Dataset preparation

1. Add this repository to your PYTHONPATH variable:

PYTHONPATH=$PYTHONPATH:/path/to/fork

2. Run script:

• for Cityscapes: Download the dataset here.

cd datasets/cityscapes
python prepare_dataset.py --data_dir=/path/to/data --gt_dir=/path/to/labels --save_dir=/path/to/save

• for PASCAL VOC2012: Download the dataset here.

cd datasets/voc2012
python prepare_dataset.py --data_dir=/path/to/data --save_dir=/path/to/save

Training

• for Cityscapes:

python train.py --config_path='/path/to/fork/config/cityscapes/densenet.py' --batch_size=5 --num_iters=3000

• for PASCAL VOC2012:

python train.py --config_path='/path/to/fork/config/voc2012/densenet.py' --batch_size=5 --num_iters=3000

Evaluation

For evaluation please refer to the new version of the repository:

https://github.com/ivankreso/LDN