Currently trying to build #6. Just for private use. My hope is that by throwing a bunch of highly personalized information in a VLM it will provide reasonably first estimates. (E.g. if you see a bowl lentils I will probably have rice below etc.). And then iterate on the main ingredients -> fetch the macros of main ingredients from a DB. If its within 20% that would be enough for me.
I have tried some off-the-shelfe solutions and they currently do not seem to cut it, or are too complex for my use case.
I looked at this field a while back and I'd caution that estimates are dramatically off because high and low calorie foods are often identical visually.
Think of a diet soda vs a sugared one - it can be 10 vs 1000 calories easily. Almost all diet options are designed to look like the non-diet options.
I have tried some off-the-shelfe solutions and they currently do not seem to cut it, or are too complex for my use case.