Ask HN: How to structure ML model to give 1 output per row for an image?

I'm trying to use machine learning to find a path from the top of an image to the bottom. The image itself contains a noisy line from top to bottom and i'd like the model to predict which column in each row of the image is closest to the line. I have lots of sample data to train on where each sample consists of the image and then an array of columns of length number of rows. I could create a "category" for each column and have it try to predict which category fits best but is this the correct way? Thanks