UNet for own RGB dataset for background removal in keras

Question

I have 1000 cat images with background. I want to try UNet to remove background from those images. Any recommendations are welcome.

Should I create annotations or other files like Json files? I have not found a proper way to do it.

Answer 1

One of a great way to implement background removal is end2end methods by using cGAN or pix2pix image to image translation (you can use U-Net in your gan architecture)

I strongly recommend you to read this article: https://github.com/eti-p-doray/unet-gan-matting/blob/master/exploringImageMattingReport.pdf

and also you can use pix2pix method :

https://phillipi.github.io/pix2pix

http://www.k4ai.com/imageops/index.html