/ladder-densenet

Primary LanguageC++MIT LicenseMIT

Dataset preparation

  1. Add this repository to your PYTHONPATH variable:
PYTHONPATH=$PYTHONPATH:/path/to/fork
  1. 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