Skip to content

sergeyprokudin/deep_direct_stat

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

76 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deep Directional Statistics: Pose Estimation with Uncertainty Quantification.

Installation

bash scripts/install.sh

This will create a virtual environment for the project (located in "$PROJECT_DIR/py_env" folder) and install all necessary dependencies (TensorFlow, Keras, etc.).

To work with available notebooks, run:

bash scripts/start_notebook.sh

Datasets

PASCAL3D+

Download the preprocessed data and place it into "$PROJECT_DIR/data" folder.

Note: all angles are stored in biternion (cos, sin) representation. Converters to degrees\radians are available at utils/angles.py

See demo notebook for an example of loading.

Training

To train on one of PASCAL3D+ classes,run:

source py_env/bin/activate
python training_scripts/train_pascal3d.py CLS_NAME

where CLS_NAME is one of the PASCAL classes (aeroplane, car, ...)

Alternatively, see the demo notebook for a step-by-step training procedure.

Pre-trained Models

Download pretrained models.

See demo notebook for an example of loading, predicting and evaluating pre-trained PASCAL3d+ models.

Citing

@conference{deepdirectstat2018,
  title = {Deep Directional Statistics: Pose Estimation with Uncertainty Quantification},
  author = {Prokudin, Sergey and Gehler, Peter and Nowozin, Sebastian},
  booktitle = {European Conference on Computer Vision (ECCV)},
  month = sep,
  year = {2018},
  month_numeric = {9}
}

ArXiv preprint:

References

About

Deep Directional Statistics: Pose Estimation with Uncertainty Quantification

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published