Continual Transformers: Redundancy-Free Attention for Online Inference

Benchmark Model Rank Results
online-action-detection-on-thumos-14OadTR-b2#5mAP: 64.5MFLOPs per pred: 1075.7
online-action-detection-on-thumos-14CoOadTR-b2#6mAP: 64.4MFLOPs per pred: 411.9
online-action-detection-on-thumos-14OadTR#7mAP: 64.2MFLOPs per pred: 2513.5
online-action-detection-on-thumos-14OadTR-b1#8mAP: 63.9MFLOPs per pred: 673
online-action-detection-on-thumos-14CoOadTR-b1#11MFLOPs per pred: 10.6
online-action-detection-on-tvseriesOadTR#3mCAP: 88.6
online-action-detection-on-tvseriesOadTR-b2#4mCAP: 88.3
online-action-detection-on-tvseriesOadTR-b1#5mCAP: 88.1
online-action-detection-on-tvseriesCoOadTR-b1#7mCAP: 87.7
online-action-detection-on-tvseriesCoOadTR-b2#8mCAP: 87.6