Generate bounding boxes - Kaggle Whales Challenge

Hi everyone,

I am trying to replicate this kernel, using FAST.AI :
https://www.kaggle.com/martinpiotte/bounding-box-model/notebook

However, I have no clue how to approach the problem. So far I’ve generated a csv file containing two columns : one with the image_name, and the other with a python list containing four points (for the top and bottom, and leftmost and rightmost points).

But I don’t know what kind of dataBunch I should create. ImageDataBunch is for regression, and PointsItemList is for only one point.

I wonder if anybody would have an approach for this problem ?