In this paper, we have presented a model to predict flood-water level from images gathered from social media platforms in a fully automatized way. The prediction is done using a deep learning framework. More specifically we have build this model on top of the Mask R-CNN architecture. The proposed model performs instance segmentation and at the same time predicts flood level whenever an instance of some specific objects is detected. We further provide a method to combine the multiple object instances level predictions and obtain a single water level prediction for the entire image. The conducted experiments proved the ability of the trained model to effectively predict water level from images within an acceptable error.

For more details, see the poster below.