jackd / voc_vgg

Tensorflow implementation of vgg-fcn8s network for image segmentation of pascal voc dataset

Tensorflow implementation of VGG-FCN8s for PASCAL VOC segmentation.

Setup

1. Get the required gits and add the parent directory to your path.

cd /path/to/parent_dir
git clone https://github.com/jackd/tf_template.git  # project structure
git clone https://github.com/jackd/pascal_voc.git   # data
git clone https://github.com/jackd/voc_vgg.git      # implementation

2. Add parent directory to your PYTHONPATH.

export PYTHONPATH=$PYTHONPATH:/path/to/parent_dir

Running

cd voc_vgg/scripts/
./vis_inputs.py
./main.py --action=train
tensorboard --logdir=../_models
./main.py --action=eval
./main.py --action=vis_predictions

Customize

To customize, create your own params file and modify the existing deserializing functions to accomodate.