DriverGaze360: Omnidirectional Driver Attention with Object-Level Guidance

This repository hosts the pretrained weights for DriverGaze360, a model for omnidirectional driver attention prediction with object-level guidance.

Available Checkpoints

Checkpoint	Training Dataset	Input Size (H × W)	Notes
drivergaze360.pt	DriverGaze360	224 × 1120	Panoramic/omnidirectional input
ondada.pt	DADA-2000	224 × 224	Standard front-facing camera input

Citation

If you use these weights in your research, please cite:

@article{govil_2025_cvpr,
  title        = {DriverGaze360: OmniDirectional Driver Attention with Object-Level Guidance},
  author       = {Shreedhar Govil and Didier Stricker and Jason Rambach},
  year         = {2025},
  eprint       = {2512.14266},
  archivePrefix= {arXiv},
  primaryClass = {cs.CV},
  url          = {https://arxiv.org/abs/2512.14266}
}