FgSegNet : Foreground Segmentation Network

This repository contains source codes and training set for the following paper:

"Foreground Segmentation Using a Triplet Convolutional Neural Network for Multiscale Feature Encoding" by Long Ang LIM, Hacer YALIM KELES

Paper Link : https://arxiv.org/abs/1801.02225

Citation

If you find FgSegNet useful in your research, please consider citing:

@article{lim2018foreground,
  title={Foreground Segmentation Using a Triplet Convolutional Neural Network for Multiscale Feature Encoding},
  author={Lim, Long Ang and Keles, Hacer Yalim},
  journal={arXiv preprint arXiv:1801.02225},
  year={2018}
}

Requirements

This work was implemented with the following frameworks:

• Python 3.6.3
• Keras 2.0.6
• Tensorflow-gpu 1.1.0

Usage

Easy to train! Just a SINGLE click, gooo!!!

1. Modify the following files in your installed dir with the files in utils dir:

   • <Your Keras DIR>\layers\convolutional.py
   • <Your Keras DIR>\backend\tensorflow_backend.py
   • <Your Keras DIR>\keras\losses.py
   • <Your Keras DIR>\metrics.py
   • <Your PYTHON 3.6>\site-packages\skimage\transform\pyramids.py
     replace
     out_rows = math.ceil(rows / float(downscale))
out_cols = math.ceil(cols / float(downscale))
     with
     out_rows = math.floor(rows / float(downscale))
out_cols = math.floor(cols / float(downscale))

2. Download VGG16 weights from HERE and place it in appropriate directory (e.g. FgSegNet dir), or it will be downloaded and stored in /.keras/models/ automatically.

3. Download our train+val frames from HRER and CDnet2014 dataset, then place them into corresponding directory.
   

   Example:

     FgSegNet/
          FgSegNet/FgSegNet.py
                  /FgSegNetModule.py
          FgSegNet_dataset2014/
                    baseline/
                         highway50
                         highway200
                         pedestrians50
                         pedestrians200
                         ...
                     badWeather/
                         skating50
                         skating200
                         ...
                    ...
           CDnet2014_dataset/
                     baseline/
                          highway
                          pedestrians
                          ...
                     badWeather/
                          skating
                          ...
                     ...      
   

Evaluation

We perform two separated evaluations and report our results on two test splits (test dev & test challenge):

1. We compute our results locally. (on test dev dataset)
2. We upload our results to Change Detection 2014 Challenge. (on test challenge dataset where ground truth values are not shared with the public dataset)
   (Both results are reported in our paper. Please refer to it for details)
   

Compute metrics locally using CDnet Utilities

Note:

• test dev: by considering only the range of the frames that contain the ground truth labels by excluding training frames (50 or 200 frames)
• test challenge: dataset on the server side (http://changedetection.net)

License

For research purpose only. For commercial license, please contact the authors.

Contact

lim.longang at gmail.com
Any issues/discussions are welcome.