Attention Bottlenecks for Multimodal Fusion

Open paper
Benchmark Model Rank Results
action-classification-on-kinetics-400MBT (AV)#81Acc@1: 80.8Acc@5: 94.6
action-classification-on-kinetics-soundsMBT (AV)#1Top 1 Accuracy: 85Top 5 Accuracy: 96.8
action-classification-on-moments-in-timeMBT (AV)#10Top 1 Accuracy: 37.3Top 5 Accuracy: 61.2
action-recognition-on-epic-kitchens-100MBT#20Action@1: 43.4Verb@1: 64.8Noun@1: 58
audio-classification-on-audiosetMBT (AS-500K training + Video)#9Test mAP: 0.496
audio-classification-on-vggsoundMBT (A)#12Top 1 Accuracy: 52.3Top 5 Accuracy: 78.1
audio-classification-on-vggsoundMBT (V)#14Top 1 Accuracy: 51.2Top 5 Accuracy: 72.6
audio-classification-on-vggsoundMBT (AV)#18Top 5 Accuracy: 85.6

Rank counts only results with a code link.