The current repository is based on pyTourch Geometric Toolbox (Python3 Notebook) https://pytorch-geometric.readthedocs.io/en/latest/

For any query email to [email protected]

If you find the code to be useful, please cite the following paper
NeuRoRA: Neural Robust Rotation Averaging, P. Purkait, T.J. Chin and I. Reid, European Conference on Computer Vision 2020
https://arxiv.org/pdf/1912.04485.pdf

The current

• pip install --verbose --no-cache-dir torch-scatter
• pip install --verbose --no-cache-dir torch-sparse
• pip install --verbose --no-cache-dir torch-cluster
• pip install --verbose --no-cache-dir torch-spline-conv (optional)
• pip install torch-geometric

The scripts to generate the synthetic datasets is adapted from the rotation averaging package of CV Lab of IISC Bangalore http://www.ee.iisc.ac.in/labs/cvl/research/rotaveraging/

To generate synthetic data for NeuRoRA, run the following scripts in order

• Example_generate_data_pytourch.m
  ** This will generate the view-graphs and can directly fed to CleanNet for Training and Evaluatuion.
  ** Note that you need to execute the code twice to generate separate training and testing datasets.
• Outlier_detect_initialization.m
  ** It requires output of CleanNet to generate data for FineNet.
  ** It does so by generating an initial solution of absolute pose from the cleaned graph.

To train the CleanNet model, call:

• Run jupyter notebook on terminal
• Open CleanNet.ipynb and run all
  ** the CleanNet model

To train the FineNet model, call:

• Open FineNet.ipynb and run all
  ** the FineNet model

The analogous files CleanNet.ipynb and FineNet.ipynb can be incorporated

• The above scripts have separate cells for training and evaluation

• test_synthetic.m
  ** This reports the prediction accuracy

For any query email to [email protected]