Classifying Flooded image vs non-flooded Image

[gcosne]

Developing a good classifier that would recognize a Flooded image vs a non-flooded one could be useful for many applications.

We intend to use the classifier as a supplementary information to guide the image-to-image translation process at training time.

Architecture proposed: As we would like a light model that generalizes well with a small amount of training examples, a good baseline would be to fine-tune a small resnet architecture trained on image net as presented in the following tutorial pytorch-tutorial.

[gcosne]

First version of the flood classifier: cc-ai/floods-gans/flood-classifier