CVPR 2020 Continual Learning in Computer Vision Competition: Approaches, Results, Current Challenges and Future Directions

In the last few years, there has been renewed interest in continual learning with deep neural networks aimed at making AI systems more adaptive, efficient and autonomous. Despite progress in addressing catastrophic forgetting, benchmarking different approaches remains challenging due to proliferation of settings, protocols, metrics and nomenclature. The first Continual Learning in Computer Vision challenge at CVPR 2020 evaluated different algorithms on shared hardware with consistent metrics across 3 settings based on the CORe50 video benchmark. The competition included 79 registered teams, 11 finalists and $2,300 in prizes. This paper reports competition results, winning approaches, current challenges and future research directions.