Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.idea Pascal training fixed Sep 5, 2018
datasets remove virtualenv installation from install.sh Aug 19, 2018
models revert to stable version Aug 22, 2018
notebooks Pascal training fixed Sep 5, 2018
scripts remove virtualenv installation from install.sh Aug 19, 2018
training_scripts revert to stable version Aug 22, 2018
utils
view_estimation matlab eval WIP Jul 24, 2018
.gitignore pascal training script added Aug 12, 2018
README.md Pascal training fixed Sep 5, 2018
setup.py Pascal3d demo added Aug 12, 2018

README.md

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

You can’t perform that action at this time.