Awesome
segmentation-over-web
My attempt at making an interactive app that works from browser
and trains an ML model.
I could not find a good referneces to do training on tensorflow.js. I found many sketching libraries that would work for annotation that could help to build datasets. I choose segmentation because it was easier, just sketch object and background. If I had to choosen detection then I would have to generate xml files of bounding boxes and cordinates.
Basically user takes snaps from webcam.Lets you annotate it on the webpage and send it over to flask server. Server stores both images. The model gets trained using segmentation_keras.
Right now it can take snaps. Can build simple datasets. and infer once you are done.
To do
- May be add tensorflow serving
Running
sudo apt install python3-pip
pip3 install -r requirements.txt
python3 main.py