Self-supervising Action Recognition by Statistical Moment and Subspace Descriptors

Benchmark Model Rank Results
action-classification-on-charadesDEEP-HAL with ODF+SDF (AssembleNet++)–MAP: 62.29
action-classification-on-charadesDEEP-HAL with ODF+SDF (I3D)–MAP: 50.16
action-recognition-in-videos-on-hmdb-51DEEP-HAL with ODF+SDF(I3D)–Average accuracy of 3 splits: 87.56
egocentric-activity-recognition-on-epic-kitchens-55DEEP-HAL with ODF+SDF (AssembleNet++)–Actions Top-1 (S1): 35.8Actions Top-1 (S2): 27.3
scene-recognition-on-yup-plus-plusDEEP-HAL with ODF+SDF (I3D)–Accuracy (%): 94.4