dangzheng / eig-free-release

Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses (ECCV 2018)

Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses (ECCV 2018)

This repository is a reference implementation for Z. Dang, K. Yi, Y. Hu, F. Wang, M. Salzmann, and P. Fua, "Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses ", ECCV 2018. If you use this code in your research, please cite the paper.

This repo contains the code for PnP task, for Keypoint Matching task pls check Eig-Free-Keypoint-Matching-realse

Installation

This code base is based on Python3. For more details on the required libraries, see requirements.txt. You can also easily prepare this by doing

pip install -r requirements.txt

Preparing data

This repo only provide the way to create the PnP dataset. Find the file pnp_outliers_dataset.ipynb in the folder datasets. Run it. You will find the dataset in the ./datasets which contains data for training, testing and validation.

Training

While we also provide our trained models, you can also easily train your own models. Simply run:

python main.py --run_mode=train

See config.py for more options in running the software. Try it yourself. Nearly all parameters that we changed in the paper should be there.

The default place to store the results is ./res/logs. To change this, use res_dir to set the base directory, log_dir for the suffix for the training configurations. test_log_dir is used to if you want to change the suffix for storing results. For example, log_dir can store the training configuration, and test_log_dir can store which training configuration is used on which testing dataset.

Testing

Again, testing is quite simple. After training is done, run:

./main.py --run_mode=test

You could find our the model which trained by us in the folder ./model_best. You have to move these two files to the dir ./res/logs/.

P.S. I changed the experiment setting, fix the number of matches to be 200 in the training process, for the reason that we found this will give us better result. The parameter what we are using is alpha=10., beta=5.e-3.