SpaceYNet: A Novel Approach to Pose and Depth-Scene Regression Simultaneously

• IEEExplore: Paper IEEExplore link.
• At 2020 International Conference on Systems, Signals and Image Processing (IWSSIP)

• Python 3
• Tensorflow
• Matplotlib
• OpenCV
• Scikit
• Numpy.

To run train project workflow, it is possible by the following command:

$ python main.py \
    train \
    --path_data_train ../dataset/laser/ \
    --output_path ../output/

Explain you folder strucure

• train: SpaceYNet train method.
• path_data_train: training dataset to work on the training step.
• output_path: outcome from the classification model using validation data.

OuOur model achieves the following performance on the LaSER dataset

SpaceYNet is a multitasking model able to regress the robot's pose and the depth-scene image simultaneously.

We compared SpaceYNet to PoseNet - a state-of-the-art robot pose regression using CNN - to evaluate the pose outcomes.

The outcomes produced by the network and its comparison with PoseNet and ground-truth is displayed below:

Depth-scene regression:

6-DoF pose regression:

If you use SpaceYNet code in your research, we would appreciate a citation to the original paper:

@INPROCEEDINGS{Aragao2020,
      author={Aragão, Dunfrey and Nascimento, Tiago and Mondini, Adriano},
      booktitle={2020 International Conference on Systems, Signals and Image Processing (IWSSIP)}, 
      title={SpaceYNet: A Novel Approach to Pose and Depth-Scene Regression Simultaneously}, 
      year={2020},
      volume={},
      number={},
      pages={217-222},
      doi={10.1109/IWSSIP48289.2020.9145427}}