Learning Label Encodings for Deep Regression

Deval Shah,, Tor M. Aamodt

This repository contains code for the work on "Learning Label Encodings for Deep Regression" to be presented in the ICLR 2023 (Spotlight presentation).

Table of Contents

Head pose estimation with ResNet50: Training and inference code for head pose estimation.

Facial landmark detection with HRNetV20W18: Training and inference code for facial landmark detection.

Age estimation with ResNet50: Training and inference code for age estimation.

End-to-end autonomous driving with PilotNet: Training and inference code for end-to-end autonomous driving with PilotNet feature extractor. Trained models can be downloaded from https://drive.google.com/file/d/1mWZiNyXcUrwLvCetdtQykhMj2qUmy7Fa/view?usp=sharing Download the models to trained_models/ directory. You can use the following commands to download trained models.

mkdir trained_models cd trained_models gdown https://drive.google.com/uc?id=1mWZiNyXcUrwLvCetdtQykhMj2qUmy7Fa unzip trained_models.zip cd ../

Citation

If you find this project useful in your research, please cite:

@inproceedings{ShahICLR2023,
  author    = {Shah, Deval and Aamodt, Tor M. },
  booktitle = {International Conference on Learning Representations},
  title     = {Learning Label Encodings for Deep Regression},
  url = {https://openreview.net/pdf?id=k60XE_b0Ix6},
  month     = {May},
  year      = {2023},
}