Research Article
An Improved Convolutional Neural Network Algorithm and Its Application in Multilabel Image Labeling
Table 2
Comparison of the labeling accuracy when two models were trained 100 times on the Pascal VOC 2012 dataset.
| Image category | CNN | DCCNN |
| Plane | 0.983 | 0.999 | Bike | 0.877 | 0.973 | Bird | 0.918 | 0.984 | Boat | 0.920 | 0.972 | Bottle | 0.722 | 0.892 | Bus | 0.920 | 0.980 | Car | 0.819 | 0.939 | Cat | 0.916 | 0.970 | Chair | 0.668 | 0.804 | Cow | 0.999 | 1.0 | Dining table | 0.570 | 0.757 | Dog | 0.894 | 0.971 | Horse | 0.927 | 0.978 | Motorbike | 0.849 | 0.931 | Person | 0.871 | 0.957 | Potted plant | 0.729 | 0.881 | Sheep | 0.960 | 0.993 | Sofa | 0.618 | 0.827 | Train | 1.0 | 0.999 | TV monitor | 0.746 | 0.866 | MAP value | 0.845 | 0.934 |
|
|