Research Article
Model Lightweighting for Real-time Distraction Detection on Resource-Limited Devices
Table 3
Comparisons with the state-of-the-art methods in the literature for the AUC dataset.
| Model | Source | Top-1 acc (%) | Params (M) |
| AlexNet [20] | Original AUC | 93.65 | 62 | Skin segmented | 93.60 | 62 | Face | 84.28 | 62 | Hands | 89.52 | 62 | Face + hands | 86.68 | 62 | InceptionV3 [20] | Original AUC | 95.17 | 24 | Skin segmented | 94.57 | 24 | Face | 88.82 | 24 | Hands | 88.82 | 24 | Face + hands | 90.88 | 24 | GA weighted ensemble of all 5 [20] | 95.98 | 120 | VGG [32] | Original AUC | 94.44 | 140 | VGG with regularization [32] | Original AUC | 96.31 | 15 | Our method | Original AUC | 94.77 | 2.78 |
|
|