Food Classification Model

Inspiration

Over the last couple of years, fitness has become a large part of society and because of this, we have seen a rise in people being interested in their dietary choices in order to improve their overall health. As a result of this, we have seen a rise in the use of calorie-tracking apps. However, the calorie tracking apps are limited to ready-made foods or foods from restaurants. Hence, people will have difficulties tracking their own homemade food. Thus, we decided to create a model which will allow people in the future to take photos of their foods and the model will predict the ingredients in the food based on the image instead of users having to individually input every ingredient. Hence, improving efficiency and being a more user-friendly approach and at the same time indirectly promoting better health choices.

What it does

The CNN model requires a user. Firstly, the user will input an image of a certain food group such as rice, eggs, and others based on the dataset we have been given. Next, the model will predict the food groups that are present in the image whilst outputting the probability of that food group being in the inputting image given by the user.

How we built it

We built the CNN model based on the knowledge we grasped throughout all the workshops given. For example, one of the workshops was specific to deep learning and the concept of CNNs and we preprocessed the data using the other workshops which were related to computer vision and others.

Challenges we ran into

One of the challenges that we ran into was the ability to upload the given Kaggle dataset onto our Google Colab in order to process it and train the model using the images in the dataset. This caused us to use different resources and try to fix the problem in order to be able to train the model. In the end, we were able to fix the issue by asking one of the facilitators to re-explain the concept of uploading the dataset. Another challenge we encountered was trying to reduce our value loss. After cuntless changes to the dropout rate, the learning rate and more we were able to drop the value loss from greater than 20 to less than 3.

Accomplishments that we're proud of

The main accomplishment was that the model was able to work after numerous failed attempts. Furthermore, the model achieved a value accuracy of around 40%-50%.

What we learned

We learned how to create a ML model which the whole group has never done so. Furthermore, we learnt the fundamentals of Machine learning and the importance it has in improving the future of human culture.

What's Next for Food Detection Model

The broader picture for the model would be using it in tandem with another AI model which allows the user to inputs other metrics such as weight, gender, body fat percentage, and exercise levels. This is vital as it will allow the model to provide personalized recommendations and hence be more useful towards society. To further depict, our current model will use its predictions generated in order to calculate the estimated calorie, protein, fats, etc in the food in the image based on the inputs given to the model in order to recommend whether a food fits with the user's diet and nutrition plan.

Built With

• cv2
• matplotlib
• numpy
• pandas
• python
• scikit-learn
• tenserflow