First one

Inspiration## Inspiration

Honestly, we were just tired of guessing how many calories are in that "healthy" salad versus that "decadent" cheeseburger. Counting calories sucks, and nobody has time to weigh every lettuce leaf and tomato slice. We figured, why not let AI do the hard work? The idea was simple: snap a picture of your food and let the magic of machine learning do the rest, so you can focus on enjoying your meal without doing math.

What it does

This project is basically your own food detective. You upload a picture of whatever deliciousness (or disaster) you're about to eat, and the model tries its best to figure out what it is and how many calories are in it. It’s like Shazam, but for your plate! The AI uses a trained model to recognize different foods and gives you an approximate calorie count, saving you from playing a guessing game with your diet.

How we built it

We used TensorFlow and EfficientNetB0 to train our food-recognizing genius. We fed it a ton of pictures of food (thanks, Food-101 dataset), and then added even more food categories because, well, people eat weird stuff. Google Colab was our trusty sidekick for training and testing, and all the calorie info was stored in a CSV on Google Drive. The result? A Colab notebook where you can upload your food pics, and boom — instant calorie detective work.

Challenges we ran into

Food looks similar — like, really similar. Is that a chicken burrito or a beef burrito? The AI often couldn’t tell either. Training the model to tell these apart was a challenge, and it didn't help that some foods look wildly different depending on the cook. Plus, keeping the Google Colab session alive and not losing progress was a headache — nothing like having to re-upload everything because the session timed out!

Accomplishments that we're proud of

We’re pretty stoked that this thing actually works. It recognizes a bunch of different foods and can give calorie estimates that are surprisingly close to reality. We also managed to make it easy for users — you upload a picture, you get your answer. No mess, no fuss. Getting all the calorie data linked up and working smoothly without any major hiccups was also a big win for us.

What we learned

We learned that training AI to recognize food is kind of like teaching a kid to tell the difference between shades of green — tricky, frustrating, and full of surprises. We also learned that cloud-based resources are great, but they come with their own set of challenges, like keeping data persistent and sessions running smoothly. And most importantly, we learned the value of making machine learning accessible and fun for people who aren't data scientists.

What's next for the Food Calorie Estimator

Next up, we want to make this even smarter — more diverse datasets, better accuracy, and fewer mix-ups between lasagna and baked ziti. We also want to turn this into a real web app that anyone can use, not just those savvy enough to use Google Colab. And how cool would it be if it could recognize multiple foods on one plate? That’s our ultimate dream: one picture, one calorie count for everything on your plate — no food left behind.

Built With

• efficientnetb0
• food-101-dataset
• google-colab
• google-drive-(for-data-storage)
• matplotlib
• tensorflow
• ython