michael-fonder/rnn_depth_pose

Recurrent Neural Network for (Un-)supervised Learning of Monocular VideoVisual Odometry and Depth

RNN_depth_pose

This is a Tensorflow implementation of our paper:

Recurrent Neural Network for (Un-)supervised Learning of Monocular VideoVisual Odometry and Depth

Rui Wang, Stephen M. Pizer, Jan-Michael Frahm

arxiv preprint: (https://arxiv.org/abs/1904.07087)

Prerequisites

This codebase was developed and tested with Python3.6 Tensorflow 1.12.0, CUDA 10.1 and Ubuntu 16.04.

Preparing training data

Download KITTI raw data and depth data. Then process using the provided script in data/KITTI folder.

Training

Once the data are formatted properly, you should be able to the model by running the following command

python main.py --dataset_dir=/path/to/tfrecords --checkpoint_dir=/path/to/output_checkpoints/

You can visualize training result using tensorboard

tensorboard --logdir=/path/to/output_checkpoints/ --port=8888

TODO

The code will continue to be cleaned up and more comments will be added.

Unsupervised training version.

Demo with pretrained model will be added.

Reference

@inproceedings{wang2019recurrent,
  title={Recurrent Neural Network for (Un-) supervised Learning of Monocular Video Visual Odometry and Depth},
  author={Wang, Rui and Pizer, Stephen M and Frahm, Jan-Michael},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={5555--5564},
  year={2019}
}