Rasoee

Landing Page of Website
Type Description and Cuisine of Dish
Result of Dishes matching specification from storage
Upload Image for Search
Result of Image Search
Application Launch Screen
Analysis page which shows recipes
Details page

Rasoee

We are a group of friends studying together, and we all happen to be foodies. So this is a project quite close to our heart, or maybe stomach. To be honest we didn't really think a lot about this topic, after we registered we met online for a brainstorming session, and of course everyone liked the idea related to food, so we went with it.

So, all of us had varying levels of familiarity with Machine Learning and PyTorch as well as knowledge in other fields, and so we divided the tasks among ourselves accordingly. One guy made the website, one guy made the app, the others looked for datasets and models that we could train to make this work. We wanted to be ambitious enough to make out own dataset despite there being Food-101 available online, so we went ahead with that and scraped to the best of our abilities to make a workable dataset. The cleaning of the dataset was quite a gruelling task, followed by the question of which model to use. We tested a few small scale models as we wanted this to be deployable on mobile devices, and in the end got a good enough accuracy with one to go ahead. This was followed by the usual nitpicking in the looks of the website and the application along with the addition of auxiliary features like provision of recipe and list of ingredients.

Finally we made something, that we are proud of, and hope it prevents many food lovers out there from racking their brains during identifying dishes, and use more of their time eating.

Built With

Submitted to

PyTorch Summer Hackathon 2020
- Winner Second Place

Created by

I worked on the web scraping for the dataset and the cleaning of the dataset. I also wrote the general code to train a PyTorch model, and used it along with the others to test various models for the best accuracy. I also worked on the CSS part of the Django website's front-end.

Arijit Gupta
I worked on training and optimizing the classifier model for our custom food dataset. I enjoyed the ease of using PyTorch itself and appreciated the convenience of deploying the model on various different platforms.

Dev Churiwala
I worked with Pytorch to build the model. I had only used Tensorflow until now and this has been a great opportunity for me to get into Pytorch as well.

Smitesh Patil
I worked on the mobile application for the Rasoee project. It was my first time using Kotlin, and using the MVVM architecture. The model evaluates on the edge, using PyTorch Libraries.

Ameya L
Anish Mulay