Model for Predicting Car Orientation from Images

Description

Repository: Car Orientation Prediction Repository
Paper: Car Orientation Prediction Paper

Model Description

This model is a ConvNext architecture fine-tuned for predicting car orientations from images. It represents the top-performing model in our experiments from the CFV Dataset paper.

It was pre-trained on ImageNet and fine-tuned on the CFV Dataset, consisting of 23K images of vehicles with diverse orientations.

The model expects a batch of images with sizes 224x224 and will output a vector consisting of scalar values that represent the pitch angle in degrees.

Citation Information

If you utilize this model for any project or research, please cite our paper:

@article{catruna2023car,
  title={Car Full View Dataset: Fine-Grained Predictions of Car Orientation from Images},
  author={Catruna, Andy and Betiu, Pavel and Tertes, Emanuel and Ghita, Vladimir and Radoi, Emilian and Mocanu, Irina and Dascalu, Mihai},
  journal={Electronics},
  volume={12},
  number={24},
  pages={4947},
  year={2023},
  publisher={MDPI}
}