ReLit

Prototype
Actual Deployed Web App
YOLO Intermediate Result
Statistical Data (1)
Statistical Data (2)

ReLit is a PWA(Progressive Web App) that identifies the recyclable garbage in an image, encouraging an environmentally friendly way of disposing our waste.

Inspiration

News like this really made us think that people should be more informed about garbage recycling. Research suggests that recycling 1kg of plastic could save as much as 5.774 kWh of energy, or 1500 gallons of gasoline(Greentumble). Therefore it’s imperative that we are able to identify the waste we are producing everyday. This inspired us to build a PWA(Progressive Web App) that help people do just that.

What it does

This PWA is able to use the camera on your device(PC/Mobile) to identify objects in realtime and offer appropriate tips on recycling.

How we built it

We utilized front-end frameworks such as Vue.js for UI/UX design, and Node.js in the backend. We also experimented with numerous computer vision frameworks like Tensorflow MobileNet, YOLO, and txAPI.

Challenges we ran into

Integrating computer vision APIs into the web app, experimenting with other APIs, and training our own CV models.

Accomplishments that we're proud of

We put a lot of efforts in UI/UX design, prototyping, and research. We are also proud of our integration of backend, frontend, and computer vision APIs.

What's next for reLit

Future features for ReLit includes a statistics on how much energy the user saved by recycling, and a comparison with the national average.

Works Cited

US, EPA, “Advancing Sustainable Materials Management: 2015 Tables and Figures .” Advancing Sustainable Materials Management: 2015 Tables and Figures, EPA, 2018, pp. 1–4.

“U.S. and World Population Clock.” Population Clock, www.census.gov/popclock/.

“How Does Recycling Save Energy?” Greentumble, 10 Sept. 2018, greentumble.com/how-does-recycling-save-energy/.

“Barrel(Bbl).” BusinessDictionary, www.businessdictionary.com/definition/barrel-bbl.html.

Built With

html
node.js
tensorflow
vue.js

Submitted to

SD Hacks 2019

Created by

I am responsible for selecting CNN models, cleaning data and training the models for our classification system. I have prepared technical breakdown demonstrations and training process of our own customized learning models on Jupyter and google colab.

Xiaochen Li
I contributed as a full-stack engineer, building and deploying server with node.js, as well as building client-side web app with Vue.js from ground up.

Baichuan Wu
I tested multiple CV libraries like MobileNet and Tensorflow, as well as their viability in a web app. I also participated in building backend functionalities.

Hanshu Liu
I worked on source research, presentation document, and video editing.

Fei Yin