A Hybrid RNN-HMM Approach for Weakly Supervised Temporal Action Segmentation

Hilde Kuehne, Alexander Richard, Juergen Gall

URL

Proceedings: TPAMI

Year: 2019

Peer Reviewed : Yes

Results: 
Breakfast Dataset /BF Action seg. weakly (MoF): 36.7% (GRU+length prior)
Breakfast Dataset /BF Action alignment weakly (IoD): 52.4% (GRU+length prior)
Hollywood Extended /HE Action seg. weakly (MoF): 12.3% (GRU+length prior)
Hollywood Extended /HE Action alignment weakly (IoD): 46% (GRU+length prior)
Breakfast Dataset /BF Action seg. fully (MoF): 61.3% (GRU+length proir)