I haven’t finished Part 2 yet, so this might come up. But when covering the bounding boxes i had this idea that you probably could use a heatmap created by the convnet to detect where the objects are, and then simply set a box at the (topmost left, bottom mostright) (along with some variable to decide how “hot” the object should be to be included in bounding box) to create a bounding box without having to actually mark anything manually.
Could this work? Has this been done before? I googled around a little but couldn’t find anything.