DATA: Include "negative examples" in our stop_signs dataset.

[TCrawley11]

A dataset team member will be responsible for adding "negative examples" in our stop_signs dataset. Negative examples are images where a stop sign is not present, indicating that there are "negative" detections of stop signs.
The added images should be a mix of road scenarios with no signs in frame, as well as images that have other signs such as yields or speed limit signs.
The "negative_examples" set should be about 1/6 of the size of the stop_signs set for both training and validation. Not too large so that training is skewed, but not too small that our model is not robust enough. This split may need tweaking.
Images should be gathered from KITTI set.