reinfocus 0.0.2

Environments where agents can focus cameras on rendered objects.

Reinfocus

reinfocus is a python package that makes it easy to create reinforcement learning environments that use ray tracing to simulate camera focus. See the examples for an impression of how it can be used.

A ppo trained agent acting in DiscreteSteps-v0. The checkerboard target is positioned at some depth, and the agent decides what depth to focus on. It receives as input the focus depth, focus value, and the change in both those since the last time step.

Installation

To install reinfocus, use pip install reinfocus.

For GPU support, install the lastest NVIDIA graphics drivers. Next you will need to install cudatoolkit; how you do that depends on what type of python installation you use:

• anaconda or variants (recommended): use conda install cudatoolkit
• plain python (untested): install cudatoolkit, then set CUDA_HOME

Special Thanks

reinfocus.graphics is a numba translation of the wonderful Ray Tracing in One Weekend in CUDA by Roger Allan.