SA-DVAE: Improving Zero-Shot Skeleton-Based Action Recognition by Disentangled Variational Autoencoders

Benchmark Model Rank Results
generalized-zero-shot-skeletal-actionSA-DVAE#2Harmonic Mean (5 unseen classes): 66.27
generalized-zero-shot-skeletal-actionSA-DVAE + augmented text#4Random Split Harmonic Mean: 75.51
generalized-zero-shot-skeletal-action-1SA-DVAE#1Harmonic Mean (10 unseen classes): 60.42
generalized-zero-shot-skeletal-action-1SA-DVAE + augmented text#4Random Split Harmonic Mean: 50.72
zero-shot-skeletal-action-recognition-on-ntu-120SA-DVAE#4Accuracy (10 unseen classes): 68.77Accuracy (24 unseen classes): 46.12
zero-shot-skeletal-action-recognition-on-ntu-120SA-DVAE + augmented text#8Random Split Accuracy: 57.16
zero-shot-skeletal-action-recognition-on-ntu-60SA-DVAE#3Accuracy (12 unseen classes): 41.38Accuracy (5 unseen classes): 82.37
zero-shot-skeletal-action-recognition-on-ntu-60SA-DVAE + augmented text#8Random Split Accuracy: 87.61
zero-shot-skeletal-action-recognition-on-pkuSA-DVAE#3Random Split Accuracy: 66.54