I recently stumbled upon zero-inflated regression and it peaked my interest so started looking into it more. Through my search, I have found out that the a common way to solve zero-inflated regression problems are tackled in two stages. The first is to treat it like a classification task to classify if the regression output is zero. And finally a regression model on the non-zero(continuous) outputs. However all of the examples I have come across involve tabular data. Is there any way of doing the same if the dataset is that of images?

To apply zero-inflated regression to images, the process can be adapted in two stages: first, use a convolutional neural network (CNN) to classify whether the target output is zero or non-zero (classification task). For non-zero predictions, apply another CNN (or the same model with a regression head) to predict the continuous value. This can be done with either two separate models or a single model with two heads for classification and regression .

Thank You