Fully Convolutional Networks for Semantic Segmentation

Sample Output

Dataset

• Traning Data: BSD [link]
• Testing Data: VOC 2012 [link]
  
  Both BSD and VOC 2012 Dataset contains 21 categories of object. [Category list]
  

Segmentation Class

[Class list]

Prerequisites

1. Tensorflow Version --v1.4 with CUDA > 8.0
2. Numpy --v1.15
3. OpenCV --v4.0
4. Matplotlib --v2.0

Training

Once the dataset is obtained, the model properties can be configured in config_files directory:

config_files
  │ config_fcn16s_train.yaml
  | config_fcn32s_train.yaml
  | config_fcn8s_train.yaml    

and then run the following command:

./run0.sh config_files/config_fcn32s_train.yaml

or

./run0.sh config_files/config_fcn16s_train.yaml

or

./run0.sh config_files/config_fcn16s_train.yaml

Testing

./run0_test.sh config_files/config_fcn32s_test.yaml

or

./run0_test.sh config_files/config_fcn16s_test.yaml

or

./run0_test.sh config_files/config_fcn16s_test.yaml

Visualization

jupyter notebook
  │evaluate_and_viz.ipynb

Paper citation

@article{ShelhamerLD17,
  author    = {Evan Shelhamer and
               Jonathan Long and
               Trevor Darrell},
  title     = {Fully Convolutional Networks for Semantic Segmentation},
  journal   = {{IEEE} Trans. Pattern Analysis and Machine Intelligence},
  volume    = {39},
  number    = {4},
  pages     = {640--651},
  year      = {2017}
}