Squirrel Lens

The UofM is best known for its super **CHONKY** squirrels, but how do we make them more of a "Michigan Difference"? The Squirrel Lens makes a difference! This project helps detect chonky squirrels.

Comment

Inspiration

This project is inspired by Umich students' love for squirrels, and I decided to make such a project that helps people identify chonky squirrels and potentially draw attention to the shaved squirrels issue happened recently.

What it does

This project is a chrome extension that detects chonky squirrels on web pages and highlight them. However, this project also supports a gradio interface that allows users to quickly inference by uploading images or videos.

How we built it

This project uses YOLOv8 as the base model, fined tuned for 10 epochs with 220+ handpicked chonky squirrel images from various sources including UofM reddit, instagram, twitter, and 200+ non-chonky squirrel images from squirrel data dataset. The total dataset size is 1122 images after augmentation. Base model used for this project is yolov8s, and it achieved a mAP50-95 of 45.0.

Challenges we ran into

The greatest challenge for me in this project is that it is my first time writing a chrome extension, so I had to do lots of research on how to implement even the simplest API call. Ajax requests were unavailable for background service workers and I was stuck on getting fetch responses for several hours. I really only had less than one day to learn everything I need.

Accomplishments that we're proud of

Although with a variety of difficulties and insufficiency, this project came out essentially as what I imagined it will be.

What we learned

Chrome extension development and yolov8 training pipeline

What's next for Squirrel Lens

Pack the extension and publish the extension. Solve the problem of being unable to read certain images on webpages. Deploy flask backend to a server instead of localhost.

Share this project:

Updates