Follow up work of #41

[syncle]

#41 covers basic data structure and functions for RGBD odometry application.
There are few issues additionally need to be handled.

• OpenMP boosted version is not that fast enough: I expect 4~5 times faster when OpenMP is used, but the performance gain is not obvious. Should be handled.
• Some example that shows Hybrid odometry is better than old odometry.

[qianyizh]

We should also show case a complete tutorial starting from downloading one of the TUM sequences, then ran odometry, then evaluating. In general, what we did in our paper.

[syncle]

We have better OpenMP support from the discussion of #52, but when the enhanced OpenMP code applied to RGBDOdometryJacobian, the performance gain is still not as drastic as #52. I think maybe due to extensive random correspondence search and memory reference.
One way to boost is reducing the number of correspondences: not searching correspondences for every pixels, maybe for every two pixels. This will boost 4x at least.

[syncle]

We have better OpenMP support from the discussion of #52, but when the enhanced OpenMP code applied to RGBDOdometryJacobian, the performance gain is still not as drastic as #52. I think maybe due to extensive random correspondence search and memory reference.

Fixed. See #52 and #53.

One way to boost is reducing the number of correspondences: not searching correspondences for every pixels, maybe for every two pixels. This will boost 4x at least.

Offline discussed, but making tutorial has higher priority for now.

[qianyizh]

Done. Close.