-
-
Main user interface
-
Training graphs after 10 epochs
-
Example predictions on weeds after 10 epochs
Inspiration
I've been working with computer vision -- real time object detection in particular -- for numerous projects including an autonomous weed killing robot. I found myself incredibly frustrated with web services trying to charge me for every part of the process. Every tool from annotation, augmentation, and training costed money.
What it does
I built FastYolo to provide a way to annotate, augment, and train vision systems completely locally in a streamlined process. With FastYolo, simply take a few images and draw bounding boxes over the specified areas. Once you are done, just press "Done" and the rest of the image processing will work on its own!
How we built it
For the desktop app, I used the Tkinter library for its usability on basically every platform. Behind the back I used a lot of data processing and made use of the imgaug library in Python to help with data augmentation.
Challenges we ran into
The YOLO format was incredibly confusing and it took WAY too much searching to figure out how it worked.
Accomplishments that we're proud of
I created a product that I actually intend to use in the future.
What we learned
Sometimes, you don't need to rely on the major paid web apps. It might be worth the effort to make your own!
What's next for FastYolo
There are a lot more ui and ux changes that need to be made. For example, each class should have its own color corresponding bounding box.
Built With
- imgaug
- pil
- python
- tkinter
- ultralytics
- yolo
Log in or sign up for Devpost to join the conversation.