CarFormer: Self-Driving with Learned Object-Centric Representations
Published in ECCV 2024, 2024
Recommended citation: Hamdan, S. et al (2024). CarFormer: Self-Driving with Learned Object-Centric Representations. ArXiv, abs/2407.15843. https://arxiv.org/abs/2407.15843
In CarFormer, we propose to learn object-centric representations in bird’s eye view (BEV) to distill a complex scene into more actionable information for self-driving in a self-supervised manner. Using this representation, we outperform both scene-level and object-level approaches that use the exact attributes of objects. Our model with slots achieves an increased completion rate of the provided routes and, consequently, a higher driving score, with a lower variance across multiple runs, affirming slots as a reliable alternative in object-centric approaches.
Access
Recommended citation:
- Hamdan, S. et al (2024). CarFormer: Self-Driving with Learned Object-Centric Representations. ArXiv, abs/2407.15843.