Self-Supervised Spatiotemporal Feature Learning via Video Rotation Prediction

Benchmark Model Rank Results
self-supervised-action-recognition-on-hmdb513D RotNet (3D ResNet-18)–Top-1 Accuracy: 33.7Pre-Training Dataset: Kinetics400Frozen: false
self-supervised-action-recognition-on-ucf1013D RotNet (3D ResNet-18)–3-fold Accuracy: 62.9Pre-Training Dataset: Kinetics400Frozen: false