10th AI City Challenge Concludes at ECCV 2026
Published:
The 10th AI City Challenge successfully concluded at ECCV 2026 in Malmo, Sweden, marking a decade of community benchmarking for computer vision, multimodal AI, intelligent transportation, smart cities, and large-scale video analytics.
It was especially exciting to see NVIDIA AI highlight the challenge on X. The post pointed to a core theme of this year’s challenge: visual AI systems need to recognize scenes, understand what happened, and predict what could happen next. It also highlighted that four of the top five video-forecasting solutions used NVIDIA Cosmos.
I also appreciated the chance to speak with iMerit Technology about the challenge and the broader shift from perception-only computer vision toward multimodal reasoning systems. The conversation covered cross-city generalization, traffic anomaly reasoning, sim-to-real evaluation, and why edge cases remain one of the hardest parts of deploying reliable visual AI.
Across its ten editions, the AI City Challenge has engaged 2.5K participating teams across 48 countries and regions, growing from early traffic perception tasks into a broad benchmark suite for multi-camera 3D perception, transportation safety understanding, traffic anomaly reasoning, natural-language visual search, generative traffic video forecasting, and cross-city object detection.
This year’s ECCV workshop brought together organizers, participants, and winning teams across six main tracks and two out-of-domain leaderboards. I am grateful to have helped lead the organizing committee and to see the community continue pushing toward visual AI systems that are more robust, grounded, and useful in real-world environments.
