2025 Challenge Winners – AI CITY CHALLENGE

Challenge Winners 2025

Track 1:

Winner: Team65 ZV

Multi-Camera 3D Object Tracking via 3D Point Clouds and Re-Identification

Runner-up: Team15 SKKU-AutoLab

DepthTrack: Cluster Meets BEV for Multi-Camera Multi-Target 3D Tracking

The following table shows the top teams from the public leader board of Track 1 with paper submissions by the challenge submission deadline.

Track 2:

Winner: Team145 CHTTLIOT

TrafficInternVL: Understanding Traffic Scenarios with Vision–Language Models

Runner-up: Team1 SCU_Anastasiu

Multi-Agent Cooperation for Traffic Safety Description and Analysis

The following table shows the top teams from the public leader board of Track 2 with paper submissions by the challenge submission deadline.

Rank	Team ID	Team Name	Score
1	145	CHTTLIOT	60.0393
2	1	SCU_Anastasiu	59.1184
3	52	Metropolis_Video_Intelligence	58.8483
4	137	ARV	57.9138
5	121	Rutgers ECE MM	57.4658
6	68	VNPT_AI	57.1133
7	60	BAO_team	55.6550
10	49	MIZSU	45.7572

Track 3:

Winner: Team16 UWIPL_ETRI

Warehouse Spatial Question Answering with LLM Agent: 1st Place Solution of the 9th AI City Challenge Track 3

Runner-up: Team57 HCMUT.VNU

Multimodal and Multi-task Fusion for Spatial Reasoning

The following table shows the top teams from the public leader board of Track 3 with paper submissions by the challenge submission deadline.

Track 4:

Winner: Team5 SmartVision

Augmentation, distillation and optimization: A practical pipeline for fisheye object detection on edge devices

Runner-up: Team15 SKKU-AutoLab

Data Augmentation Is All You Need For Robust Fisheye Object Detection

The following table shows the final ranking based on the multi-step docker evaluation by Dr. Gochoo and his team.

All Docker submissions have been evaluated on the Jetson AGX Orin 64GB platform, configured with a 30W power mode and maximum frequency settings. The ranking in the 1st table is based on the harmonic mean of the normalized frames per second (FPS) and F1-score on FishEy1Keval dataset. All teams met the FPS > 10 requirement. The top2 teams from the 1st table are further evaluated by retraining and re-evaluating on the in-house dataset by Dr. Gochoo’s team and the results are shown in the 2nd table.