XKD: Cross-modal Knowledge Distillation with Domain Alignment for Video Representation Learning

Benchmark Model Rank Results
self-supervised-action-recognition-on-kinetics-1XKD (ViT-B/112/16)#1Top-1 accuracy %: 77.6Top-5 Accuracy %: 92.9
self-supervised-action-recognition-on-hmdb51XKD (ViT-B/112/16)#8Top-1 Accuracy: 69
self-supervised-action-recognition-on-hmdb51XKD-Modality-Agnostic (ViT-B/112/16)#14Top-1 Accuracy: 65.9
self-supervised-action-recognition-on-ucf101XKD (ViT-B/112/16)#83-fold Accuracy: 94.1Pre-Training Dataset: Kinetics400
self-supervised-action-recognition-on-ucf101XKD-Modality-Agnostic (ViT-B/112/16)#133-fold Accuracy: 93.4