Overview / Usage

As college students, we are very susceptible the problem of not having time to think of what to cook, random leftover groceries, and boring repeated meals. Chef.ai aims to solve this problem by enabling users to upload a photo of their groceries, classify the food, and return up to 10 fun recipes that include that food. In addition, users can look at caloric intake, health labels, other required ingredients, and a link featuring more information about the recipe.

Methodology / Approach

We used a backend Stack of Flask, Celery, and Redis. The Flask backend is implemented asynchronously to accommodate for heavy traffic. The flask backend uses the OpenVINO python API in order to respond with prediction given a requested input image.

Additionally, use of task IDs and interval function calls are used to track statuses and results.

Our object detection model is based off of images from the internet, and self labeled by Christian. The process by which the model was curated mimics the general workflow of a ML software developer developing on Tensorflow's Object Detection API. In regard to deployment, we used the Intel Distribution of OpenVINO to help with inferencing and accelerating development with low power.

In order to create a user friendly, engaging experience, we used React to create a responsive and minimalistic-styled webpage for the app. The test react app demonstrates how to properly use the Flask backend, and connects the backend to the frontend.

Lastly, the Edamam API and Axios were used to find the recipes. Card displays are used to render 10 recipes that a user can choose from to make a delicious decision!

Technologies Used

• Backend:
  • flask
  • celery
  • openCV
  • OpenVINO 2019 R1.1
  • numpy
  • base64
  • PIL.Image
  • io
  • redis
• Frontend:
  • React dependencies
  • axios
  • react-webcam

Repository

https://github.com/fcr3/chef_devmesh

Collaborators