Non-standard dependencies: Torch package matio

Uses trajectory-graph-based sequence augmentation. Several architectural types are included.

A sample script is provided to train a basic model architecture on one KITTI odometry subsequence (downsampled to 307x93) for one epoch. The resulting model is then used to estimate the translation on a KITTI test subsequence. Errors and the estimated trajectory are plotted. As expected, it performs pretty poorly after a single epoch on this limited data!

To run the sample script, first download the demo data and unpack it in the main project directory.

##To run the sample script: ./sample-train-test.sh

##To train a model with the default settings (only recommended when using full KITTI): th ./train-kitti-net.lua

Command-line options are listed in setup.lua.