From the paper (https://openaccess.thecvf.com/content/CVPR2021/papers/Thames...):

> We asked them to estimate the mass of each ingredient present on the plate and subsequently converted these values into nutrition estimates using the same USDA [9] values we used to create our dataset

I get that there's a linear relationship between the mass of a food and its calories, but I'd expect that nutritionists would be better at estimating the calories in a plate of food than the mass of a food item. Most people aren't doing the math in their heads, they're using a frame of reference that recognizes calories. Did you have this in mind? Is there any research on this?