Skip to content

Latest commit

 

History

History
46 lines (27 loc) · 1.31 KB

README.md

File metadata and controls

46 lines (27 loc) · 1.31 KB

Image matting

Image Matting is a project that's inspired by Apple's Photo Subject copy feature in their iPhones. This is an Image segmentation model built on top of a U-Net architecture, the primary purpose of the model is to predict the main subject in the image segment and using the segmented image given by the model, the original image is compared to the segmented image, and the subject in the original image is removed.

This model is trained with around 950 images for 300 epochs with a batch of 32

Input shape:

(128,128,3)

Modules to install

python3 -m pip install TensorFlow

pip install opencv-python

pip install matplotlib

pip install numpy

Note

Make sure to change all the directory names to ur directory path in the code

Model working

Original image

Original image

Segmented image

Segmented image

Region image

Region image

Model loss and accuracy

Accuracy

Accuracy

Loss

Loss