ARL 37 – Programmer, Synthetic Image Generator for Deep Learning Using Unity 3D

Project Name
Creation of Synthetic Annotated Image Training Datasets Using Computer Graphics for Deep Learning Convolutional Neural Networks

Project Description
Work as part of a team on a project to develop and apply DLCNN on field deployable hardware.
Purpose: Accelerate deep learning algorithms to recognize people, behaviors and objects relevant to military purposes using computer graphics generated training images for complex environments.
Product: A training image generator which creates a corpus of automatically annotated images for a closed list of people, behavior and objects. Optimized fast and accurate machine learning algorithms that can be fielded in low-power, low-cost and low-weight fieldable sensors.
Payoff: Create an inexpensive source of military related training data and optimal deep learning algorithm tuning for fieldable hardware, which could be used to create semi-automatic annotated datasets for further training and be scalable for the next generation machine learning algorithms.

Job Description
Develop a Unity 3D based image generator to create “pristine” and sensor degraded synthetic data suitable for training and testing DLCNN’s, e.g. Caffe, TensorFlow, DarkNet…. assets such as personnel, vehicles, aircraft, boats and other objects will be rendered under a variety of observation and illumination angle conditions, e.g. full daytime cycle, weather conditions (clear to total overcast, low to high visibility, dry and rain, snow).

Preferred Skills
– Familiarity with Unity3D gaming engine
– Able to program in C#, shaders
– Self-motivated and able to work with existing code, github

Apply now.

Go back to the summer research program list.