/U-net_for_Multiclass_Segmentation

Primary LanguagePythonMIT LicenseMIT

U-net_for_Multiclass_Segmentation

UNET PAPER: https://arxiv.org/abs/1505.04597

U-Net Architecture The architecture follows ‘U’ shaped path. It is a unit network architecture which is built upon the fully connected network in CNN.

I have used the Oxford IIIT Pet dataset, which consists of three classes:

  1. Main object (Cat or Dog)
  2. Border
  3. Background

I took only 10 epochs in training, each epoch took almost 1 hr to train, but I've trained the model for just 2 hrs and still the results are pretty good :)
The first image is the input, second one is the masked image and third one is the output we got, train the model for longer and you'll get better results.


Dataset: Oxford IIIT Pet dataset