## Announcement of Challenge Winners We're excited to reveal the top performers in the Anticipating Vehicle Accidents Challenge! The following participants will be awarded certificates, and the first-place winner will receive an additional prize. Congratulations!!! --- 1st Place: Matthew1012, with a score of 0.67142 Solution: Predicting Car Accidents with YOLOv7 Object Detection and Object Relationships 2nd Place: ALLAccept, with a score of 0.61428 Solution: Improved Dynamic Spatial-Temporal Attention Network for Early Anticipation of Traffic Accidents --- ## Private Leaderboard Score Released We are pleased to announce the top three performers on our private leaderboard for the recent competition. These participants have achieved remarkable scores. --- 1st Place: Matthew1012, 0.67142 2nd Place: Huang Yu Hung, 0.61904 3rd Place: ALLAccept, 0.61428 4th Place: asd3200asd, 0.59047 5th Place: Host(not included) 6th Place: abida gurman, 0.58571 --- We will send an email to the above participants to notify them to submit a link to their technical report (1 page) and source code. Please stay tuned for this important message. All of the aforementioned participants are required to submit their challenge papers. Please note the deadline for paper submission and we encourage all participants to submit their work as well. **4/21 Latest News!!** 1. The deadline for **Final Submission** is based on UTC time (8:00 a.m. UTC+8). Please ensure to upload the files in time, thank you! 2. The paper submission deadline has been moved up from **May 1** to **April 28** due to ICME committee decision. We will send out the acceptance notice of your papers on April 29. Participants are kindly reminded to take note of this change. 3. Camera-ready deadline is May 1 UTC time. ## Goal of the Competition Video analysis research is crucial for autonomous driving as it can assist with accident prediction. Dashcam videos can offer valuable information for this task. However, there is a scarcity of well-annotated, large-scale datasets for accident prediction in Asian countries. We collected and labeled a large-scale vehicle collision dataset from Taiwan to address this gap for frame- and video-level predictions. The dataset aims to inspire participants from academia and industry to enhance safety in autonomous driving. The track of this challenge is vehicle accident prediction. Please try to predict whether there will be a car accident within the next 20 frames. More information can be found on the website [https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0](https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0) **In order for award recipients to participate in the IEEE ICME2023 Grand Challenge AVA Special Session and present their work, they are required to submit challenge papers and give their consent. Failure to submit the paper or submitting a paper that is less than 3 pages in length will result in the cancellation of the award.** ## Context According to the research, self-driving cars can only reduce about one-third of major car accidents that occur in the United States out of 5,000 incidents. People typically assume that autonomous driving systems can eliminate human errors while driving. However, many car accidents are still caused by human factors, including perception errors, driver distraction, poor visibility, or slow reactions. Other possible causes are prediction errors, such as misjudging the distance to other vehicles or pedestrians' movements. Some errors may be related to planning and decision-making, such as inappropriate evasive actions or overcompensation when controlling the vehicle. Other accidents can be caused by impairment due to alcohol, drugs, or drowsiness. There are avoidable accidents that are caused by vehicle malfunctions. Based on the above, we propose a topic for participants to use existing data to predict whether a car accident will occur in the next second to facilitate crisis response in self-driving and improve traffic safety. The competition host is Advanced Computer Vision Lab (ACVLab), a research team led by Professor Chih-Chung Hsu at National Cheng Kung University. Our lab focuses on introducing machine learning/deep learning into various applications in computer vision and cultivating students with high practical and applied capabilities in deep learning. Your efforts in this competition could help extend the benefits of road safety. Greater access could further reduce the occurrence of car accidents worldwide. ## Contact Information If you have any questions, please contact the following ACVLab members. Yun-Zhong Jiang (yunzhong1105@gmail.com) Wei-Hao Huang (robert20000831@gmail.com) ## Rank The valid submission only if the participants have submitted their report, code, and successful submission on kaggle could be considered as the winner. For the winners of the competition (top three), we will invite the top three to five to share their winning solutions at ICME and award certificates to the top three. For other information could refer to our official website https://sites.google.com/view/tvcd-tw/ava-challenge?authuser=0