AssembleNet++: Assembling Modality Representations via Attention Connections

Benchmark Model Rank Results
action-classification-on-charadesAssembleNet++ 50#4MAP: 59.8
action-classification-on-charadesAssembleNet++ 50 without object#7MAP: 54.98
action-classification-on-toyota-smarthomeAssembleNet++#5CS: 63.6