Add kitti dataset

[Sirokujira]

Issue #139

use package : pykitti package(0.2.4/0.3.0) parseTrackletXML.py (Methods for parsing tracklets (e.g. dataset labels), originally created by Christian Herdtweck.)

Function : Display labels and boxes

not implemented Grayscale processing

[yuyu2172]

Hi! Thank you for your PR! Some comments.

1. Could you change default data directory from KITTI to kitti? (similarly for COCO, we name the directory as coco)
2. Could you raise more friendly error when date or drive_num are not specified? (label_resolution option of CityscapesSemanticSegmentationDataset is a good example for situation like this)
3. Could you change the shape of label to (0,) when there is only one instance? https://github.com/chainer/chainercv/blob/master/chainercv/utils/testing/assertions/assert_is_bbox_dataset.py#L51
4. Is it possible to drop support for gray scale image? If we drop support, can we use pykitti==0.3.0 for all settings? I personally prefer to set 0.3.0 as the recommended pykitti version.

[yuyu2172]

Do you know any experiments in academic papers that can be supported by this dataset?

[Sirokujira]

1. current data folder name Lower case(and download destination folder)

2. add check logic (check date, drivenum)

3. fixed code(label shape 2 dim -> 1 dim)

4. fixed use pykitti >= 0.3.0 (dropped function grayscale image process.)

Do you know any experiments in academic papers that can be supported by this dataset?

source of a quote? It is the item "raw dataset" of "Citation" at http://www.cvlibs.net/datasets/kitti/. (add comment kitti_bbox_dataset.py)

I do not know about the research citing the data set.