Skip to content

Implementation of W-Net: A Deep Model for Fully Unsupervised Image Segmentation with soft n-cut loss

Notifications You must be signed in to change notification settings

detecttechnologies/W-Net

Repository files navigation

W-Net

W-Net: A Deep Model for Fully Unsupervised Image Segmentation

Tensorflow Implementation of W-Net

w-net

Both the loss functions Soft-N-Cut-Loss and Reconstruction Loss has been Implemented

Batch Processing is enabled

As this model is unsupervised, we will be needing a huge amount of dataset to train properly.

To start the training, run

python soft_n_cut_loss.py

For tensorboard visualizations,

cd checkpoints/logs
tensorboard --logdir=.

Loss Description

Reconstruction loss

N-Cut-loss

Soft-N-cut-loss

Training Procedure

Results

Pipeline of Tasks

  • Encoder Decoder Architecture
  • Reconstruction Loss
  • Soft N Cut Loss
  • Batch Processing
  • Post Processing
    • Heirarchical Segmentation
    • Conditional Random Fields

About

Implementation of W-Net: A Deep Model for Fully Unsupervised Image Segmentation with soft n-cut loss

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages