MBRVO: Constructing Blur Robust Visual Odometry Based on Blurring Artifacts

This motion blur dataset was synthesized using Unreal Engine 5. The dataset contains 230,000 sharp and blurred image pairs, in addition we also provide the groundtruth optical flow, depth, and camera trajectory.

MBR-Code

The code of this paper will be released here after the paper is accepted.

MBR-Dataset

Currently, only a portion of the sequences of this motion blur dataset is publicly available. We will open source the all dataset when the paper is accepted. Click on this link to download the test data.
【Baidu】 https://pan.baidu.com/s/1JNKzgmytF8yhRwvHnEMzGQ?pwd=jj2s Password: jj2s
【Google】 https://drive.google.com/drive/folders/12DNUOU8DFfUSkbXx4wY5p-54-MUKD6f1?usp=sharing

Demo

============Blur Image=====================Sharp Image===========
blur image image

===========Optical Flow=======================Depth=============
image image

Camera Parameters

The camera parameters used to generate this dataset are shown in Table 1.

Data Structure

-Dataset
 |--train
 |    |---scenes
 |    |    |----image
 |    |    |     |-----blur
 |    |    |     |-----sharp
 |    |    |----depth
 |    |    |----optical flow
 |    |    |----poses   
 | --test
 |    |---scenes
 |    |    |----image
 |    |    |     |-----blur
 |    |    |     |-----sharp
 |    |    |----depth
 |    |    |----optical flow
 |    |    |----poses