haven-ai/counting-benchmark

Build a Counting Benchmark in 3 minutes

1. Install dependencies

pip install -r requirements.txt
pip install git+https://github.com/ElementAI/LCFCN

This command installs pydicom and the Haven library which helps in managing the experiments.

2. Download Datasets

• Trancos Dataset

  wget http://agamenon.tsc.uah.es/Personales/rlopez/data/trancos/TRANCOS_v3.tar.gz
  

3. Define Hyperparameters

EXP_GROUPS['trancos'] =  {"dataset": {'name':'trancos', 
                          'transform':'rgb_normalize'},
         "model": {'name':'lcfcn','base':"fcn8_vgg16"},
         "batch_size": [1,5,10],
         "max_epoch": [100],
         'dataset_size': [
                          {'train':'all', 'val':'all'},
                          ],
         'optimizer':['adam'],
         'lr':[1e-5]
         }

4. Train and Validate

python trainval.py -e {EXP_GROUP} -d {DATADIR} -sb {SAVEDIR_BASE} -r 1

• {DATADIR} is where the dataset is located.
• {SAVEDIR_BASE} is where the experiment weights and results will be saved.
• {EXP_GROUP} specifies the exp_group such as trancos training hyper-parameters defined in exp_configs.py.

5. View Results

> jupyter nbextension enable --py widgetsnbextension --sys-prefix
> jupyter notebook