FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation

Benchmark Model Rank Results
action-segmentation-on-breakfast-1FACT (efficient hybrid of convolution and transformer model)#1Average F1: 74.7F1@50%: 66.2F1@25%: 76.5F1@10%: 81.4Edit: 79.7
action-segmentation-on-gtea-1FACT#1F1@50%: 87.5F1@25%: 95.6F1@10%: 96.1Acc: 84.5Edit: 93.5