As the comments point out, apart from immediate use to consumers, it can also help understand underlying temporal trends and perhaps with some data analysis suggest healthier substitutes based on user tastes.
However, as part of a course project in using Convolutional nets for image recognition, I was wondering, what if, we could simply take a picture of a rack in a grocery store and augment the image with the overall "healthiness" heatmap.
And now to the meat of the problem, data. Is there a readily available dataset on which the image recognition can be trained? Note that each product will need to have a images taken from a variety of angles and lighting conditions. There exist methods to automatically generate alternate sets of images, but they are not perfect and we would still need data.
Yes, we have a tagged dataset, and image recognition would just be awesome.
Can you send a mail to pierre@openfoodfacts.org so that I can invite you to the OpenFoodFacts slack, so that we can discuss it further ?
However, as part of a course project in using Convolutional nets for image recognition, I was wondering, what if, we could simply take a picture of a rack in a grocery store and augment the image with the overall "healthiness" heatmap.
And now to the meat of the problem, data. Is there a readily available dataset on which the image recognition can be trained? Note that each product will need to have a images taken from a variety of angles and lighting conditions. There exist methods to automatically generate alternate sets of images, but they are not perfect and we would still need data.