Agglomeration Estimation
Estimating the agglomeration in the project is made from adaptations of a model created and pre-trained by Xiyang Liu in his github. Thus, it's used to extract the number of people present from an image and its count map, as shown in Figures 1 and 2:
Figure 1: Image with the people count.
Figure 2: The people count map related to the image in Fig. 1.
The model proposed by Xiyang Liu uses a new learning target called Local Counting Map (LCM), a map divided into patches in which each patch has a value representing the number of people present in it. Thus, it is possible to estimate the total number of people by adding the values of all the patches. A regression structure is used for counting, as shown in Figure 2.
Figure 3: The overview of the agglomeration estimation model.
See the original article for more information on how the model works.