IR Object Detection
AI Challenge 2024
Infrared Dataset
Object Detection Challenge 2024
This competition aims to develop AI models for object detection using datasets acquired from infrared sensors in autonomous driving scenarios. By advancing AI model development, the goal is to enhance autonomous driving safety by enabling accurate environmental perception regardless of conditions, allowing for effective responses to hazardous situations through precise object detection.
Each team is tasked with predicting bounding boxes for eight predefined classes within infrared autonomous driving videos. The competition comprises two evaluation phases to determine the final winners:
-
First Evaluation: Teams submit predictions on a public evaluation dataset, with results displayed on a leaderboard. The top 10 teams from this phase advance to the next round.
-
Final Evaluation: The top 10 teams are assessed using a private evaluation dataset. Based on this evaluation, the top 3 teams are selected as the final winners.
IR Sensor
Object Detection Dataset 2024
-
The dataset comprises 10,883 infrared images captured using Hanwha Systems' Quantum RED thermal imaging sensor across various environments, times, and locations. The dataset is divided as follows: 7,018 training images, 600 validation images, 1,636 test open images, and 1,629 test private images. To evaluate model robustness across diverse conditions and mitigate overfitting, the test data is split into test open and test private subsets.
Schedules
24.08.14
-
Challenge Announcement
24.08.19
-
Dataset Release
-
Leaderboard Open
24.09.20
-
Leaderboard Close
-
Code Submission by Top 10 Teams
24.10.18
-
Final Rankings Announcement
24.11
-
Award Ceremony (Korean Artificial Intelligence Association Autumn Conference)
Dataset Samples
AI Challege 2024
Partners
