fbarriosr/madagascar

# Counting penguins with density map estimation ## Environment - python >=3.6 - pytorch >=1.0 - opencv-python >=4.0 - scipy >=1.4.0 - tqdm - h5py >=2.10 - pillow >=7.0.0 - pandas ## Construct dataset from annotation With all the original data (Four folders: Thomas, Jack, Maisie, Luke), and annotations(JSON folder) in the same root, simply run the following code. This code file includes the way to generate density maps. ```python generate_dataset.py ``` Then split the dataset into the train, Val and test sets. ```python train_val_test.py ``` If everything works well, you will get a new folder called 'processed_data' under the current root. ## Train the model ```python train.py ``` ## Test the model ```python test.py ``` We provide a pre-trained model and it can be downlowded here: drive.google.com