Challenge Results

Twenty-nine teams from academic labs in Brazil, China, Greece, India, Italy, Japan, Turkey and the United States signed up to participate in the IEEE Smart World NVIDIA AI City Challenge. After the annotation phase, which generated more than 150,000 annotated video keyframes and 1.4 million annotations, 18 teams were chosen to compete in two tracks: modeling and applications.

Track 1: Modeling

Winner: Team 24, University of Illinois at Urbana Champaign.

The winning team was selected by a panel of judges based on overall performance on object detection, localization, and classification. Since the AIC540 and AIC1080 datasets represent the same set of videos, the formula used by the judges to rank teams in this track was the mean of a team's AIC480 mean average precision (mAP) score and the best of the AIC540 or AIC1080 mAP scores. The mAP scores were computed as the mean of per-class average precision (AP) scores. The tables below show the AP scores in each class for each team's best submitted model.

The 'Challenge' tab below shows results for models teams submitted up to the challenge deadline. After the challenge workshop, teams could continue improving models and submit additional results for two weeks prior to the camera-ready paper deadline. Click on the 'Camera Ready' tab to see scores from updated models the teams submitted prior to the camera-ready deadline.

AIC480: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

AIC540: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

AIC1080: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

AIC480: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

AIC540: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

AIC1080: Average Precision


Note: Click on the class or model name to display the associated Precision-Recall graph.

Track2: Applications

Winner: Team 4, University of Washington.

Honorable Mention: Team 5, SUNY Albany.

The winning team was selected by a panel of judges based on novelty (33.3%), value added to a Smart City (33.3%), and the quality of the demonstration (33.3%).


For any questions, please email nvidiaAICitychallenge@gmail.com.