The code for "Simple and Effective Approaches for Uncertainty Prediction in Facial Action Unit Intensity Regression" at FG 2020
All building blocks are part of this repository:
- DWAR
- GP-VFE
- MLP ensemble
- Loss Attenuation
- MLP Model for dropout, U-MLP, and the Multi-Task MLP
git clone [email protected]:twoertwein/UncertaintyRegression.git
cd UncertaintyRegression
poetry update
poetry run pip install torchvision --extra-index-url https://download.pytorch.org/whl/cu116Run the grid-search for a primary model to predict facial action units (AU) intensities. Some primary models can also estimate their uncertainty.
python train.py --method dropout --workers 4 --dataset mnistTrain a secondary model to estimate the uncertainty of the primary dropout model.
python train_secundary.py --uncertainty umlp --workers 4 --dataset mnist