Fig. 6

A flowchart for the whole process of the proposed algorithm. At the training phase, crowdsourced photos are collected and manually segmented to train one-stage and two-stage models. In the inference phase, the newly collected crowdsourced photos will be processed by the (one-stage or two-stage) image segmentation model and a photogrammetry model to obtain a segmented 3D model, which can be used for further estimation of area of interest