DepthNetwork

An implementation of Deep Monocular 3D Reconstruction for Assisted Navigation in Bronchoscopy in Python using Keras.

Keras supports multiple backends, and this implementation has only been tested with TensorFlow, but it might work with others, though it may need minor modifications.

The actual implementation of the network and other related library functions can be found in the depth_network/ folder. All executable scripts are located in the tools/ folder.

More information about each component can be found in comments at the top of the files.

Dependencies

• Python 3
• Keras
• scikit-image
• TensorFlow
• h5py
• matplotlib
• PyYAML
• openexrpython (optional, for converting depth maps from the Unity renderer)

Using sample data from paper

A set of shell scripts are provided in the tools/paper directory that will automatically download and prepare the data collected by Visentini-Scarzanella et al., as well as train and test the network.

Run the scripts in this order:

download_data.sh

prepare_data.sh

train.sh

test.sh