Table 7.
Obtained accuracies for the benchmark datasets with hand-crafted methods and deep learning methods.
| METHOD | UCF101 | HMDB51 | Weizmann | |
|---|---|---|---|---|
| Hand-crafted | Hierarchical [55] | - | - | 72.8% |
| Far Field of View [63] | - | - | 100% | |
| HOOF NLDS [46] | - | - | 94.4% | |
| Direction HOF [45] | - | - | 79.17% | |
| iDT [76] | - | 57.2% | - | |
| iDT+FV [76] | 85.9% | 57.2% | - | |
| OF Based [81] | - | - | 90.32% | |
| Edges OF [15] | - | - | 95.69% | |
| HOG features [87] | - | - | 99.7% | |
| Deep learning | Slow Fusion CNN [110] | 65.4% | - | - |
| Two stream (avg) [117] | 86.9% | 58.0% | - | |
| Two stream (SVM) [117] | 88.0% | 59.4% | - | |
| IDT+MIFS [162] | 89.1% | 65.1% | - | |
| LRCN (RGB) [121] | 68.2% | - | - | |
| LRCN (FLOW) [121] | 77.28% | - | - | |
| LRCN (avg, 1/2-1/2) [121] | 80.9% | - | - | |
| LRCN (avg, 1/3-2/3) [121] | 82.34% | - | - | |
| Very deep two-stream (VGGNet-16) [123] | 91.4% | - | - | |
| TDD [126] | 90.3% | 63.2% | - | |
| TDD + iDT [126] | 91.5% | 65.9% | - | |
| C3D [127] | 85.2% | - | - | |
| C3D + iDT [127] | 90.4% | - | - | |
| TwoStreamFusion [129] | 92.5% | 65.4% | - | |
| TwoStreamFusion+iDT [129] | 93.5% | 69.2% | - | |
| TSN (RGB+FLOW) [131] | 94.0% | 68.5% | - | |
| TSN (RGB+FLOW+WF) [131] | 94.2% | 69.4% | - | |
| Dynamic images + iDT [135] | 89.1% | 65.2% | - | |
| Two-StreamI3D [138] | 93.4% | 66.4% | - | |
| Two-StreamI3D, pre-trained [138] | 97.9% | 80.2% | - | |
| LTC (RGB) [139] | 82.4% | - | - | |
| LTC (FLOW) [139] | 85.2% | 59.0% | - | |
| LTC(FLOW+RGB) [139] | 91.7% | 64.8% | - | |
| LTC(FLOW+RGB)+iDT [139] | 92.7% | 67.2% | - | |
| DB-LSTM [141] | 91.21% | 87.64% | - | |
| Two-Stream SVMP(VGGNet) [143] | - | 66.1% | - | |
| Two-Stream SVMP(ResNet) [143] | - | 71.0% | - | |
| Two-Stream SVMP(+ iDT) [143] | - | 72.6% | - | |
| Two-Stream SVMP(I3D conf) [143] | - | 83.1% | - | |
| STPP + CNN-E (RGB) [145] | 85.6% | 62.1% | - | |
| STPP + LSTM (RGB) [145] | 85.0% | 62.5% | - | |
| STPP + CNN-E (FLOW) [145] | 83.2% | 55.4% | - | |
| STPP + LSTM (FLOW) [145] | 83.8% | 54.7% | - | |
| STPP + CNN-E (RGB+FLOW) [145] | 92.4% | 70.5% | - | |
| STPP + LSTM (RGB+FLOW) [145] | 92.6% | 70.3% | - |