Table 5 Comparison of accuracy across different models.
From: A Large Crowdsourced Street View Dataset for Mapping Road Surface Types in Africa
Deep learning Model | Accuracy | Type | Precision | Recall | F1 score | Support* |
|---|---|---|---|---|---|---|
Resnet-50 | 0.922 | Paved | 0.919 | 0.924 | 0.921 | 1,100 |
Unpaved | 0.940 | 0.913 | 0.926 | 1,100 | ||
Unknown | 0.908 | 0.929 | 0.918 | 1,100 | ||
VGG-16 | 0.920 | Paved | 0.921 | 0.923 | 0.922 | 1,100 |
Unpaved | 0.927 | 0.917 | 0.922 | 1,100 | ||
Unknown | 0.912 | 0.920 | 0.916 | 1,100 | ||
Swin Transformer | 0.924 | Paved | 0.947 | 0.905 | 0.925 | 1,100 |
Unpaved | 0.932 | 0.928 | 0.930 | 1,100 | ||
Unknown | 0.896 | 0.939 | 0.917 | 1,100 | ||
Yolo v7 | 0.917 | Paved | 0.900 | 0.929 | 0.915 | 1,100 |
Unpaved | 0.953 | 0.900 | 0.926 | 1,100 | ||
Unknown | 0.900 | 0.922 | 0.911 | 1,100 | ||
ConvNeXt | 0.918 | Paved | 0.928 | 0.920 | 0.924 | 1,100 |
Unpaved | 0.930 | 0.911 | 0.921 | 1,100 | ||
Unknown | 0.898 | 0.925 | 0.911 | 1,100 |
