lppllppl920/cnn_desmoke

Learning Smoke Removal from Simulation

Before

After

This repo contains the implementation of CNN based smoke removal from simulation.

If this code is useful for your project, please consider to cite:

Chen, L., Wen, T., John, N.W
Unsupervised Learning of Surgical Smoke Removal from Simulation. 
Hamlyn Symposium on Medical Robotics. 2018.

Requirements

This code was tested with Blender, Tensorflow, CUDA 8.0 and Ubuntu 16.04.

Data Preparation

We use the da Vinci dataset from Hamlyn Centre Laparoscopic / Endoscopic Video Datasets

render_smoke.py contains the python script to render smoke on images

Usage:

1. Modify the data path in render_smoke.py
2. ./blender --python ./render_smoke.py

Training

desmoke_main.py

Testing

desmoke_test.py