For ML practitioners, how much more/less difficult is this than predicting calorie content/macros from pictures/video of food? Because that seems like the holy grail of “calorie counting” apps, to seamlessly track (within a small margin of error) what your macros for the day are so far, thus helping you make intelligent choices about what to eat & portion size. Especially since those apps are great for sticking to diets, if it weren’t for all the mind-numbing data entry.

Hotdog or not hotdog

This seems to use only one image of each recipe (unless I'm misreading the code).

That seems to be a pretty big flaw.

It's sad to see people waste time on such pursuits. Getting a recipe is very easy. Just google it. Or even reverse image search it.

I do get that these exercises add a lot of know how to the deep learning process, but I really wish these smarts are applied to more practical and immediate problems.

Solving practice problems are one small step towards learning how to solve practical problems.

The project was done as a "Matura-Arbeit" which is the equivalent of a final project in highschool.