Multideep Feature Fusion Algorithm for Clothing Style Recognition

Figure 10

The final model (ours) is compared with other models. As shown in (a), it can be found that no matter whether adding a YOLO model to the traditional residual network or just using improved ResNet, the best experimental results have been obtained with our proposed models. As shown in (b), the improved residual network of this paper is compared with the traditional residual network and the mAP of our model is better than that of the traditional residual network. When the epoch was around 140 and 200, MAP dropped sharply, forming two valleys and peaks.

(a) The final model (ours) is compared with two other improved models (YOLO+ResNet: YOLOv3+ResNet_101; Imp ResNet: improved ResNet_101 without YOLOv3)

(b) The improved ResNet_50 compared with ResNet_50