CA^2ST: Cross-Attention in Audio, Space, and Time for Holistic Video Recognition

Benchmark Model Rank Results
action-classification-on-kinetics-soundsCA2ST(B/16)–Top 1 Accuracy: 93.3
action-classification-on-kinetics-soundsCAVA(B/16)–Top 1 Accuracy: 92.9
action-recognition-in-videos-on-ucf101CA2ST(B/16)–3-fold Accuracy: 97.2
audio-classification-on-vggsoundCA2ST(B/16)–Top 1 Accuracy: 68.3
audio-classification-on-vggsoundCAVA(B/16)–Top 1 Accuracy: 68.2