Table 4 Comparison of experimental results of different 3D-CNN networks.
From: Improved 3D-ResNet sign language recognition algorithm with enhanced hand features
Method | FLOPs (G) | Parm (M) | Top1 (%) | Top2 (%) | Top3 (%) | Top4 (%) | Top5 (%) | AVG (%) |
|---|---|---|---|---|---|---|---|---|
C3D | 32.998 | 78.405 | 78.34 | 88.44 | 91.70 | 93.72 | 94.80 | 89.40 |
C3D + local | 98.961 | 145.514 | 87.42 | 94.08 | 96.30 | 97.40 | 98.20 | 94.68 |
3D-ResNet10 | 5.654 | 14.445 | 76.82 | 86.04 | 90.30 | 92.34 | 93.20 | 87.74 |
3D-ResNet10 + local | 16.961 | 14.496 | 84.84 | 92.46 | 95.14 | 96.68 | 97.62 | 93.35 |
3D-ResNet34 | 12.696 | 63.548 | 76.58 | 83.62 | 86.24 | 88.10 | 89.26 | 84.76 |
3D-ResNet34 + local | 38.088 | 63.599 | 87.90 | 94.04 | 96.14 | 97.30 | 97.84 | 94.64 |
P3D | 4.192 | 25.073 | 77.50 | 88.74 | 93.32 | 95.10 | 96.08 | 90.15 |
P3D + local | 12.577 | 25.278 | 72.32 | 83.18 | 87.94 | 90.26 | 92.38 | 85.22 |
3D-ResNet18 (Baseline) | 8.308 | 33.246 | 81.06 | 88.98 | 92.22 | 94.00 | 94.78 | 90.21 |
3D-ResNet18 + local | 24.924 | 33.297 | 88.66 | 94.30 | 96.42 | 97.44 | 98.02 | 94.97 |
