DeepTracking: Seeing Beyond Seeing Using Recurrent Neural Networks
This is an official Torch 7 implementation of the method for the end-to-end object tracking from occluded sensor measurements using neural network presented in the academic paper:
- author: Peter Ondruska, Mobile Robotics Group, University of Oxford
- email: ondruska(at)robots.ox.ac.uk
- paper: http://www.robots.ox.ac.uk/~mobile/Papers/2016AAAI_ondruska.pdf
- webpage: http://mrg.robots.ox.ac.uk/
For any questions about the code or the method please contact the author.
Installation
Install Torch 7 and the following dependencies (using luarocks install [package]):
- nngraph
- image
- cunn (optional for training on a GPU)
Data
Download and unzip the training data for the simulated moving balls scenario:
http://mrg.robots.ox.ac.uk:8080/MRGData/deeptracking/DeepTracking_1_1.t7.zip
This is a native Torch 7 file format.
Training
To train the model run:
th train.lua
Training of the neural network using provided data takes about 12 hours on Nvidia Titan X. Every 1000 iterations the training error is logged to log_model.txt, network weights are saved to weights_model and the visualisation of its performance is stored to video_model.
Optional parameters
| Flag | Description |
|---|---|
| -gpu [id] | use GPU [id] (0 to use CPU) |
| -model [file] | neural network model |
| -data [file] | data for training |
| -iter [number] | the number of training iterations |
| -N [number] | the length of training sequences |
| -learningRate [number] | learning rate |
| -initweights [file] | initial weights |
| -grid_[minX/maxX/minY/maxY/step] [number] | 2D occupancy grid parameters |
| -sensor_[start/step] | 1D depth sensor parameters |
License
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.
Release Notes
Version 1.0
- Original version from the academic paper.
Version 1.1
- Native decoding of raw 1D depth data into 2D input.
- Larger NN network.