Food-Image-Recognition

Table of Contents

• Food-Image-Recognition
  • Table of Contents
  • About the Project
    • Overview
    • Built With
    • Dataset
  • Results
    • Demo
    • Accuracy
    • Loss
    • Testing on random images.
    • Visualization of different layers.
    • Heat-Map & Class-Activation-Map
  • Contributing
  • License
  • Contact
  • References
  • Acknowledgements

About the Project

Photo by Jay Wennington on Unsplash

Overview

• Each year, approximately 6,78,000 deaths are caused in the United States of America due to unhealthy diet.
• A typical American diet is too high in calories, fat, sugars, sodium, etc.
• Hence, people have became more proactive when it comes to health matters.
• Services like eating habit recorder and calorie/nutrition calculator have became extremely popular.
• They can make users aware of problems like obesity, cancer, diabetes, heart-disease, etc. that can be caused by unhealthy diets.
• Most of these services require the users to manually select a food item from a hierarchical menu which is a time consuming process and not so user friendly.
• An user-interactive system that takes food images as an input, recognizes the food automatically and gives the nutritional-facts as an output will save a lot of time.
• This system can be used in various areas such as social network, health-care applications, eating-habit evaluations, etc.
• For food image recognition we will be using transfer learning to retrain the final layer (with 101 additional food-classes) of Inception-v3 model which is already trained by Google on 1000 classes.
• It almost took 10-11 hours to train the model on Google Colab.

Built With

• Python
• Jupyter Notebook
• Google Colab

Dataset

Food Images Source: The Food-101 Data Set

• The data set consists of 101 food categories, with 1,01, 000 images.
• 250 test images/per class and 750 training images/per class are provided.
• All the images were rescaled to have a maximum side length of 512 pixels.

Nutrition Information Source: Food Data Central API

• U.S. Department of Agriculture, Agricultural Research Service. FoodData Central, 2019. fdc.nal.usda.gov.

Results

Demo

Accuracy

Loss

Testing on random images.

Visualization of different layers.

Heat-Map & Class-Activation-Map

Contributing

Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/amazing-feature)
3. Commit your Changes (git commit -m 'feat: some amazing feature')
4. Push to the Branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Maharsh Suryawala - Portfolio

Project Link: https://github.com/MaharshSuryawala/Food-Image-Recognition

References

• https://cspinet.org/eating-healthy/why-good-nutrition-important
• https://www.tensorflow.org/api_docs/python/tf/keras/applications/InceptionV3

Acknowledgements

• Img Shields
• Iconify
• MIT License
• Unsplash
• Badgen
• For The Badge