Kitti Eigen Split

Question

I am wondering what is the Kitti Eigen Split. Why is it used in depth estimation? I am working with the Kitti dataset for my computer vision research. However, I am a beginner in computer vision terminology.

Thanks,

score 3 · Accepted Answer · answered Aug 20 '20 at 20:50

I have searched google a little more and I found that Eigen Split is a term from the paper:

David Eigen, ChristianPuhrsch,andRobFergus.Depth map prediction from a single image using a multi-scale deep network. In NIPS*2014, pages 2366–2374.

and here a brief summary about Kitti Split and Eigen Split:

The KITTI-Raw dataset contains 42,382 stereo image pairs grouped into 61 scenes. Image sizes 1242×375.

In the paper above they split Kitti in two ways:

Kitti Split:

28 Scenes - testing.

33 Scenes - training and validation.

29,000 images -> training

1,159 images -> validation

Eigen Split:

29 Scenes - testing.

32 Scenes - training and validation.

22,600 images -> training

888 images -> validation

score 0 · Answer 2 · answered Sep 10 '20 at 15:48

The Eigen split is, as you've written in your own answer, a split proposed by Eigen et al. and used by subsequent works to be able to relate to their results in a fair manner. As far as I know, this split contains 697 test images. Please refer to this file for an exhaustive list. You should find all the entries in the raw KITTI dataset.

Kitti Eigen Split

2 Answers2