Algorithm 2 Symptoms detection procedures |
Input: Only face image Output: The symptom recognition
-
1:
Using CSPDarknet53 transforms feature maps of different sizes in different convolutional layers.
-
2:
Use SPP to transform feature maps of any size into feature vectors of fixed size to improve the perceptual field of view.
-
3:
Use PAN to blend three feature maps of different sizes.
-
4:
Pass the result of step 4 to the head of YOLO and output the three feature maps: (19,19,num_anchor*(num_classes + 5)), (38,38,num_anchor*(num_classes + 5)), (76,76,num_anchor*(num_classes + 5)).
-
5:
Three different-size feature maps are used to calculate the prediction BB.
-
6:
The IoU is compared with the ground-truth BB to calculate the IoU loss.
-
7:
End.
|