Video Understanding and Soccernet
Motivation
Understand videos semantically and temporally. Videos and events can then be tagged and warehoused. Result videos can be used for retrieval.
Results
At the 2021 CVPR ActivityNet workshop, our team won first place in both the Soccernet Event Detection and Replay Grounding challenges. We released the features from our model and subsequently open-sourced an equivalent training pipeline based on Baidu's deep learning framework, PaddlePaddle. The features we released became the foundation for the methods that achieved top three positions in all the challenges during 2022 and 2023."
In 2022, we co-hosted the tracking challenge and provided the baseline methods for detection and tracking.
Publications
- Feature Combination Meets Attention- ASM-Loc: Action-Aware Segment Modeling for Weakly-Supervised Temporal Action Localization
- SoccerNet-Tracking: Multiple Object Tracking Dataset and Benchmark in Soccer Videos
- SoccerNet 2022 Challenges Result
- SoccerNet game state reconstruction: End-to-end athlete tracking and identification on a minimap