and check again.