Action Recognition on CMU subset 664 sequences (3-fold cross validation)
93.22AccuracyClips + CNN + MTLN
Evaluation results
| Method | Links | |
|---|---|---|
| Clips + CNN + MTLNinput=Clips, architecture=MTLN2017.03 | 93.22 | |
| Frames + CNNinput=Individual Frames2017.03 | 91.53 | |
| Clips + CNN + Concatenationinput=Clips, aggregation=Concatenation2017.03 | 90.97 | |
| Clips + CNN + Poolinginput=Clips, aggregation=Pooling2017.03 | 90.66 | |
| Deep LSTM + Co-occurrence2017.03 | 88.4 | |
| Deep LSTM2017.03 | 86 | |
| Coordinates + FTPinput=Coordinates, feature=FTP2017.03 | 83.44 | |
| Hierarchical RNN2017.03 | 83.13 |