Reinforcement Learning as Applied to Training DNNs for the Automotive Theater

Paper #:
  • 2017-01-0099

Published:
  • 2017-03-28
Abstract:
Reinforcement learning is a tailored deep neural network (DNN) training approach which uses an iterative process to support the learning of DNNs by targeting their specific mis-classification and detections. The process begins with a DNN that is trained on freely available image data, (which we will refer to as the base model), where a few of the categories for the classifier are related to the automotive theater. A small subset of video capture files taken from drives with test vehicles are selected, (based on the diversity of scenes, objects of interests, incidental lighting, etc.), and the base model is used to detect/classify images within the video files. An in-house software application allows for the capture of frames from the video where the DNN has made mis-classifications. These images, and the corresponding annotation files are subsequently \textit{corrected} to eliminate mislabels. The corrected annotations and corresponding images are then collated and used to re-train the base DNN model. This results in a new model that is more tailored to the automotive environment. The process is subsequently repeated, where the newly trained model, (which we will refer to as the first cycle model), is used to review the same subset of video capture files as used in the base model training. The process is repeatedly indefinitely until a satisfactory level of accuracy is achieved, where each cycle of training results in an incrementally improved model.
Access
Now
SAE MOBILUS Subscriber? You may already have access.
Buy
Attention: This item is not yet published. Pre-Order to be notified, via email, when it becomes available.
Select
Price
List
Download
$22.00
Mail
$22.00
Members save up to 36% off list price.
Share
HTML for Linking to Page
Page URL

Related Items

Standard
2011-06-01
Technical Paper / Journal Article
2013-04-08
Technical Paper / Journal Article
2013-04-08
Article
2016-02-02
Article
2016-02-02
Article
2016-02-02
Training / Education
2011-04-20
Article
2016-02-02
Technical Paper / Journal Article
2014-04-01