Driving In The Matrix
DESCRIPTION
Deep learning has rapidly transformed the state of the art algorithms used to address a variety of problems in computer vision and robotics. These breakthroughs have however relied upon massive amounts of human annotated training data. This time-consuming process has begun impeding the progress of these deep learning efforts. By training machine learning algorithms on a rich virtual world, we can illustrate that real objects in real scenes can be learned and classified using synthetic data. This approach offers the possibility of accelerating deep learning’s application to sensor based classification problems like those that appear in self-driving cars.
RELATED Publication
M. Johnson-Roberson, C. Barto, R. Mehta, S. N. Sridhar, Karl Rosaen,and R. Vasudevan, “Driving in the matrix: Can virtual worlds replace human-generated annotations for real world tasks?,” in IEEE International Conference on Robotics and Automation, pp. 1–8, 2017. [PDF]
@inproceedings{johnson2017driving,
title={Driving in the Matrix: Can virtual worlds replace human-generated annotations for real world tasks?},
author={Johnson-Roberson, Matthew and Barto, Charles and Mehta, Rounak and Sridhar, Sharath Nittur and Rosaen, Karl and Vasudevan, Ram},
booktitle={2017 IEEE International Conference on Robotics and Automation (ICRA)},
pages={746--753},
year={2017},
organization={IEEE}
}Dataset
Data is provided in PASCAL VOC format. The 200k archives below contain everything, we provide links directly to the 10k archives in case you would like to save bandwidth or storage.
You can obtain the entire archives from the FCAV Simulation Dataset Deep Blue Data resource. The images sets archive contains image sets for 200k, 50k and 10k.
10k images (2.4G) annotations (1.8M) and pixel level segmentation images (29M)
200k images, annotations and segmentation images can be obtained directly from the Deep Blue data resource.
50k: subset of 200k archives specified by 50k image set. 50k images, annotations and segmentation images can be obtained directly from the Deep Blue data resource.
Note this data can only be used for non-commercial applications.
Code
See our capture code and steps to reproduce training results on Github for more details and help working with these files.