Deep-Tracking

Project Description

The Optophysiology laboratory of Professor Ilka Diester investigates the motor system in rodents. Understanding cortical motor activity and how it can be influenced by optogenetic means might pave the way for a brain-machine interface for prostheses with neural feedback. A precise measurement of the animal‘s behavior is essential to relate neural recordings and neural manipulation to the animal’s motor output. In collaboration with Professor Thomas Brox we apply artificial neural networks (Deep Learning) for automatic object recognition, semantic segmentation and precise behavior tracking.

Software

• Matlab (Neural Network Toolbox, R2016a or newer).
  Teaching example of a Convolutional Neural Network with Matlab.
• Caffe.
  Teaching example of a Convolutional Neural Network with Caffe.
• TensorFlow
  Teaching example of a Convolutional Neural Network with TensorFlow.
• Annotation Tool (Python): v0.1, v0.2, b0.3
  Used to annotate data for training the (Fully) Convolutional Neural Networks.
• Video Grabber (C++)
  For high speed cameras from Basler (e.g. acA640-750um) .
  Optionally interfacing Caffe framework for RT applications.
  Based on pylon APIs

• Synchronous Video Grabber (C++)
  For high speed cameras from Allied Vision Technologies.
  Based on Vimba SDK
  Note: Vimba v2.0 not compatible with boost library v1.57
• CinePlex Studio
  Commercial video tracking software by Plexon

Hardware

• NVIDIA TitanX (Maxwell) , GTX1080, GTX1070
• High speed cameras from Basler (acA640-750um)

• High speed cameras from Allied Vision Technologies (Mako-U-029B, Stingray F-033)

Collaboration

• Professor Thomas Brox, Computer Vision Laboratory, University of Freiburg, Germany

Acknowledgment