An example inference
This project was started by dewball345
Object detection libraries are everywhere, and you can find python, web, and other implementations of them. But I have yet to find a comprehensive object detection for ear detection, let alone one that can do inference in the web. So I created ear.js, a javascript library for real-time ear object detection.
- Ear.js is easy to implement in your code, with prebuilt classes for preprocessing, postprocessing, and drawing results to canvas.
- Ear.js is very flexible; it only gives you what you need to have(the final inference data), and from that you can use it in many ways
- On a relatively high end PC performance was about 15 fps.
- On literally any other device the performance was a bout 3 fps.
So no, this is not realtime.
Please test your fps on your device using the web demo.
Type git clone https://github.com/dewball345/ear.js.git. You will see an index.html and index.js file with sample code. Keep the import statements in the html and javascript files.
If you want to use the project externally, you can use CDN. In your javascript file type:
import {EarModel, GeneralUtils} from https://cdn.jsdelivr.net/gh/dewball345/ear.js/ear.js
to use the Ear model and have access to some other utilities that may be useful(like drawing to a canvas)
A tutorial is currently being developed. You can view the tutorial HERE
This project is still under development, so you can expect things to change. In addition, the model, which can be found HERE is not ready for use in production. I have yet to train it with other data besides me.
This model is based on the yolo-v5s architecture and retrained to detect ears.
However, this model is far from done; it was only trained on pictures of me(with the intent of using a dataset like celeb-faces or something later), so the model will not work on everyone.
Don't expect optimal performance right now, especially if you look different than I do. I will update these files once I retrained my model on a more diverse set of faces.
And after I update my files, I may decide to add metrics like MAP. Right now, assume that it's performance is similar to yolov5 in ultralytics' repository
Please write your issue in the issue tracker
dewball345/ear.js is licensed under the BSD 3-Clause "New" or "Revised" License